Bill Miller’s career is one of the most instructive arcs in modern investing because it contains all three acts: a historic rise, a public collapse, and a credible redemption. At Legg Mason Value Trust, Miller became a legend by beating the S&P 500 for 15 consecutive years. Then came 2008, Miller decided to bet heavily in financial stocks during the panic, but these companies did not recover and some went bust. This mistake devastated his fund, losing roughly 70% of its value and destroying much of the prestige he had built over decades. For years since his departure from Legg Mason, most investors on Wall St would remember him as the man who flew too high and took too much risk—flying too close to the sun guaranteed his failure.

For many investors, that would have been the end. It was not. Miller rebuilt his career through Miller Value Partners, continued to benefit from some of his most important long-term holdings, and later made an early investment in bitcoin that became one of the defining wins of his second act. That is why his career is so worth studying. It is not useful because it is clean or flattering in hindsight. It is useful because it contains all three lessons at once: how distinctive thinking can drive extraordinary success, how the wrong kind of risk can overwhelm even a great investor, and how a serious setback can force a deeper understanding of markets, risk, and survival.

Focus on the Roots, Not the Branches

The first lesson comes from Miller’s rise: he saw businesses through multiple lenses, not just financial screens. The best example is Amazon. In May 1999, Barron’s ran its now-infamous “Amazon.bomb” cover story, arguing that investors were only beginning to realize how many problems the company had. By conventional value standards, Amazon looked completely wrong. It was not statistically cheap, its future was highly uncertain, and many people doubted it would survive. Miller leaned in anyway. His fund became the third-largest Amazon shareholder, behind only Jeff Bezos and MacKenzie Scott. In a 2001 Fortune profile, Miller recalled meeting Amazon executives at the Santa Fe Institute in 1999 and coming away deeply impressed by the company’s economic model. He saw a business that could achieve major economies of scale, avoid the burden of physical retail inventory, and operate with a structurally lower cost base. More importantly, he believed Amazon had the potential to become a winner-take-most business rather than a conventional near-term earnings story. His timing was poor, since he bought near the peak of the dot-com bubble, but the underlying thesis was extraordinarily right.

That instinct helps explain why Miller outperformed for so long. He did not think investing was about buying the lowest multiple on a spreadsheet. He tried to understand a business the way you might understand a person: its character, its incentives, its resilience, and what it could become over time. Much of his work was a form of future-casting. He would ask where the company might be five or ten years out, then work backward to judge whether the business had the quality to survive the path in between. This is what made him look like a maverick. He was often buying what others rejected, not because he simply enjoyed being different, but because he was focused on underlying circumstance rather than surface attributes. Michael Mauboussin captured this distinction well when he described Miller as a circumstance-based investor rather than an attribute-based one.

Onlookers often describe the investment strategy of successful investors as eclectic. Perhaps it is more accurate to describe their approach as circumstance-based, not attribute-based. Bill Miller is a good case in point. Miller’s approach is decidedly circumstance based, yet he is routinely criticized for straying from an attribute-based mindset

Michael Mauboussin

That mindset was reinforced by two important influences outside traditional finance. One was philosophy, which trained Miller to question assumptions, hold competing ideas in tension, and distinguish reality from convention. The other was his long involvement with the Santa Fe Institute, where complexity theory offered a richer view of markets and businesses as adaptive systems with many interacting parts, feedback loops, and occasional regime shifts. From that perspective, value is not simply “cheap on a multiple.” It is the gap between what the market believes is happening now and what the business may actually become. Miller’s first lesson, then, is that great investing often comes from seeing the deeper structure of a business before the numbers fully reveal it. The investor who only studies static attributes may miss what the company is in the process of becoming.

Anything Multiplied by Zero is Still Zero

The second lesson comes from Miller’s fall, and it is harsher: even a great investor can be destroyed by the wrong type of risk. His collapse in 2008 was not a normal drawdown where earnings weaken and valuations compress. It was a systemic crisis in which the market stopped asking what a company was worth over time and started asking whether it could survive next week. Miller moved decisively into “ground zero” of part of the financial meltdown and loaded up on distressed financial companies such as Bear Stearns, AIG, and Citigroup. This was consistent with an instinct that had worked for him before. When fear becomes extreme, bargains appear. But in 2008, many of these stocks were not simply misunderstood assets. They were highly levered institutions facing a true liquidity and solvency problem. These companies can move from cheap bargains to worthless in days.

Miller’s own reflection is unusually blunt…

The thing I didn’t do from day one was properly assess the severity of this liquidity crisis. I was naive.

Bill Miller reflecting on his 2008 experience

That is the key point. He did not merely get a valuation wrong. He misidentified the nature of the event. A systemic liquidity crisis is not a normal market dislocation. It changes the rules of the game. When funding disappears and confidence breaks, intrinsic value can become temporarily irrelevant because survival becomes the only thing that matters. Miller was still thinking like a long-term investor in businesses, but many of his investments went bankrupt or were acquired at very low prices. In that environment, being directionally right about long-term value was not enough.

The damage was made even worse by Miller’s fund structure. As investors rushed to take money out of his fund, Miller was forced to sell positions at the worst possible time. That is what turned a bad investment environment into a collapse. His experience reinforces two hard truths. First, the risk that matters most is not volatility but permanent loss of capital. No matter how cheap you buy a business for, you will lose money if it goes bankrupt. Second, your fund structure can force you to betray your own thesis. Even if you are right over a three- to five-year horizon, client redemptions, mandates, and reputational pressure can destroy your ability to hold on long enough for the thesis to work. Miller’s fall is a reminder that brilliance in picking stocks does not insulate you from Black Swans—you should make sure you can survive even the worst disasters.

You Cannot Win the Infinite Game

The third, and perhaps most important, lesson from Miller’s redemption is that investing is an infinite game. What matters most is not chasing quick wins or building an impressive streak, but staying in the game long enough for your best decisions to thrive. One fatal mistake can end that process altogether.

Miller’s second act was not simply a lucky recovery. It reflected a change in his views on risk. After the crisis, he became more sensitive to rare events that can cause permanent impairment. He also placed greater value on managing capital on his own terms, with less vulnerability to the panic and redemptions that had magnified the 2008 collapse. That matters because independence is not just psychological. It is structural. An investor whose money can be pulled at the worst time is playing a fundamentally different game from one who can ride through distress. Buffett and Munger benefited enormously from this at Berkshire Hathaway, where insurance float gave them a durable capital base that could not be redeemed in a panic. Miller understood more clearly in his later career that controlling your investing structure is part of controlling your destiny.

His redemption also says something deeper about conviction. Miller did not abandon bold thinking after 2008. He remained capable of making large, unconventional bets when he believed the market was missing the bigger picture. His later success with bitcoin is one example. So is the fact that he continued to benefit from some of the long-duration insights that had always defined his best work. The lesson is not that he became conservative. It is that he became more aware of the difference between a misunderstood opportunity and a fragile situation where one bad scenario can wipe you out. That distinction is subtle, but it is everything. His second act suggests that the goal is not to avoid being wrong, since all investors are wrong sometimes. The goal is to avoid being wrong in a way that ends the game.

That is why Bill Miller’s career remains so useful to study. The rise shows the power of seeing businesses as complex systems rather than static financial objects. The fall shows how catastrophic risk and structural fragility can overwhelm even an extraordinary mind. The redemption shows that surviving a great mistake can deepen an investor’s understanding of risk, time horizon, and capital structure. Miller’s story is not inspiring because it is triumphant. It is valuable because it is honest. It shows that investing is not just about insight. It is also about survival.

Grail’s rollercoaster backstory

Grail is a cancer-screening biotech spun out of Illumina in June 2024. Its flagship product, Galleri, is one of the first multi-cancer early detection (MCED) blood tests. The pitch is straightforward: one blood draw that can detect multiple types of cancers. Ideally, this technology can catch tumors earlier, when treatment is typically more effective and less costly.

When Grail first began trading after the spin, the stock was around ~$12/share. Over the last year, even without profitability, the company built real commercial momentum—volumes grew, margins improved, and the stock ran to roughly ~$100/share at its peak. Grail also raised a $325M private placement (PIPE) that included Samsung, Hims & Hers, and several life sciences investors.

Then came today. Shares fell to around ~$50 after topline results from the NHS-Galleri study in England were released. As the price action suggested, the market considered it a major setback for Galleri’s path to broad adoption. Is it time to run for the exit? Or is there more to the story?

The NHS-Galleri topline is a disappointment

The clean “win” investors wanted didn’t happen: the trial did not meet its primary endpoint—showing that using Galleri at scale leads to fewer people being diagnosed only after cancer has already reached a late stage (Stage III–IV) compared with usual screening alone. In short, the study didn’t deliver the clear population-level evidence that Galleri meaningfully shifts diagnosis earlier in a way payers can bank on.

That miss matters because it undercuts the core “clinical utility + cost-effectiveness” thesis for MCED. The bet is that earlier detection means simpler treatment and lower total cost of care. If a large randomized trial can’t cleanly show fewer late-stage cancers, payers have an obvious reason to be cautious: the test can add costs (more testing and follow-up) without reliably producing the downstream savings and outcome improvements needed to justify broad coverage.

If payers won’t reimburse Galleri, is it game over?

Lack of coverage would clearly cap the ceiling for Galleri. But it doesn’t necessarily kill the business, because Galleri is increasingly being commercialized as a consumer-centric product in the US, funded primarily by cash pay, especially among higher-income consumers. We’ve already seen consumers step in where coverage is patchy, with GLP-1s via DTC channels and telehealth being the most visible example. The same consumer behavior is starting to show up in proactive screening: if someone wants an MCED test and has the means, the buyer can simply be the individual.

That’s why Grail’s distribution strategy matters. Galleri is now being sold through consumer-friendly on-ramps like Hims & Hers, Function Health, and Everlywell. Among these, Hims & Hers is the clearest scale signal because it’s a large consumer health platform that can turn screening into a productized, repeatable experience.

And the early numbers suggest this strategy can work. In Q4 2025, Grail reported it sold more than 185,000 Galleri tests in 2025, grew U.S. Galleri revenue 26% YoY to $136.8M, and EBITDA is getting close to breakeven—despite operating with no Medicare/Medicaid or commercial insurance coverage. If Grail can keep expanding through consumer channels, it can remain viable even if broad reimbursement takes longer—or never fully materializes.

It’s also reasonable for insurers and providers to stay skeptical of MCED until the evidence is overwhelming. But investors should run the counterfactual: what does Grail look like if broad reimbursement stays limited? The NHS-Galleri headline makes that scenario feel more plausible, but it’s not automatically a disaster if consumer demand continues to do more of the work.

Buy the dip? Not so fast

It depends on how you value Grail. The company still has no positive gross profit, EBIT, EBITDA, or even adjusted EBITDA, so there’s no earnings anchor. The balance sheet doesn’t help much either: most of Grail’s ~$2.2B book value comes from ~$1.8B of intangible assets (developed technology, trademark, and R&D). That looks overvalued vs. its cancer diagnostics peers (Exact Sciences: $920M, Guardant Health: $5M, Natera: $0), especially given Grail is effectively a one-product company today. At ~$50/share (~$2B market cap), Grail is still about 5x tangible book value and 13x sales… not exactly screaming cheap.

It’s unlikely to go bust in the next few years, but I don’t see enough margin of safety (at $50/share) to call this a no-brainer buy. If your cost basis is less than $20/share, best to sit tight and see where it goes.

When I was a boy, there was a popular saying: ‘Don’t just sit there; do something.’ But for investing, I’d invert it: ‘Don’t just do something; sit there.’

Howard Marks

Disclaimer:

The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

But the rabbit hole was still worth it, because once you stop treating VTubers as “streamers with anime faces,” you realize you’re looking at something closer to a new kind of entertainment IP machine.

Anycolor isn’t just managing creators. It’s running a character-IP franchise business that looks closer to a modern entertainment studio like Disney and DreamWorks than a typical talent agency. The key difference is structural: the “mask” is the product. And that mask is not just a face-cover. It’s a persistent identity with a visual language, personality, lore, and a vibe that fans attach to.

In a classic long-tail market defined by winner-take-most at the top and endless niches in the fat tail, I believe Anycolor has chosen the more robust strategy. That’s why I’m comfortable calling it a high-quality compounder at a valuation that still feels reasonable.

This post lays out the thesis.

What Nijisanji is, and how Anycolor makes money

Nijisanji is the flagship brand under Anycolor: a large roster of VTuber talents who stream, perform, collaborate, and build fandoms under a unified umbrella.

The typical VTuber agency model works like this:

• Talents apply to join an agency.

• The agency provides training, production support, managers, moderation, and coordination.

• The agency provides distribution leverage: a built-in audience, cross-pollination between talents, and a platform-like ecosystem that pushes discovery.

• Importantly, VTuber agency constructs a bespoke avatar and identity for the talent, think personality, lore, and genre.

Now the crucial VTuber twist: fans aren’t just bonding with a “person streaming with a webcam.” They’re bonding with a human-avatar hybrid character that evolves organically, often beyond what either a human or machine can create on its own.

That identity is powered by increasingly affordable technologies:

• 2D/3D computer animation

• Real-time motion capture

• AR live performance capabilities

As these technologies become better, faster, and cheaper, operating a VTuber platform is more scalable than ever before.

Over time, the avatar becomes a portable brand container that can be monetized across channels: content, merch, events, sponsorships, and other fan-oriented products.

One more structural point that matters a lot for the business model: unlike traditional entertainment where a performer can often leave an agency and keep their personal brand intact, VTuber agencies typically maintain far stronger leverage over the “character” identity. When a talent leaves, they generally cannot just take that same character IP with them and continue as-is. The switching cost is huge because “starting over” means rebuilding an identity, a content graph, and a fandom.

That leverage is uncomfortable for some people (and we should name that openly), but economically it creates a very real lock-in dynamic that doesn’t exist in most creator ecosystems.

The industry: from Japan’s internet to global mainstream

VTubing came out of Japanese internet culture in the mid-2010s and became mainstream in Japan in the next decade. By 2020, the category has matured into a full stack of content formats: gaming, music, comedy, variety shows, live concerts, brand sponsorships, and more.

Today, there are two dominant players in the agency model:

• Cover Corporation (best known for Hololive Production)

• Anycolor (best known for Nijisanji)

There are plenty of smaller and emerging agencies in South Korea and China (among others), but they’re sub-scale relative to the top two. Business models rhyme across the category, so focusing on the leaders is a clean way to study the structure.

Tale of Two Agencies: “focused idols” vs “diversified portfolio”

Cover / Hololive (as it is commonly perceived, especially overseas) has leaned into a more “idol + music” orientation with a tighter set of highly prominent tentpole talents and major moments. When it works, it’s extremely powerful: high per-talent output, massive merch/event monetization, and strong international brand recognition.

Anycolor / Nijisanji has taken a different route. It looks less like a tightly curated idol group and more like a multi-category portfolio:

• Idols and musicians

• Comedians and talk-show styles

• Competitive gamers

• Collab-heavy social streamers

• Niche anchors that may never be “global famous,” but are deeply beloved inside their community

This matters because the VTuber market behaves like a long-tail distribution. A small set of talents will always dominate view counts and mindshare, but the real surface area of fandom is spread across a fat tail of niches. The tail is where people find “their person,” and once they do, they stick.

If you accept the long-tail premise, Anycolor’s strategy is basically: own more niches, more consistently, with more redundancy.

Anycolor’s Moat: switching costs + portfolio anti-fragility

1) The agency has unusually strong leverage over the talent identity

As discussed earlier, the character-based identity is hard to port. Leaving often means losing the thing the audience actually bonded with: that exact “mask” and the social graph around it. That creates switching costs that are rarer in the broader creator economy.

2) Anycolor’s roster strategy creates robustness

Graduations and retirements are inevitable in this business. People burn out. Life happens. Scandals happen. Tastes shift.

If your revenue is concentrated in a handful of superstars, your business can look amazing until it suddenly doesn’t. Volatility is the tax you pay for concentration.

Anycolor’s larger cohort strategy acts like a steady cash-cow portfolio:

• One talent can decline or quit without breaking the company

• Multiple niches can grow independently

• The company can replenish the roster over time without relying on a tiny set of perfect bets

Some people interpret “more revenue per talent” as superior efficiency. That’s true in a narrow sense. But in an industry with constant churn and unpredictable hit dynamics, I’d rather own a business designed to be resilient than one that requires consistently big and accurate bets to keep compounding.

This is what I mean by anti-fragility: you don’t need everything to go right for the business to keep working.

Why I think the stock is attractive

Here’s the simple version of the investment case:

• Structurally robust strategy in a churn-prone category: diversified roster reduces single-talent concentration and revenue volatility.

• Founder-led with real skin in the game: founder Riku Tazumi still owns ~40% of Anycolor (as of this writing).

• High-quality economics: 60–70% ROIC.

• Bulletproof balance sheet: ~¥22B net cash.

• Good capital allocation: repurchased ~¥10B of shares in 2024 when the stock was near all-time lows; shifted toward dividends in 2025 after the stock rebounded.

• Undemanding valuation (near historic lows): ~18x P/E and ~12x EV/EBIT.

This combination has all the right ingredients. The market often prices entertainment/creator companies like fragile hit factories. Anycolor is closer to a scaled, repeatable portfolio model with strong control points.

The risks

No thesis is complete without naming what can break it.

1. Talent relations and reputation risk
   This is a people-driven business. If management over-optimizes for monetization and under-invests in talent well-being, you can trigger a slow bleed (low morale → high churn → eroding brand → lower monetization). Even if the IP is controlled, the human factor still matters.

2. Platform dependence
   Even with omni-channel monetization, a lot of attention still routes through platforms like YouTube. Policy changes, algorithm shifts, monetization rules, or cultural backlash can disrupt the business overnight.

3. Pipeline dilution risk
   Scaling a roster can lower average quality if the support system (training, management bandwidth, production) doesn’t scale with it. Portfolio only works if the platform keeps producing and maintaining appealing characters.

4. Overseas execution risk
   Expanding outside Japan isn’t just translation. It’s cultural fit, moderation, community management, and the ability to build local relevance without breaking what made the brand work. It took decades for K-pop to find success in the US, why would it be any easier for VTubers?

Bottomline

To be clear, this isn’t a “Cover is bad” post. Cover has built an incredible brand and has a strategy that clearly works.

But if you force me to choose, I’d rather own the business designed to survive imperfect outcomes.

In a long-tail entertainment market where churn is guaranteed and hit-making is never fully controllable, Anycolor’s diversified portfolio strategy looks more durable to me than a model that relies on consistently creating and maintaining a small set of mega-stars.

At ~18x P/E with net cash and 60–70% ROIC, that durability is exactly what I want to pay for.

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

I was already familiar with Zen and the Art of Motorcycle Maintenance, particularly Pirsig’s idea that the most important motorcycle you’re working on is yourself. But the connection between Robert Pirsig and Christopher Alexander wasn’t immediately clear to me.

Recently, I read The Way of Excellence, and it landed as a clear evolution of Pirsig’s teaching. The author, Brad Stulberg, frames quality as a lot like intimacy—and that’s when it clicked. “Quality,” across architecture, motorcycle maintenance, and nearly any human endeavor, really does seem to share a common root.

Alexander’s view is that a building is beautiful when it matches the living patterns of the community it serves—the building must be intimate with the world it inhabits. Pirsig and Stulberg argue something parallel at the personal level: to achieve excellence in your life’s work, you have to be close to your values. Even in engineering, control theory tells us that a good regulator of a system must contain a good model of that system. The pattern is the same: you thrive when you’re close to what you’re working on (your values, a building’s inhabitants, your motorcycle), and you wither when you’re disconnected from that source of truth.

Turning back to investing, you can see the same dynamic in the corporate world. We’ve watched powerful corporations rise and fall, and the failures often rhyme with “disconnection” while the successes look like “intimacy.” When companies stay close to customers, suppliers, employees, and shareholders, they tend to compound well over the long term. When they become disconnected from too many stakeholders, they turn into a house built on sand. When a company overuses debt to maximize near-term performance, it disconnects from future obligations. When it raises prices without offering more value, it disconnects from customers. When management enriches itself by screwing over owners, it disconnects from shareholders. A lot of what we call low-quality behavior is, at root, just different flavors of disconnect.

We’ve also seen what high-quality behavior looks like. I’d put Berkshire Hathaway close to the ideal: closely aligned with shareholders (e.g., plain-language communication and a long-held bias toward long-term compounding over short-term optics), closely aligned with the businesses it acquires (generally preserving operating autonomy rather than suffocating them with HQ “synergies”), closely aligned with customers (a reputation for straightforward business practices rather than opportunistic price-gouging), and closely aligned with its future (notably cautious about interest-bearing debt, preserving resilience and optionality).

My thinking on “quality businesses” used to anchor on metrics like return on invested capital (ROIC) and free cash flow return on invested capital (FCFROIC), because over the long run they’re among the best quantitative signals for how “good” the underlying business is.

The recurring problem is that this lens can inadvertently reward zero-growth (or even slowly dying) businesses that look great in steady state. There’s no real drive to improve or expand; and sometimes that’s low-quality behavior. Does it really make sense to ignore customers’ need for a better product just to preserve your precious ROIC? Through the “quality is intimacy” lens, that’s low quality precisely because it’s a form of disconnect.

Once I started thinking about quality this way, I could evaluate a much broader set of companies without over-relying on the numbers—like younger businesses investing heavily in brand (low or negative ROIC), or fast growers with heavy CapEx (low or negative FCFROIC).

I’ll leave you with this: Stulberg argues that modern, mainstream views of excellence over-emphasize quantitative metrics (EPS, revenue growth, EBITDA margin). Taken to an extreme, the proxies hijack the real thing. If we’re looking for truly wonderful businesses, perhaps we should spend far more time on the qualitative reality—history, customer relationships, and how management behaves in good times and bad—because those are closer to the real quality than ratios derived from financial statements.

Over the past 18 months, I’ve studied more AI enterprise software companies than I can count. Early on, every pitch felt exciting. Then the wave of trillion-dollar TAM slides, startup/microcap “AI disruptors,” and slick investor decks hit—and somewhere along the way I went from bullish, to bearish, to something closer to neutral.

What confused me most was the disconnect between theory and reality. Most of these AI enterprise software companies are essential business process outsourcing (BPO), the difference is that humans are replaced by AI. In theory, BPO is an ugly industry: hyper-competitive, low switching costs, and little differentiation except for scale. If everyone has equal access to AI, nothing should change—everyone can automate the same workflows with the same models. Eventually, competition would compress margin back to historical norms.

But in practice, a handful of AI-native operators—many still sub-scale—are winning real contracts and growing quickly. Why are some players breaking out while others stall? What’s durable, and what’s just early-mover noise?

Below is the five-rule framework I that distills my thinking so far. It is still evolving, and I’ll probably change my mind again. I admit the title is bit misleading. But “5 rules framework” is more appealing than “5 lessons subject to revision,” and it captures the patterns I see across the most promising operators.

None of these rules stand alone (great companies usually check several of them at once) nor do they replace valuation work. But they help separate durable business models from AI slop.

Rule 0 — Execution still matters

Before anything else: if a company can’t win customers or ship excellent products, the rest is irrelevant. A lot of the early “AI BPO winners” are probably just good operators riding temporary momentum. Without the other rules, that momentum won’t last.

Rule 1 — Own a mission-critical workflow with asymmetric cost of failure

The product must run a workflow where failure creates regulatory, legal, financial, or operational damage so severe that switching becomes irrational.

• Cost of failure must be ≥10× the benefit of switching.

• Common in financial reporting, compliance, settlement, legal filings, credit decisioning, and liability-sensitive workflows.

• Also includes essential but non-core work customers must outsource (e.g., disclosure systems, registry lodgement, compliance controls).

Test: If the system vanished today, would the CEO/CFO/COO call an emergency meeting within 48 hours?

Rule 2 — Control the entire workflow, not just a feature or tool

The company must act as the orchestrator or operator of the process, not just a copilot or tool. It needs to deliver the outcome, not just enable it.

• Must sit in the “system of record & action” layer. For example, AI to automate financial disclosure workflow would be intake → data cleansing → validation → document generation → submission → monitoring.

• If Microsoft/SAP/Salesforce/Epic could bolt on the same feature and kill the business, it fails.

• If removing the vendor doesn’t break the workflow, it’s just a tool.

Test: Can customers keep their existing platforms but remove this vendor without the workflow falling apart?

Rule 3 — AI improves unit economics, but is not the value proposition

AI should materially reduce variable labor and increase throughput, pushing the model toward SaaS-like margins—but customers shouldn’t buy it because it’s AI.

• AI serves as a labor substitute and throughput multiplier.

• The business should still make sense in a world where “AI” is not a pitch.

• If differentiation disappears once competitors access similar models, the advantage was not real.

Test: In a world where everyone has the latest LLM, does this company still have superior economics because of where AI is embedded and what proprietary data it can learn from?

Rule 4 — Closed-loop proprietary data + process flywheel

Operating the workflow must generate feedback loops that improve automation accuracy and deepen switching costs.

• Includes raw data (documents, filings, exceptions) and process knowledge (playbooks, escalation paths, regulator feedback).

• Data must be workflow-native. This is something general-purpose AI models cannot replicate from public data.

• Scale → More data captured → Improved product → More customers → More data captured → etc.

Test: Could a general-purpose model match performance without this vendor’s proprietary workflow data? If yes, the moat is weak.

Rule 5 — Small market share, reinvesting aggressively into the same workflow at high returns

The best companies are early in their penetration curve and reinvesting into the workflow they already dominate.

• Look for high LTV/CAC, fast payback, strong customer retention, and high incremental ROIC.

• Expansion should be depth-first (more value in the same workflow) or into adjacent areas with similar switching cost.

Test: Is every new dollar of R&D/S&M clearly aimed at strengthening the same mission-critical workflow?

At the end of the day, this isn’t a perfect or permanent framework, but it’s the clearest lens I’ve found to separate real, durable AI businesses from the noise. The gap between theory and practice in AI-enabled service business requires deep understanding of unit economics to reconcile.

The companies that matter will be the ones that anchor themselves to mission-critical workflows, build genuine moats, and let AI amplify their value. My hope is that these five rules make it a little easier to spot them.

In Superforecasting, Philip Tetlock makes a humbling observation: most of us are much worse forecasters than we believe. The cure, he argues, is not brilliance. It is the disciplined use of cognitive diversity and constant self-correction. Michael Mauboussin’s More Than You Know echoes this from the perspective of investing. The best investors draw from multiple disciplines so they can see problems from more than one angle. They mix the outside view, anchored in base rates, with the inside view shaped by their own experience.

Tetlock uses the image of a dragonfly eye. Each facet of the eye captures a slightly different glimpse of the same object. When combined, the full picture becomes sharper and more accurate. Mauboussin arrives at the same idea through two practical tools. First, start with base rates, a point borrowed from Kahneman. Second, push decision-making toward decentralization so the group does not fall into the same trap at the same time.

For an individual investor, achieving this level of cognitive diversity is difficult. You can read widely and build what Charlie Munger calls a latticework of mental models. You can train yourself to switch between frameworks. But your line of sight will always be shaped by your history, your habits, and your priors. No one person can replicate the breadth of perspective of a strong forecasting team.

That led me to two questions. Could AI give an individual investor access to something resembling a team? Can AI become more useful R2D2-like assistants that does more than just provide basic summarization?

Building a “Digital” Superforecasting Team

With AI agents now easy for non-technical users to build, I decided to try creating a small investment “team” composed of agents with different worldviews.

I started with five roles.

• Outside-View Analyst: focuses on identifying appropriate reference class and its base rates (e.g., average growth rate of companies with similar size) to provide historical context.

• Inside-View Specialist: leans on analyzing the business through the perspective on domain/industry-specific knowledge (e.g., this industry’s historical performance, competitor’s profit margin).

• Systems thinker: focused on understanding the business’ surrounding ecosystem, value chain, and how key stakeholders interact with it.

• Incentives analyst: applies a behavioral lens on the management team, particularly the compensation structure, ethical track record, and stock ownership.

• Devil’s Advocate: examines other agent’s work and looks for holes, contradictions, and weak assumptions.

Each agent researches the same company independently and forms its own thesis. Only after this first step do they “meet” to compare notes, critique one another, and revise their thinking. The dynamic is meant to imitate how Tetlock’s superforecasters iterated toward better answers: independent views first, constructive debate second, convergence last.

Here’s roughly what the workflow looks like:

The results were mixed. The team still made odd mistakes—for example, the agents mistakenly concluded that Medpace’s revenue concentration is “highly concentrated” because 90% of revenue comes from small-mid biotechs. I’m pretty sure there is more than one small-mid biotech company in the US.

However, the underlying concept feels promising. When I examined under the hood to see how each agent approached the task, their thought processes were genuinely distinct from each other. The problems emerged when I forced them into premature consensus. At that point, the quality of the analysis collapsed into generic AI output.

There’s still work to do, but the early signals suggest that a well-designed multi-agent structure can create something closer to true cognitive diversity.

Where the Prototype Started to Strain

There was one limitation that emerged almost immediately. When the agents conducted “deep research,” they kept pulling from the same public sources. That is simply a consequence of running on the same LLM with access to the same websites. Even though the frameworks differed, the inputs often converged.

To get around this, I had to manually supply them with richer information. This included financial statements, filings, proprietary transcripts, industry datasets, and primary research. Once this additional material was uploaded, the agents produced more distinct and more interesting views. The quality of their work was much closer to what I had hoped for.

It became clear that diversity of thinking only works if the underlying information is also diverse. Otherwise, the team looks different on the surface but ends up drinking from the same well.

Why This is Promising

Even with the gaps, the early results are encouraging. With the right structure, AI agents can behave a little like a superforecasting group. They generate independent perspectives. They challenge each other. They revise their views. And the output is usually less “generic” than any single AI’s first pass.

This experiment also opened up new questions.

• What mix of mental models makes for the strongest team?

• How many agents are enough, and how many are too many?

• Can one agent combine multiple mental models effectively, or does that reintroduce bias?

• Does a small, specialized team beat a larger, more diffuse one?

I don’t have the answers yet. But the idea that one person could simulate the strength of a well-functioning investment team feels new and exciting. Tetlock showed that good forecasting is the result of process and structure. Mauboussin showed that judgment improves when investors deliberately choose different vantage points. Munger argued for building a broad toolbox of mental models.

For the first time, technology might let an individual investor stitch these ideas together in a scalable way.

It is said an Eastern monarch once charged his wise men to invent him a sentence, to be ever in view, and which should be true and appropriate in all times and situations. They presented him the words: “And this, too, shall pass away.” How much it expresses! How chastening in the hour of pride! How consoling in the depths of affliction!

Abraham Lincoln, recounting the ancient Persian fable in 1859

New Oriental is a high-quality, dominant franchise in a structurally attractive niche. It survived the 2021 regulatory shock, rebuilt profitability, and is now consolidating a fragmented market. Yet Chinese after-school tutoring (AST) remains ‘untouchable’ for most investors, creating a large gap between EDU’s price and its underlying brand and scale advantages.

A scarred industry rebuilding

New Oriental is a private education provider focused on after-school tutoring and overseas college prep (e.g., SAT, TOEFL, LSAT). The AST market remains highly fragmented, with tens of thousands of sub-scale local players and intense competition. Even as the largest player, EDU only had about 1% market share in 2023. TAL is the second largest AST business and is a key competitor on a national scale.

In 2021, the K-12 AST industry was effectively crushed by the “double-reduction” policy, which briefly outlawed tutoring. Enforcement was often brutal, with anecdotes of tutors being arrested in police-raids. Thousands of weaker operators have exited or perished, while leaders like EDU and TAL have been taking share. Even though the harshest measures have eased, marketing for tutoring remains heavily constrained, which makes pre-2021 brand equity and word-of-mouth the primary demand drivers.

With regulation now more stable, EDU is positioned to keep gaining share from smaller players, driven by its brand recognition, national footprint, and scale economics.

Why EDU is structurally advantaged

EDU’s core business is highly cash generative and capital-light. The main costs are classroom rent and teacher salaries. Parents typically pay most fees up front, which creates a low or even negative cash conversion cycle. This is effectively free, non-interest-bearing funding (“float”) that can be reinvested to compound value.

Historically, the biggest weakness of many AST businesses has been talent portability: star tutors often have stronger personal brands than the institution and can walk out to start competing centers. That dynamic has kept the market perpetually fragmented.

Today’s regulatory environment changes that equation. Because AST players and independent tutors cannot openly advertise, new entrants have to rely almost entirely on word-of-mouth. That massively favors incumbents like EDU and TAL, whose brands are already top-of-mind for parents. The marketing ban therefore does two things at once:

• It pushes demand toward the largest, most trusted brands.

• It makes it much harder for sub-scale players to gain traction and become national competitors.

If these rules persist, they will likely act as a growth flywheel for EDU: existing brand awareness drives new enrollments, which further reinforces word-of-mouth and perceived quality.

In addition to its brand power, EDU also benefits from having a massive library of education content (e.g., past exams, study guides, etc.) it has built over the last two decades that can be shared across its national network. This is a scale-driven cost advantage that smaller AST providers do not have.

Financials that don’t look “distressed” at all

EDU’s recent financials underline the strength of the franchise:

• High ROIC: After adjusting for excess cash and short-term equity investments (primarily its controlling stake in East Buy, which should not be treated as working capital), EDU is generating roughly 40% ROIC.

• Earnings power above pre-crackdown levels: Aside from 2022, EDU has been consistently profitable. Operating profit has rebounded to about $500M LTM, above the $400M pre-2021 peak.

• Balance sheet strength: Through the worst of 2022, EDU still held at least $1B in excess cash while many sub-scale competitors were forced into debt-funded survival. The contrast in resilience is stark.

This is not a wounded asset limping out of a crisis; it is a leaner, more dominant EDU operating in a less crowded market—much like how fire-resistant redwoods benefit from occasional wildfires that wipe out the competing plants.

Valuation: the market is still fighting the last war

Despite this setup, investor sentiment remains skeptical.

At around $8.8B market cap, with $1.28B in cash, $3.3B in its East Buy stake, and $780M of debt, EDU’s enterprise value is roughly $5B. Over the last twelve months, it generated about $600M in free cash flow, implying an EV/FCF multiple of ~8x. For a leading consolidator with a strong balance sheet and high ROIC, that is extremely cheap.

A reverse DCF on these numbers suggests the market is pricing EDU for only 3–5% annual growth over the next decade. That is well below reasonable base rates:

• From 2012 to 2019, the number of Chinese students studying abroad grew roughly 9% per year.

• EDU is not only exposed to that demand but is also taking share from weaker players in a still-growing, partially rebuilt market.

A more plausible mid-case is growth at or above market rates, not half of them.

So why the pessimism? Most likely, investors are still traumatized by the 2021 wipe-out and are over-discounting regulatory risk—similar to how financial stocks traded for years after the 2008 crisis. The business has moved on; the narrative has not.

How value can surface even if sentiment doesn’t heal

Even if global investors stay on the sidelines, EDU does not need the market’s enthusiasm to create shareholder value. Management has multiple levers:

1. Buybacks: EDU has a buyback authorization of $300M over the next 12 months and has already repurchased about $580M of stock this year.

2. Ordinary dividend: Management has committed to returning at least 50% of net income as dividends going forward.

3. Potential special dividends: EDU could accelerate value realization by distributing more of its excess cash balance in the future, even though this has not yet been done.

At an 8x EV/FCF multiple, aggressive repurchases and disciplined dividends can compound per-share value even if the headline valuation multiple never re-rates.

A broader lesson on where to fish

New Oriental is a good example of the kind of opportunity that can emerge in high-quality small and mid caps operating in out-of-favor markets such as China, Mexico, or South Africa. These markets are often shunned for macro or political reasons, even when individual companies have dominant positions, strong balance sheets, and disciplined capital allocation.

Meanwhile, reasonably priced ideas in the US—especially in popular tech/AI stocks—are increasingly scarce as capital crowds into the same names.

The implication is straightforward: if you want attractive risk-adjusted returns over time, you need to be willing to look where others aren’t. EDU is one of those cases. Make sure you’re fishing where the fish actually are, not just where everyone else has decided to cast their lines.

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

Recent discussions in financial media and among investors show growing unease about AI infrastructure spending and whether we’re inflating one of the largest bubbles in history. 

Even insiders admit the tension. Sam Altman, hardly a skeptic, concedes that speculative dynamics are forming. Yet he and his peers are trapped in a prisoner’s dilemma: under-investing feels riskier than over-investing, and faith in long-term productivity gains justifies almost any burn rate.

Still, “bubble” and “benefit” aren’t mutually exclusive. Like the railways, canals, electric grids, or dot-com companies of past centuries, the AI boom is both an overreach and a technological foundation. Those earlier manias ended in bankruptcies but left behind assets that powered decades of growth. The better question, then, isn’t whether this is a bubble—but how the system behaves under stress.

What the AI Capital Cycle Does and Does Not Resemble

Most famous bubbles are the wrong map. The dot-com boom was about unproven websites and ad models; 2008 was about household leverage and bad collateral; Dutch tulip mania of 1636 left nothing useful behind.

A closer parallel lies in the late-1990s TMT and fiber build-out. Telecom firms installed networks years ahead of demand, many went under, yet the physical backbone became indispensable. That industrial pattern—massive capital outlays, delayed returns, eventual utility—mirrors today’s AI spree in GPUs, data centers, and power.

But to understand the interweaving technological innovation and financial psychology of the moment, the sharper analogy comes from an older episode: the Great British Bicycle Bubble of 1896.

The Great British Bicycle Bubble

In the 1890s, a wave of inventions—pneumatic tires, chain drives, lighter steel, the diamond-frame design—made the bicycle fast, comfortable, and suddenly attainable. For the first time, ordinary people could move freely without trains or horses. Demand soared. Finance followed.

Between 1895 and 1897, about six hundred bicycle firms went public in London, and share prices tripled before gravity returned. Promoter Ernest Terah Hooley embodied the mania. He bought Pneumatic Tyre for £3 million (mostly funded with bank loans), rebranded it as Dunlop, hyped it furiously, and flipped it for £5 million—a Victorian leveraged buyout masterstroke that sparked a gold rush of imitators.*

As production boomed, margins collapsed. Cheaper American bikes flooded Britain at half the price. By 1901, more than seventy per cent of local makers had disappeared. Yet the bicycle endured, democratizing mobility and paving the way for motorcycles and cars. Technology triumphed. Capital did not.

AI is walking a similar path. Transformer breakthroughs made automated reasoning cheap and scalable. If bicycles freed people from horses, AI frees them from boring work. But progress has again fused with speculation. Financialization now amplifies every genuine gain. The winner-take-all narrative—“invest or be left behind”—feeds reckless spending that assumes growth without end. History suggests otherwise: tech winners can emerge after the crash as well—Uber, Tesla, and Meta are the most prominent examples.

*Note: Hooley was later exposed as a serial fraudster—repeatedly bankrupted and convicted for false pretenses in other ventures.

The Mt. Hood Disaster

In 2002, a roped team on Oregon’s Mount Hood suffered a fatal chain fall. On hard, early-morning ice and a steep slope, the lead climber slipped. With no intermediate anchors (pickets, ice screws) to absorb the force, the rope transmitted the fall downhill, yanking each teammate off in sequence and dragging multiple parties into a pileup. Rope travel works on moderate terrain because partners can arrest a minor slip with their axes; that everyday success creates a false sense of security. But when the leader falls on hard, steep ground—where momentum builds instantly—the same rope becomes a coupling device that multiplies risk, turning one error into a system-wide failure. As one experienced Hood climber put it, “a rope without fixed protection is a suicide pact.”

Today’s AI ecosystem resembles that rope. Model developers such as OpenAI and Anthropic set the narrative—“if we build it, they will come.” Hyperscalers like Microsoft, Google, and AWS translate that belief into massive CapEx and long-term contracts. Suppliers, lenders, and startups are tied in through equity stakes, take-or-pay agreements, and debt secured by projected demand. The system can absorb small slips—a missed quarter or delayed launch—but a major stumble at the top can cascade through the chain.

As Bloomberg’s network maps show, the interconnections run deep: Microsoft and Nvidia hold stakes in OpenAI; Amazon owns part of Anthropic; OpenAI holds warrants in AMD; smaller firms like CoreWeave borrow against rapidly depreciating GPUs. What looks like diversification tightens the knot—everyone’s safety line is tied to the same deadly rope.

History in Real Time

The AI surge carries both signatures—it is an industrial and a financial bubble at once. Like the bicycle mania, the technology is transformative and will outlast the bust. But the dense web of cross-holdings, long-dated contracts, and debt-financed build-outs has turned what could have been a healthy investment cycle into a tightly coupled system. Stress at one node now travels instantly to another.

The smaller, leveraged players—GPU lessors, data-center developers, PE-backed power suppliers—are climbing without anchors. When liquidity tightens or utilization disappoints, they’ll tumble first. The giants—Microsoft, Nvidia, Amazon—are more likely to survive and consolidate the wreckage, but not before a costly shake-out.

For investors, the lesson is plain. Bubbles built on leverage and interdependence last longer than reason allows, yet when they break, they fall fast. AI’s promise is real; the coming destruction, equally so. The wise move is to admire the technology, respect the cycle, and stand clear while the music is still on.

Source: Bloomberg, Nasdaq, Deep Survival by Laurence Gonzales

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

"What is elementary, worldly wisdom? Well, the first rule is that you can’t really know anything if you just remember isolated facts and try and bang ‘em back. If the facts don’t hang together on a latticework of theory, you don’t have them in a usable form.” — Charlie Munger

My Journey Thus Far
When I started investing, I was taught investing could be reduced to spreadsheets and multiples. Therefore, investors should focus on predicting how a company’s income statement, cash flow, and balance sheet will evolve in the next 3-5 years. The path forward seemed simple: learn to use these existing tools and the rest follows.

As I venture deeper into my investing journey, I have a growing conviction that there is tremendous value not just in digging deeper, but in exploring broadly. While it is helpful to practice the basic skills of investing (how top investors think about investing, how great business leaders think about strategy, how experts dissect business models), I experienced a wall of
diminishing returns in this endeavor.

Studying the writings and philosophies of great investors, I noticed that many of the lessons can be coalesced into the following statements:

• Protect the downside first, everything else is upside

• Buy high-quality businesses at sensible prices that leave room for compounding

• Stay within your circle of competence: craft a simple, falsifiable thesis and try to disprove it

• Think in base rates and probabilities, ignoring macro noise that doesn’t change intrinsic value

• Pursue variant perception, act with conviction, let winners run, and fix mistakes ASAP

Although these concepts are most likely true, I’ve always disliked mindlessly echoing quotes or heuristics as if they’re immutable truths: they are intellectually lazy. In the past, I struggled to internalize them without addressing the what’s and why’s. Here are some of my questions:

• Why prioritize downside? Why not maximize the expected outcome?

• Is there a consistent and objective definition of quality across industries?

• How do you reliably know your circle of competence and avoid hubris?

• How stable are base rates? Can they be applied to emerging or rapidly changing sectors?

• How do you create contrarian views that are right?

In the world of investing, reliable case studies are sparse, path-dependent, and difficult to transfer, so pure “inside-view” learning hits diminishing returns. Similar to Daniel Kahneman’s discussion on outside view versus inside view, I found myself learning so much more about investing when looking at it from the “outside”—think biology, systems engineering, statistics, history, literature, etc. Studying each of these fields helped me become a better investor.

Inside View vs. Outside View
The baby steps of investing is to learn the accounting language, familiarize with setting up DCF models, take to heart the importance of capital allocation—this is the bare minimum.

Beyond corporate finance, theory, and economics, most investors would further their expertise by specializing in a handful of sectors, like software or banking. Some investors would go a step further to learn about niche investment situations like M&A and spin-offs. A minority of investors would
dive into cutting edge areas like behavioral finance and quantitative investment strategies. I believe investors should not stop here.

There are three non-finance domains I find useful for understanding investing. For each, I’ll outline what it is, why it matters, and how it relates to investing:

1. Complex adaptive systems (CAS)

The modern economy and its stock markets are among the largest CAS emerged from our species’ history. Short term macro outcomes are extremely difficult to predict—think the famous butterfly effect. This is a cliche but true to all CAS: rainforest ecology and neural networks in your brain (and LLMs) are similarly black-box-like on a macro level because their
emergent behaviors are driven by self-organized patterns.

Lesson: For fundamentals-driven investors, focus on the non-chaotic aspects of the organization you’re studying. That is: business model, returns on capital, margins, bargaining power, reputation, incentives. Each of these traits has a different level of durability, but all are more tangible than today’s stock price.

Interestingly, some CAS can also exhibit much higher levels of problem-solving ability than its components. In finance, this is the efficient market hypothesis. In nature, we see this in ant and bee colonies—tiny creatures when working together can solve highly complex issues (e.g., finding food sources, optimizing hive location).

Lesson: Having a more diverse set of mental models will amplify your problem-solving ability. Armed with multiple ways to solve a problem, even an individual can mimic the wisdom of a crowd and outperform single-minded competitors.

2. Evolutionary biology

Nautiluses, a cute sea creature that looks like a coiled seashell with a squid that lives inside it, are the last living descendants of the ancient cephalopod lineage that has persisted since the Cambrian Period (~500 million years ago). This little creature survived five mass extinction events!

As you can tell from its looks, Nautiluses are not particularly smart, fast, or strong. The nautilus survived for eons simply by occupying a deep-water refuge, adopting a low-metabolism lifestyle, and laying eggs away from the surface. Given this creature’s longevity, there may be some benefit to being the ultimate shut-in.

Lesson: It doesn’t matter how smart you are or how much financial engineering you’ve implemented. One fatal error can end your investing career. Think about the downside before dreaming about the upside. How much money will you lose if things do not go your way? What if the company falls into a sinkhole right before an M&A transaction was about to close?

You laugh, but look up what happened to Florida Cypress Gardens in 1985.

3. Metaphors & Analogies

In literature, metaphors map the familiar to the new or connect seemingly unrelated concepts. Besides helping generations of students to pad their English essays, why does metaphor matter?

Metaphors bring distinct ideas together based on structural similarities. Not all metaphors are created equal—most are just surface-level comparisons, while others are deep structural connections. See examples below:

• This memo is a brick—heavy, blunt, and hard to digest.

• Debt is a time machine—it moves future cash flow to the
  present, at the expense of the future.

As you can see, one underscores a point while the other adds a new perspective without being redundant. Perhaps that is why wise words sounds poetic.

Lesson: Analogies are some of the most powerful and versatile tools for connecting ideas. This is helpful for understanding new concepts, dissecting novel business models, assessing new product launches. And through the exercise of creating analogies, an investor inadvertently tests the limit of their knowledge.

An additional benefit to using metaphors: there is clinical evidence (Bjork, 1994) suggesting that learnings/memories can be enhanced through efforts and struggles, because neural connections are strengthened when it is connected to more neurons in the brain (new connections form whenever you experience difficulties). In other words, purposefully creating analogies is like taking on heavier weights for strength training—breaking muscle fibers will yield greater strength in the long run.

Bottom Line
These three mental models are not the silver bullet to investing but are just the starting points. The more we widen our lens, the better we see what matters and what lasts. My goal is to keep building this list, and I invite you to do the same: collect models wherever you find them, and put them to work.

For the past several months, the catchy tune "darling take my hand... nothing beats a Jet2 holiday" has gone viral across the English-speaking internet. On Tiktok, Jet2's advert is mixed with videos of chaotic vacation experiences, unruly tourists, and other travel-related disasters. Interestingly, Jet2’s meme moment spotlights a company whose disciplined, vertically integrated engine delivers reliable, good‑value travel. Unlike many other garbage meme stocks, this memetic phenomenon reveals a real compounder.

Founded as a regional airline Jet2.com, Jet2 plc is now primarily a package tour operator, better known as Jet2holidays. Jet2's business model shifted from a purely airline business to selling bundled trips that include flights, hotels, and transfers, so holidays now drive ~80% of revenue.

Why Travelers (and Competitors) Love Jet2

Importantly, Jet2 is known for its focus on quality customer experience and delivering good value for price. Customer surveys in the UK consistently rank Jet2 and Jet2holidays #1 in their respective sectors. Even competitors credit the company for lean execution, plain‑spoken promises, and delivery as advertised, which builds trust.

“I actually went on a Jet2 holiday… Everything worked seamlessly. Everything was on time, everything was as described. It was a good experience.”
– Former Director of Customer Experience at LoveHolidays

How does Jet2 achieve this level of quality consistently over the last 20+ years? Jet2's operating principle seeks complete control over critical steps across the customer/user journey -- it owns and directly operates the aircraft, baggage handling, snacks on the airplane, routing, local reps, etc. In short, Jet2 controls the entire process to reduce points of failure even in times of distress and external shock.

In 2022 and 2025, Heathrow shutdown and ATC outage paralyzed nearly all UK flights. In the midst of this chaos, Jet2 continued to operate and deliver vacationers to their destinations. Jet2's consistency is their brand signature -- and their competitors know it too.

"Jet2 actually, to be honest with you, is a really impressive business. They are extremely lean. They give very honest opinions, reviews on their holidays. They don’t have anything fancy. It’s just basically a flight, a hotel, a transfer.”
– Former regional lead at TUI

Under the Surface Thinking

At this point, I can imagine some value investors coming out with their metaphorical pitchforks shouting, "Blasphemy! Warren Buffett says to never invest in airlines!" While I agree that passenger airlines typically make terrible businesses in the long-run, Jet2 fundamentally operates differently:

1. Jet2 is primarily a tour operator, only 20% of its revenue is from seat-only sales. Expecting Jet2 to behave like a regular airline would be an extreme take. Its most relevant comparators are online travel agencies (OTAs) in Europe like TUI, On the Beach, Expedia, Booking.com.

2. While Jet2's airline business remains just as capital intensive as any other airline, it operates with better line-of-sight into its future demand. If you know where your customers are going for vacation, you can quickly adjust your routes to meet this demand. This translates into a better capital allocation in the long-run.

Besides differences in business model, Jet2 also presents a flywheel that is rare and powerful: scaled economies shared -- this is a self-reinforcing loop in which more volume lowers unit costs, those savings are passed to customers, lower prices attract more demand, and the cycle compounds not just margin but also loyalty. This model isn't a novel concept, retail giants like Costco and Walmart built their business with scaled economies shared. Still, some successful budget airlines benefit from this flywheel as well. However, the concept itself isn't difficult to adopt -- it is much harder to execute and maintain this discipline across cycles. 

Upon a closer look, we can see strong signs of this flywheel happening. Over the last decade:

• Revenue increased 7x (from £1 to £7 Billion)

• Jet2Holiday average unit price rose about 3.4% a year, a touch above UK inflation at ~2.6%.

• Operating margins and ROIC stayed steady outside the COVID years.

• ROIC averaged ~14% with minimal debt.

When examined together, these signs suggest that Jet2 transfers all the cost savings directly to keeping prices low. Importantly, this price is still high enough to generate an above-average ROIC.

Why can't other tour operators copy this formula? It comes down to discipline -- the moment the management team becomes complacent, the flywheel falls apart. Fortunately, Jet2's governance and incentives support this discipline, with the founder, Philip Meeson, retaining a meaningful stake of ~14% and the CEO holding equity that is 5.5x his cash compensation. However, I'd like to see the CEO increase his equity stake.

Investment Thesis

Disciplined Execution: The management has demonstrated a track record of focus and skills to execute and maintain the scaled economies shared flywheel. Excess capital is returned to shareholders, as appropriate. Evidence: consistently low price, high ROIC and no (net) debt.

Ownership Over Customer Experience: Jet2 owns all the critical steps along the customer journey means that it can and does deliver the most seamless, most consistent vacation experience for the price. Evidence: customers love Jet2, even the competitors.

Anti-fragile Model: Jet2 runs a decentralized base and operates its own supporting services (e.g., baggage handlers) -- this enables Jet2 to minimize disruptions from a single vendor or a major airport. This prevents the business from achieving maximum efficiency, but it pays off in the long run. Evidence: while its competitors floundered in 2022 and 2025 disruptions, Jet2 cancelled no flights and kept going.

Undervalued by Every Metric

Despite an impressive track record, the market prices Jet2 like a declining airline by implying negative long‑term free cash flow growth (about -3% p.a.). Valuation based on cash flow and earnings growth point to materially higher fair value:

DCF Based on Free Cash Flow Growth

• Assumptions: FCF growth: 7%, terminal P/FCF at year 10: 5x

• Discount rate: 12%

• Implied fair value: ~£30/share

DCF Based on Earnings Growth

• Assumptions: earnings growth: 15%, terminal PE at year 10: 7x

• Discount rate: 12%

• Implied fair value: ~£32/share

As of 8/1/2025, Jet2 trades at ~£16/share. This is a severely suppressed valuation when adjusted for its net cash position (~3x EV/FCF and 3x EV/EBIT)—despite having customer loyalty and stable returns on capital. The market is pricing it like a shrinking airline, not a resilient tour operator.

Despite the upside opportunities, I'd highlight a few items to watch:

• Does the management remain focused & disciplined in executing its not-so-secret formula?

• Is there any drift in corporate culture?

• How do competitors in UK respond to Jet2's success?

Bottomline

Jet2 is a rare breed. It scales responsibly, earns customer trust, and generates attractive capital returns. It doesn’t rely on pricing power or luxury positioning. Instead, it wins through execution, value, and discipline.

It’s the top-rated airline in the UK, one of the most beloved brands in European travel, and now the subject of a viral meme. That cultural moment only adds to its brand strength.

Call it a meme. Call it a compounder. 

Either way, Jet2 is a business worth holding on to.

Source: Jet2 Investor Relations, AlphaSense Expert Interviews, UK customer satisfaction index, Which?

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

“We often ask, 'Where are the clouds today?' These cloud can obscure the long-term, extraordinary parts of a business."
— Christopher Begg

UnitedHealth Group (UNH) sits at the epicenter of a national debate—its “pay-vider” model is now under challenge by rising regulatory scrutiny and rapidly eroding brand value. This write-up is my take on the situation. 

The Payvider Model: UNH’s strength lies in its vertically integrated payer-provider combination. In 2011, UNH launched Optum, the provider side of the business, and has since consolidated a massive network of healthcare providers, pharmacy, and data analytics.
This deliberate, decade-long build-out created an effective feedback loop: data from tens of millions of members informs insurance underwriting, risk stratification, and early interventions—driving a consistent 4–5 percentage point advantage in medical cost ratios (this roughly reflects the COGS) over peers.

Competitors have yet to replicate this model at scale, but it’s not for lack of trying:

• Cigna launched its provider division (Evernorth) in 2020 and raced to acquire assets like MDLive and Bright.md. The strategy? Fast-track integration to rival Optum’s data and care loop. The result? A costly stumble. Cigna recently reported a $300 million loss tied to its investment in VillageMD clinics—a sign that duct-taping together a network of providers doesn’t make you a payvider overnight. Moreover, Evernorth remains structurally fragmented: telehealth, pharmacy benefits, and analytics continue to operate in silos, lacking the unified coordination that defines Optum.

• Humana, another potential challenger, has made real progress—especially within Medicare Advantage—by integrating care management and clinical coordination. But its reach remains narrow. Humana lacks the broad infrastructure across PBM, clinics, and analytics that UNH operates nationally.

• Kaiser Permanente, the original pay-vider, boasts true integration. But as a nonprofit operating in just eight states and D.C., it lacks both the scale and the reinvestment cadence to match UNH.

So yes, UNH’s model is unique—built slowly and deliberately. Competitors who try to shortcut that timeline find themselves with patchwork systems and costly missteps. That depth and durability explain both UNH’s moat and why regulators are now circling.

In the broader healthcare world, combining funding and care delivery isn’t new. In fact, it's the standard in many advanced economies. The NHS of England is a good example: government funds care and provides it under one roof. But in the U.S., the payvider model only gained traction in the past decade. Why? Because insurers finally realized that if you want doctors to conserve scarce resources, you’d better give them financial skin in the game.

Controversy and Uncertainties:
In an unprecedented and shocking event, UNH’s then-CEO was murdered in NYC after its annual investor day. Since the shooting, discontent from all corners of society has poured out against UNH and other major insurers. Worse, the company now faces federal investigations into its billing practices and mounting criticism over how it uses prior authorization to control costs (that often hurts the patients).

It’d be an understatement to say that UnitedHealth’s brand has taken a reputational hit.

These aren’t just technical disputes; they may cut to the core of UNH’s operating model. Will regulators rewrite the rules? Could the very structure—tying payer and provider—be forced apart? It’s unlikely for now. The White House rhetoric hasn’t pointed that far. But that’s the worst-case tail risk investors are chewing on.

The Moat:
This is where things get more nuanced. UNH’s margin advantage isn’t built on clever accounting. It comes from a long, patient investment in infrastructure: risk models, data, provider networks. Even if regulators curb more aggressive billing strategies, the underlying engine—scale and coordination—likely remains intact. And the company’s financials? Still solid. UNH generates robust free cash flow across multiple profit centers, giving it room to absorb hits, whether from litigation or shifting policy.

Mr. Market's Mood:
Now here’s where the math gets interesting. Based on a reverse discounted cash flow model (12% discount rate, 10x terminal multiple), the market is effectively pricing in just 3.5% annual growth in free cash flow per share over the next decade. That’s well below UNH’s 10-year average of 13%. In short: the market’s baking in a worst-case future.

That disconnect creates an opportunity. Unless regulators go nuclear—splitting the company or flattening its payment structures—UNH’s core engine is likely to keep running. Maybe slower. But not dead in the water.

If the regulatory fog lifts—or just becomes easier to navigate—investors may reprice the stock upward. In that light, today’s pessimism looks like an overshoot.

Bottom line:
UNH is in a tough spot. The model is durable, but the optics are bad and the scrutiny is real. The challenge is to separate short-term noise from true structural risk. Right now, Mr. Market isn’t making that distinction.

The best investments tend to be simple, easily understood, and overlooked by others.
Seth Klarman

System of record (SOR) has created some of the most dominant and durable franchises in tech. Think Salesforce, SAP, Epic. But one of the most compelling SOR opportunities today isn’t in Silicon Valley — it’s in Osaka.

Ok, That’s Great… What is an SOR Again?

A system of record is an authoritative source of data that captures business-critical information, such as customer relations (CRM), enterprise resources (ERP), and laboratory notebook (LIM). These systems are often deeply embedded in workflows and exceptionally sticky. Why? Three forces drive their defensibility: 

• Data lock-in: Once business data resides in an SOR, switching becomes expensive and disruptive. 

• Internal network effects: As more employees engage with the system, its utility and switching cost grow. 

• Product adjacency: SOR vendors can layer on adjacent modules, expanding use cases and deepening entrenchment. 

When these forces meet, they create powerful feedback loops for profitable growth. Most importantly, this helps build an almost insurmountable moat against challengers in this space.

SOR is no stranger to the tech world. Most of Wall Street is familiar with this model as well. In fact, most equity analysts can list numerous SOR examples on the spot.

In the last decade, it has evolved from a novelty to a buzzword in the venture capital world. Countless start-ups claim to be the "Salesforce for X", but few deliver.

What if I told you that there is a rapidly growing, GAAP profitable SOR hidden in plain sight?

Unlike Salesforce, which has nearly saturated its domestic market, this hidden SOR is a micro-cap with significant runway for growth. Look no further than Osaka, Japan. Although Japan's third-largest city often flies under the radar for most Western investors, it remains a welcoming and accessible destination for foreign capital.

From Hardware to ERP: a Three-Decade Detour

In 1991, Tetsuo Iwamoto founded I’LL Inc., a small business selling PCs. During its first decade in existence, Iwamoto experimented with several businesses, including computer maintenance and IT consulting. During the Dot Com frenzy of the late 90s, the company entered into several online businesses — a job matching platform and an online healthcare business. None of these businesses reached any meaningful scale. 

After several false starts, I'LL launched Aladdin Office, an enterprise resource planning (ERP) software. This ERP helps small and medium size businesses (SMBs) manage their customer relations, purchasing, inventory, and production. In the next decade, I’LL incrementally built out its SOR ecosystem with Aladdin Office ERP as the platform to expand into B2B e-commerce, DTC e-commerce, plus brick-and-mortar retail management.

Economics of a Quiet Compounder

Fast forward to 2025, I’LL is operating a SOR platform with GAAP profitability and grew consistently at 10-15% over the last 20 years. Most notably, the company is now operating an exceptionally high ROIC with a clean balance sheet. Furthermore, as I’LL scales, margins continue to improve on multiple levels — both COGS and SG&A are stable or decreasing relative to sales. All of this indicates that I’LL’s model is sustainable at scale. For the Excel junkies, here’s some fast facts to illustrate how strong the fundamentals are:

• 20-40% ROIC: management doesn’t hoard cash either with a ~30% payout ratio 

• Gross margin up from 45% to 56% in the last 5 years 

• Operating margins up from 13% to 24% in the last 4 years 

• All of this achieved with nearly no debt with 7% debt to equity ratio 

It shouldn’t come as a surprise that an SOR business delivered these impressive results, especially when the founder (Iwamoto) still holds ~40% of the company to this day.

This time, I made sure to check the management compensation structure — Mr. Iwamoto was paid $400K in 2024 (not $14M!) — the coast is clear.

Giants Don’t Compete Here

Why don't—or can’t multinational ERP players just eat I’LL’s lunch? After all, larger providers such as Oracle NetSuite and Salesforce typically go after larger enterprises in all geographies. 

The reason is simple — it’s more cost-effective to convert large accounts with the same amount of effort (i.e., higher LTV to CAC ratio). I’LL, on the other hand, goes after the much more difficult segment of SMBs. These smaller companies typically have less standardized workflows, requiring more customization. Winning each contract is also less rewarding per dollar spent on sales & marketing. While the high-flyers rapidly captured the lucrative accounts, I’LL slowly chewed through the [more] difficult customers. 

The result? I’LL has thrived with limited competition. I’LL’s customization is also paying off. While offering more customization in the form of longer implementation time would compress margin in the short run, this also builds a powerful moat against smaller entrants that typically attempt to disrupt from the low-end with cheaper alternatives. 

Most of these cheaper, smaller SaaS competitors are also built much more rigid, designed with boilerplate features to maximize scalability. I’LL’s customer would need to suffer the painful switching-related disruptions AND a reduction in ERP quality. Switching is simply unthinkable for small-to-medium enterprises; this is the power of data lock-in. 

Even if the larger players enter I’LL’s market, they’d have a hard time dislocating I’LL because Aladdin Office has had more than a decade to entrench in many of these accounts. In other words, it’s simply not worth Oracle and Salesforce’s time to compete here. Notably, we see signs of I’LL’s moat in its financials too:

• Customer retention rate is 98% 

• Referral-led sales is 45% of new bookings. In short, this means customers almost never switch and they tell their friends about it!

Moving Up

Why does any of this matter? Given that I’LL’s formula has proven to work at scale, I think this company has potential to be a long-term compounder. It has an excellent track record, a powerful SOR-based moat, and a niche with weak competitors. 

I’LL also has plenty of room to grow given its status as a micro-cap company: 

In 2024, Aladdin Office generated ~$107M in revenue from 5,150 customers. There are approximately 3.4 Million Japanese SMBs in the same year. Even if only 10% of all Japanese SMBs would consider installing an ERP, this suggests that Aladdin Office would have only penetrated 1% of the market. Directionally, I’LL can go much higher.

Bottom Line

I’LL isn’t flashy — but it’s profitable, defensible, and still early in its market penetration. For patient investors, it might just be the rare snowball worth rolling.

Source:

• https://www.ill.co.jp/ir/material/Company_Information_Material_FY7_2023.pdf

• https://www.ill.co.jp/ir/fisco_report/20250512.pdf

• https://www.ill.co.jp/ir/financial_results/for_the_Nine_Months_Ended_April30_2025.pdf

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

"Invert, always invert" - Charlie Munger

Background

Medpace (MEDP) is a US-based contract research organization (CRO) focused on helping small, often single-asset biotech firms bring their drugs to market. These firms typically have no revenue and rely entirely on the success of a single clinical program. Medpace offers a vertically integrated, full-service model that takes products from pre-clinical through approval. This integrated structure allows for streamlined project execution and cost control — making Medpace especially attractive to biotech companies seeking capital efficiency and execution certainty.

The Financial Outperformance

Despite its smaller scale, Medpace has significantly outperformed its larger CRO peers. As of the latest data (LTM):

• EBIT margin: Medpace leads with 21.2%, compared to ICON’s 14.5% and Syneos Health’s 7.5%*.

• Leverage: Medpace carries just 25.3% debt-to-equity, lower than ICON’s 37.8% and far below Syneos’s 74.8%*.

• ROIC: Medpace reports a stellar 61.2%, handily beating ICON’s 8.6% and Syneos’s estimated 5.2%*.

This raises the question: Why is Medpace generating far better returns — while employing far less leverage — than peers nearly double its size?

*Based on Syneo Health's last known full-year financials of 2022

Focus

Medpace has built a culture centered on doing one thing extremely well: serving small biotech clients. The company doesn’t chase big pharma logos or diversify outside its core — all growth has been organic. Equally important is what Medpace chooses not to do: it is highly selective in the projects it takes on.

Scientific rigor is baked into its process. The Medical Directors Committee — made up of medical key opinion leaders (KOLs) — acts as the final gatekeeper on which programs the company accepts. Unlike the rest of the CRO industry, where landing large accounts is the only thing that matters, Medpace puts the science first. This brand of selectivity builds immense trust in a risk-averse industry where most clients only have one shot at success.

An added benefit: Medpace’s medical directors often become champions for their clients’ programs. As respected KOLs, their advocacy can help accelerate uptake and downstream commercial success.

Strategic Discipline

Medpace’s peers, notably ICON and Syneos, chased growth through M&A during the COVID-fueled CRO boom. They acquired smaller CROs in an effort to scale up — but many of these roll-ups destroyed value. When niche CROs focused on small biotech clients are absorbed into a larger platform, the cultural alignment erodes. These boutique teams are suddenly incentivized to prioritize large accounts like Pfizer or Amgen. The result? Service quality for smaller clients deteriorates. Many of these acquisitions ultimately turned into costly misfires.

In contrast, Medpace stuck to its knitting. No acquisitions. No flashy pivots. Just consistent, disciplined execution.

Culture: The Root of the Advantage

This all ties back to culture. Medpace remains founder-led — CEO August Troendle still owns 17% of the company. There’s no dual-class structure. The culture is lean, intensely focused, and customer-obsessed. Medpace often builds in-house tools rather than rely on vendors, simply because they believe they can do it better themselves.

This alignment is rare — and powerful. It creates a moat that isn’t easily copied.

Two Moats: Brand Equity and Process Power

Medpace’s culture manifests in two durable competitive advantages:

• Brand equity: Trust built through clinical discipline and selective execution. Clients know Medpace will treat their product like it’s their own.

• Process power: A well-honed internal process — from opportunity screening to execution — that keeps trial timelines tight and quality high.

These moats compound over time. Medpace isn’t winning because it’s the cheapest — it wins because it delivers.

The AI Wildcard

The most pressing question for CROs today: how will AI reshape clinical trials?

There are two dimensions to consider.

1. Market Dynamics: One risk is that AI enables pharma companies to bring trials in-house. But this seems unlikely at scale. Running trials is operationally complex and not core to pharma’s identity. Most would rather partner with a trusted CRO than build clinical ops themselves.

2. Nature of AI: The vast majority of AI tools — from recruitment software to trial simulations — are commoditized. They’re useful but interchangeable. In this world, it’s the operators — not the toolmakers — who win. Medpace is exactly the kind of operator that will benefit: disciplined, streamlined, and execution-focused.

There is, however, a second category: AI applications built on proprietary insight. These are harder to copy and can scale a company’s unique “secret sauce.” This is where Medpace could shine. It already has a deep and unique understanding of how to run trials for biotech clients. Layer AI onto that, and you could see powerful leverage.

Final Thoughts

Medpace is not your typical CRO. It chose NOT to grow through debt-driven roll-ups, NOT to participate in FOMO-driven acquisitions, NOT to take on lucrative but low-quality projects. The result? A company with best-in-class margins & ROIC and a brand that biotechs trust with their most important asset.

In an industry full of sprinters trying to grow at all costs, Medpace is a marathon runner — built to reach the finish line.

The core idea is this: you can't abstract away emergent behavior. If you want to truly understand the system, you have to rebuild its micro-foundations. This memo captures the shift in my perspective through recent modeling efforts.

ABM as a Better Fit for Complex Questions

I used to try breaking systems into high-level summary metrics: average market share, churn rates, adoption curves. But real-world systems don't obey linear approximations. Markets are shaped by bounded agents, asymmetries in information, lock-in effects, herd behavior, and policy inertia. ABM gives me a way to mimic all that.

I keep coming back to one insight: emergent outcomes can't be faked. You can only "get" path-dependence, self-reinforcing feedback loops, or tipping points if your agents act in a way that reproduces them.

Case Studies from My Own Projects

1. Modeling AI App Uptake in Healthcare

   • Health system buyers behave differently depending on peer behavior, internal ROI thresholds, and contract lock-ins. These aren't bugs; they're features of the system.

   • When I introduced heuristics like "go with the herd" or "copy the most profitable peer," I started seeing much more realistic diffusion patterns.

2. Simulating Competitive Dynamics Among CROs

   • Traditional firm modeling assumes rational optimization and continuous strategy space. In reality, CROs evolve strategies via imitation, inertia, and partial information.

   • ABM let me build in reactive strategies: copy pricing of competitors, expand capacity only after seeing three cycles of unmet demand, etc.

3. Vendor-Buyer Feedback in EMR Markets

   • Adding contract length as a lever wasn't just cosmetic. It created second-order effects—stability for incumbents, fragility for startups, and oscillations in price pressure.

Why This Approach Changed How I Think

Building an ABM forces me to ask: how does the behavior I see emerge from actual decision rules? This has rewired how I approach strategic questions. Instead of "What is the optimal price?" I ask "What kind of pricing rules survive competition and customer response over time?"

This means spending a lot of time, like A LOT, trying to understand how each agent thinks about their decisions—including figuring out their heuristics.

It also disciplines my assumptions. When a model outcome surprises me, I can't wave it away—I have to trace back the agent logic, and either fix the rule or embrace the implication.

Looking Ahead

I see ABMs not as predictive machines but as structured environments for stress-testing beliefs. Every new simulation is an invitation to ask: "What if the agents acted differently? What invisible pressures shape their behavior? What feedback loops are we missing?"

In that sense, ABMs are not just models—they are arguments. And like any good argument, they are only as strong as their assumptions, their logic, and their ability to make us see familiar dynamics in new ways.

Quick Take: Ningi's Claims vs. My View

• 🟢 PPV is overstated: Valid. The 43% PPV from PATHFINDER 1 reflects an enriched population. Real-world PPV in average-risk groups is likely closer to 5–10%.

• 🟡 Galleri does not detect 50+ cancers: Partially valid. Technically accurate, but most clinical value comes from the top 10–20 cancers where test performance is strongest. Unclear how this impacts Galleri's commercial value, if at all.

• 🔴 FDA approval is unlikely: Disagree. Based on the data and Grail's reported FDA feedback, approval odds remain high.

• 🔴 Lack of reimbursement: Misleading. Galleri isn’t FDA approved yet—no test is reimbursed pre-approval. This is standard, not a red flag.

• 🔴 Grail has cultural issues: Unverifiable and anecdotal. These claims don’t materially impact the regulatory or clinical outlook.

• 🟡 No mortality data = No clinical utility: Somewhat misguided. While this has been a point of contention voiced by some oncologists, it’s not the benchmark FDA, CMS, or NHS use to evaluate early cancer screening. Stage shift and cost-effectiveness are the operative endpoints.

PPV, Specificity, and the Real-World Picture

Ningi is right to push on Galleri’s positive predictive value (PPV). The widely quoted 43% PPV came from the PATHFINDER 1 study, which included many high-risk participants. That matters: in a general population, where cancer prevalence is lower, the test’s PPV likely drops to the high single digits (5-8%). It’s still strong—comparable to or better than fecal immunochemical test (FIT), mammography, and low-dose CT (LDCT) on a per-cancer basis—but not 10x better, and certainly not immune to false positives. I continue to view 43% PPV as a high watermark, but one that must be caveated. In practice, Galleri’s real-world utility will depend on how clinicians handle the 5–10% of patients who receive a false signal. This is where protocols and physician education matter.

What hasn’t changed is Galleri’s specificity. At ~99.5%, it has one of the lowest false positive rates of any screening tool, and that’s a legitimate strength. It means fewer follow-up scans, fewer unnecessary procedures, and less friction to adoption. Importantly, Grail's management has repeatedly emphasized that Galleri is a rule-in test (low false-positives), not rule-out (low false-negatives).

Sensitivity, Breadth, and What “50+ Cancers” Really Means

Galleri does not detect all cancers equally well. Sensitivity varies by cancer type and stage. It performs best in the deadliest cancers: pancreatic, liver, ovarian, head and neck, esophageal. It performs modestly in breast and lung, and poorly in prostate and thyroid. That’s not surprising—these more indolent tumors shed less DNA into the bloodstream, especially early on. The "50+ cancer" number is technically true, but clinically misleading. The real story is Galleri’s strength in detecting ~10–20 high-lethality cancers that currently have no other approved screening tools.

Should we discount the "50+ cancers" headline? Yes. Does that mean the product is dead in the water? Very unlikely. The value in Galleri isn’t that it’s perfectly comprehensive—it’s that it addresses the biggest unmet need in oncology screening. These are the cancers that kill people because we don’t find them until it’s too late. If Galleri can catch just a subset of those earlier, it moves the needle.

Regulatory Outlook: Prior Thesis Holds Up Here

I remain highly confident that Galleri will be approved under the FDA’s PMA pathway, barring any major public policy disruptions. Why? Because Galleri has shown a consistent analytical profile across studies (CCGA, PATHFINDER 1), and because the FDA’s reported feedback—as paraphrased by Grail management—has been straightforward: replicate your test performance in a larger, non-enriched cohort. That’s what PATHFINDER 2 and the NHS-Galleri trials aim to do.

The exact label is harder to predict. It may end up narrower than the "50+" banner, but even a label focused on the top 10–15 cancers would capture most of the clinical and economic value. These cancers are the ones driving mortality and cost, and where payers will likely be most receptive.

Aside from directional "guesses", I think it is a complete waste of time to speculate further on what the label will or won't include. There is no historical precedence in MCED.

Reimbursement: The Real Bottleneck

My thesis around reimbursement hasn’t fundamentally changed since Ningi’s report. This remains the most complex part of underwriting Grail’s long-term value. I agree with Ningi that it will be an uphill battle—not just to convince payers, but also physicians and guideline bodies—to adopt a new paradigm of cancer screening. Mass screening with an MCED followed by site-specific diagnostic workups is a major shift, and it will take time for the oncology community to adapt.

Reimbursement is critical to Galleri’s success. Based on published cost-effectiveness analyses, I believe Medicare and NHS will ultimately find the test acceptable around the $500/test range. But this is not a foregone conclusion. We’ll need to see how the Medicare and NHS outcome studies land. In the meantime, I continue to model a slower uptake curve, reflecting these structural frictions and the likely phased rollout via targeted coverage policies (e.g. age 65+, high-risk cohorts).

What has changed:

• I expect Galleri’s PPV in general population studies like PATHFINDER 2 to come in around 5–10%, not the 40%+ seen in enriched cohorts like PATHFINDER 1.

• There’s a real risk the FDA label includes only the top 10–12 cancers Galleri detects best. But the economics don’t hinge on detecting 50. Galleri’s value lies in population-level impact. Dropping low-sensitivity cancers may actually improve trust and cost-effectiveness by reducing false leads and unnecessary workups.

What has not changed:

• FDA approval is still the base case.

• Galleri’s value is strongest in high-mortality, unscreened cancers, and those account for the majority of cancer deaths.

• With PPV and specificity this strong, the signal-to-noise ratio is good enough for clinical adoption in a focused segment.

• Cost-effectiveness modeling still supports $500–800 pricing for seniors—even if real-world validation is ongoing.

A Word on Speculation

Some of the claims in Ningi’s report veer into the speculative. Allegations about company culture or internal morale are anecdotal at best and hard to independently verify. More to the point, they don’t impact the test’s clinical utility or regulatory odds. Similarly, the point that Galleri isn’t reimbursed yet is irrelevant—it’s not FDA approved, so of course (almost) no payer covers it yet. This isn’t a hidden red flag, it’s how the regulatory timeline works.

Bottomline

The Ningi short report helped sharpen my thinking. It drew clearer lines between narrative and evidence, and forced me to better distinguish what’s speculative versus what’s grounded in data. Some of their critiques were fair, particularly around adoption friction and over-reliance on enriched trial data. But others—like Galleri not being reimbursed yet—missed the mark entirely. I continue to see Grail as a compelling investment opportunity. The science is credible, and the probability of FDA approval is very high. What remains uncertain is the commercial arc: how quickly physicians adopt it, how CMS and NHS decide to pay for it, and how the broader oncology community embraces a new screening paradigm. This won’t be quick. It will require patient capital, steady execution, and a long view.

Note: I'll likely follow up with a part 2 to walk through the modeling and valuation to make this story more comprehensive.

Source:

• https://www.galleri.com/hcp/galleri-test-performance

• https://www.bmj.com/content/386/bmj.q1706/rapid-responses

• https://www.medrxiv.org/content/10.1101/2024.02.14.24302576v1

• Interviews with clinicians, former employees, competitors

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

This memo was initially written around May 21st, 2025

The Pitch

SBC Medical (ticker: SBC) is a high-margin franchising machine leading the Japanese aesthetic clinic market. It’s expanding into the US and Southeast Asia, backed by a founder who owns nearly the entire company. At ~$5/share (PE of ~10x), it’s still attractively priced. But there’s one big wrinkle: the CEO’s pay package is... jaw-dropping.

What’s SBC?

SBC Medical franchises aesthetic clinics across Japan. Think Botox and hair removal—packaged with brand support, admin services, and group-buying discounts. Since going public in 2022, it’s scaled to 250+ clinics, holding strong with 40% EBITDA margins and a 70% customer return rate.

But Japan’s aesthetic market is getting crowded, and regulators aren’t exactly handing out gold stars. Clinic bankruptcies are up, and competition in major cities is heating up. Still, SBC remains profitable and is pruning underperforming units.

Growth Game Plan

• Southeast Asia: Building a regional hub via its 2024 acquisition of AHH in Singapore.

• US: Entered Irvine, CA with plans to grow via MedSpa M&A.

• Japan: Doubling down on regional cities with comprehensive clinics, plus price hikes to offset cost pressures.

Moat Check

• Brand Loyalty: A 70% repeat rate is no joke. It implies trust in brand value and customer mindshare. I also checked SBC clinics' google review in Japan: they're incredibly high - nearly all exceeding 4 (out of 5). As a side note, Japanese reviewers often start their evaluations at 3 stars, viewing it as a neutral or satisfactory rating. Higher ratings are reserved for experiences that significantly exceed expectations.

• Scale Advantages: Bulk purchasing = cost edge.

• Operational Know-How: Years of running clinics at scale is hard to replicate overnight.

Given SBC's competitive advantages (and the track record to show for it), SBC remains profitable despite headwinds in Japan and is shedding non-core operations. Importantly, if SBC can endure the pain of a down cycle, it has plenty of opportunity to consolidate and emerge as a stronger player in Japan.

Management Scorecard

• Aikawa still owns ~90% of the company. That’s great for alignment—but not so great for stock liquidity.

• The financials look solid: high ROIC (>30%), low debt, growing earnings.

• Then there’s the comp issue. Again: $14M in 2024, $17M in 2023. Not stock. Not options. Cash. This isn’t just excessive—it’s a flashing red flag. If unchecked, it undermines the investment case entirely.

Less Beautiful Numbers

In 2024, CEO Yoshiyuki Aikawa took home $14 million in cash comp—about 3% of SBC’s entire market cap. That’s the kind of paycheck you expect from a Big Tech exec, not the boss of a micro cap business. To put it plainly: this is not just high—it’s egregious. It's more than most S&P 500 CEOs earn in cash comp. If this continues, it raises serious questions about governance.

It is difficult to speculate Aikawa's motives/philosophy around his compensation. While we know that the founder is highly aligned to how the business (and the stocks) performs, this compensation package is just too much. The worst-case scenario? Aikawa treats SBC like his personal piggy bank while minority shareholders are left holding the (empty) bag.

Capital Allocation & Risks

• SBC is reinvesting—expanding aggressively abroad. Generally speaking, investors should be happy to see reinvestment given SBC's high ROIC.

• But with growth comes risk:

  • Distraction: Managing Japan, the U.S., and SEA is a lot. We should monitor if the management loses focus moving forward.

  • Diworsification: New markets may not offer SBC's favorable unit economics. And we've seen SBC entering segments with poor outcomes. That said, management has shown discipline, cutting loose what doesn’t work (e.g., regenerative medicine).

Bottom Line

SBC is a bit of a mixed bag. On one hand, it’s a rare gem in the aesthetic space—40% EBITDA margins, strong brand loyalty, and undisputed market leadership in Japan. On the other hand, the CEO’s outsized cash comp is hard to ignore and raises real governance concerns.

Sources: IR presentation, Proxy Statement.

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

This memo was initially written around May 29th, 2025

Business Overview
Costamare Bulkers (CMDB) is a dry bulk shipping company formed via a spin-off from Costamare Inc. (CMRE). It owns 38 vessels and charters 48 more through its CBI (Costamare Bulkers Investments) platform, giving it a combined fleet of 86 ships totaling ~10.8 million DWT. CMDB generates revenue via a mix of spot and time charters, complemented by freight trading and hedging through the CBI platform. The company emphasizes operational flexibility and aims to capitalize on dry bulk market volatility.

Spin-Off Situation
The spin-off aimed to isolate Costamare’s dry bulk business (SpinCo) from its container operations (RemainCo), enabling more transparent performance evaluation. CMDB was recapitalized post-spin with ~$205M in liquidity (cash and funding from RemainCo) and ~$172M in debt, including $85M in forgiven related-party loans. The Konstantakopoulos family, which controls Costamare Inc., took a ~60% stake in CMDB—demonstrating insider alignment and confidence in the new entity.

For those of you who are familiar with special situations - you'll realize this is a spinoff + recap situation.

Current Valuation

• 2024 Revenue: $1.2 billion

• Estimated Book Value: ~$705M (Note that CMDB claims to have a NAV of ~$800M. I have made a few conservative adjustments)

• 2024 Operating Cash Flow: Negative $55.5M

• Market Cap (as of 5/29/2025): $227.76M

• P/S Ratio: 0.19

• P/B Ratio: 0.32

These standalone numbers mean little without context. Compared to peers in dry bulk shipping:

CMDB trades at a steep discount to peers. The discount likely stems from CMDB's unprofitability and uncertain recovery timeline. Plus, CMDB surely suffers from the selling dynamics triggered by spinoff.

Valuation Scenarios Based on Peer Multiples

• Downside (DSX levels):

  • P/S = 0.72 → $860.4M

  • P/B = 0.34 → $239.7M

• Mid-Range (SHIP/SB/GNK average):

  • P/S = 1.12 → $1.338B

  • P/B = 0.49 → $345.5M

• Upside (GOGL/SBLK average):

  • P/S = 1.67 → $1.995B

  • P/B = 0.85 → $599.3M

To better understand what this distribution means in aggregate, let’s run a Monte Carlo simulation (10,000 simulations) using peer valuation distributions (P/S mean = 1.12, P/B mean = 0.49). Results are as follows:

• Mean: $964M →323% Upside (from $227.76M)

• Median: $964M →323% Upside

• 5th Percentile: $596M →161% Upside

• 95th Percentile: $1.3B →486% Upside

To better reflect CMDB's lack of profitability, an adjusted simulation was run using capped distributions (P/S ≤ 1.5, P/B ≤ 0.6) and a 30% weighting for low-multiple scenarios:

• Mean: $721M →216% Upside (from $227.76M)

• Median: $818M →259% Upside

• 5th Percentile: $185M →19% Downside

• 95th Percentile: $1.2B →427% Upside

The result is a bimodal distribution: one peak (left) representing CMDB remaining distressed, and another (right) assuming operational improvements. Even in poor outcomes, downside is limited—heads I win, tails I don’t lose much.

Investment Thesis
CMDB offers asymmetric upside due to:

1. Spin-off Dynamics: As Joel Greenblatt noted, spin-offs often result in mispricing. CMDB share price has declined since the spin, likely due to forced selling and limited coverage.

2. Conservative Balance Sheet: $160M in cash vs. $172M in debt post-spin.

3. Cyclicality Tailwind: Dry bulk is in a downcycle—any rebound in rates could sharply improve cash flow and sentiment.

The core assumption is CMDB must return to cash flow breakeven. Its clean balance sheet, refocused mandate, and scale give it a fair chance.

Risks

• Management Quality: The Costamare family has a mixed track record; CMRE stock has stagnated over the past decade. Their control of SpinCo is a concern.

• Cash Burn: Persistent operating losses could erode equity value if market conditions remain weak.

• Macro Headwinds: Recession or trade slowdowns could weigh on shipping demand. However, CMDB’s focus on raw materials mitigates exposure to consumer trade shocks.

Conclusion
CMDB is a classic cigar-butt, spin-off special situation. Despite valid execution risks, its deeply discounted valuation, strong insider alignment, and potential operational catalysts create a compelling opportunity for contrarian investors. Downside appears contained; upside is substantial if execution improves.

Sources: CMDB IR Presentation, SEC Filings.

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

Info

This memo was initially written around January 29th, 2025

Kaspi (KSPI) is a conglomerate based in Kazakhstan that operates a super app called “Kaspi.kz,” which includes banking, payments, e-commerce, and government services. Its model is similar to WeChat and Line, minus the social media and instant messaging elements.

Kaspi’s net income is equally divided between its three main segments: fintech, payments, and marketplace.

• Payments: Kaspi QR, household bills, P2P

• Fintech: buy now pay later, Kaspi deposit, loans

• Marketplace: e-commerce, m-commerce (offline payment via QR), Kaspi travel, e-grocery

Kaspi has seen remarkable growth since its founding 15 years ago. By 2024, the Kaspi super app boasts 14 million monthly active users, capturing ~70% of Kazakhstan's total population (20 million). Mirroring its user growth, Kaspi’s net income has surged at a CAGR of over 70%, from $26.6 million in 2016 to $2.2 billion in 2023. Although the growth rate has decelerated in recent years, Kaspi still reported a 44% increase in net income from 2022 to 2023.

Kaspi presents two major competitive moats – brand power and network effect.

• Brand power: Kaspi is the most well-known financial services and e-commerce company in KZ. With a 70% penetration rate, it's no surprise that many Kazakhs rely on Kaspi for their daily needs, such as groceries, online shopping, and payments.

• Network effect: For many super-apps, network effect is an essential growth driver. WeChat, for example, leveraged its instant messaging platform to achieve this. Kaspi benefits similarly through its P2P and merchant payment platform. Each new user enhances the experience for both current and future users. Moreover, with widespread adoption, the cost of acquiring customers for new services drops significantly. However, it's important to note that network effects can also create a negative feedback loop if the company fails to maintain high-quality service and users start to abandon the super-app.

The company’s management team has demonstrated good track record and is highly incentivized to produce adequate long-term shareholder returns.

• Extremely high insider ownership: the two co-founders of Kaspi own about 40% of the outstanding shares, without a dual class share structure.

• Stellar track record: Kaspi has consistently generated high and increasing returns on invested capital (ROIC) since 2016, ranging from 40% to 80%. In 2023, ROIC exceeded 100%. Impressively, this growth was achieved despite decreasing debt and maintaining a net cash position for the past five years.

• Rational capital allocation: historically, the management team has not hesitated to return excess capital to shareholder in the form of dividends.

Here are some risk factors that can negatively impact investment results:

• Growth stagnation: Kaspi may face challenges if it exhausts expansion opportunities within KZ and fails to enter other markets, despite attempts to break into Turkey. So far, Kaspi has successfully diversified within KZ, generating new income streams through e-grocery and travel services.

• Economic downturn: Kazakhstan's economy can face significant disruptions due to nearby conflicts (Russia, China, or other central Asian countries) or fluctuations in raw material prices, such as petroleum and uranium, given its heavy reliance on exports.

• Government intervention: Kazakhstan's weak rule of law could lead to government actions that negatively impact Kaspi. However, Kaspi currently enjoys a positive and strong relationship with the government, which has showcased Kaspi as a success story of its digitization and fintech-friendly policies.

Given Kaspi’s current market valuation (enterprise value) of $18 billion, I estimate the long-term expected ROI to be around 12% to 21% per year.

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.

This memo was initially written around January 31st, 2025

A Dino store has everything you need, nothing more.

Dino Polska (Dino) is a discount retailer chain based in Poland. Dino offers basic staples, like bread, vegetables, fruits, and fresh meat alongside non-food everyday items like detergent. As a discount retailer, Dino aims to be the most affordable and convenient grocery market in its vicinity.

Dino achieves affordability by adopting a pure-play strategy focused on rural and suburban locations, avoiding competitive urban markets. This approach targets a homogenous consumer base, simplifying inventory and supply chain with an SKU count of around 5,000. Additionally, Dino builds and owns every store to save on rent long-term and keeps each store layout identical, maximizing repeatability, accelerating operational learning curves, and lowering building costs. Their vertically-integrated meat processing plants enable them to perfectly coordinate a supply of affordable fresh meat with minimal spoilage rates.

For convenience, Dino employs a small store format of 400 sq meters, designed to serve about 2,500 residents. This decentralized network of stores allows them to be located within walking or cycling distance to customers while remaining economically viable even in remote towns with only a few thousand residents.

Fresh meat counters are valued for their perceived quality in Polish cuisine, which emphasizes chicken and pork. Most Polish discount groceries lack fresh meat (and offer packaged meats) due to coordination challenges in preventing spoilage while keeping costs low.

The combination of small store format and offering products at a good value keeps Dino highly competitive against its discounter peers, as shown in the comparison below:

Dino’s track record speaks for itself. Since its founding in 1999, it has grown into a $7 billion revenue business in 2024 with more than two thousand stores across western and central Poland. Dino’s net income has also grown at a CAGR of 35% in the last 10 years with an ROIC above 20% since 2018.

Dino has one clear competitive moat and a less tangible one.

• Counter position: Dino’s strategy of focusing on rural areas with low-density populations is a core aspect of its small store format model. In contrast, Dino’s main competitors like Biedronka and Lidl struggle to expand into rural areas due to differing customer preferences and spending power. Urban and rural customers have distinct tastes, necessitating either a reduction in product variety or increased inventory in distribution centers to meet diverse demands. By avoiding urban markets, Dino perfects its unit economics in rural areas, whereas competitors can only achieve this balance by risking their overall operating margin. In short, major retailers with existing city presence have limited staying power in villages.

• Brand power: Dino's brand power is evident through social media discussions among Polish residents, who often regard Dino as a quality retailer despite its low costs. For instance, Dino's fresh meat counter is highly praised, as Poles consume a lot of meat and associate a traditional meat counter with quality. In contrast, Biedronka, which only offers packaged meat, struggles to maintain this perception of freshness.

  • Another strength of Dino’s brand is its straightforward approach to customer service. During recent price wars among discount retailers, competitors offered complex daily bundles that, while seemingly cheap, were purposefully designed to be confusing to customers. Some Poles perceived these marketing strategies as deceptive. However, Dino's daily offers are clear and simple, such as "buy X bundle for Y price today," which fosters greater trust and satisfaction among customers.

The company’s management team has demonstrated good track record and is highly incentivized to produce adequate long-term shareholder returns.

• Extremely high insider ownership: the founder, Tomasz Biernacki, owns about 51% of the outstanding shares, without a dual class share structure.

• Stellar track record: Dino has consistently generated good and increasing returns on invested capital (ROIC) since 2014, ranging from 16% to 25%. Since 2018, ROIC exceeded 20%. For a capital-heavy, real-estate heavy business model, achieving 20%+ ROIC is quite impressive and would be impossible without a tight control on costs and disciplined capital allocation.

• Rational capital allocation: historically, the management team has not paid dividend or repurchased shares. Dino has focused on investing in new stores and distribution centers to capture new markets. Given its ability to maintain high ROIC, it’s clear that these reinvestments have good incremental returns.

Here are some risk factors that can negatively impact investment results:

• Real estate costs: Dino’s expansion model heavily relies on acquiring land. If real estate prices were to appreciate significantly, this can reduce new store’s ROI.

• Intensified competition: If Lidl, Biedronka, or other competitor successfully replicates Dino’s small store / rural model, Dino can face much greater headwind.

• Growth stagnation: Although Dino still has significant room for growth in Poland (mainly eastern regions), the remaining whitespace is not infinite. Sooner or later, international expansion would be needed to maintain its current trajectory.

• War: Poland can face significant disruptions due to nearby conflicts (Ukraine, Russia) or an outright invasion by Russia. The implications would be disastrous.

Given Dino’s current market valuation (enterprise value) of $11 billion, I estimate the long-term expected return on investment to be around 6% to 11% per year. Despite Dino’s excellent business qualities and growth prospects, current valuation (price to free cashflow of >100x) assumes Dino to achieve very high growth rate in the next decade. High expectations is a source of downside risk if valuation turns pessimistic.

Disclaimer:
The information provided in this content is for informational and educational purposes only and should not be construed as financial or investment advice. The opinions expressed are those of the author and do not constitute a recommendation to buy or sell any securities or financial instruments. While efforts are made to ensure accuracy, the information may become outdated or incomplete over time. Investing involves risk, including the potential loss of principal. Always conduct your own research or consult with a licensed financial advisor before making any investment decisions. The author may hold positions in the securities discussed.