The problem with measuring the moat

There is a seductive simplicity to the moat as a concept. You picture the fortress, you estimate the water’s depth, you tick the box. Brand: present. Switching costs: present. Network effects: present. The checklist tells you the business is defensible, and you move on.

The problem is not that moats are fiction. It is that measuring a moat at a point in time tells you almost nothing about whether it compounds or collapses over the years that follow. A static reading on a dynamic system is, at best, incomplete. At worst, it is actively misleading.

This is not an abstract concern. It is the central failure mode we have observed in otherwise rigorous investors — and the one we have worked hardest to design around in how we evaluate businesses.

Static versus dynamic

The language of moats is a language of position. It describes where a business stands today relative to competitors. What it does not describe is whether the ground beneath that position is rising or sinking, or whether the competitive game itself is changing in ways that render today’s position irrelevant.

A business can have every traditional moat characteristic intact — strong brand, high switching costs, dominant market share — while its underlying substrate is quietly eroding. This is what we call the Kodak camouflage pattern: fundamentals that look fine on current metrics, masking a structural deterioration that only registers in the numbers late. By the time it becomes visible in earnings, the thesis is already broken.

The implication is that any evaluation framework worth using must account not just for the strength of a competitive position, but for the health of the system in which that position operates. Is the underlying market growing or shrinking? Is the business model gaining surface area or defending a contracting base? These are questions about trajectory, not snapshot — and they require a different set of tools than the standard quality checklist.

Two moats, not one

One distinction we have found consistently useful is between what we think of as defensive and propulsive moats — and specifically, whether both are present simultaneously.

A defensive moat is what most quality frameworks measure: the barriers that prevent customers from leaving. Switching costs, data lock-in, brand trust, regulatory position. These slow erosion.

A propulsive moat is something different: the mechanism by which the business actively expands its surface area over time. The feature set that gets better with more users. The data advantage that compounds with each transaction. The distribution that makes the next product launch easier than the last.

The distinction matters because a business can have one without the other. A company with strong defensive characteristics but no propulsive engine is a slowly shrinking asset — durable but decelerating. What we are looking for are businesses where the two engines are genuinely separate, because independence creates robustness. When a single capability is asked to do both jobs simultaneously — defend the existing position and drive expansion — you get the Intel failure mode: a process node that was both moat and growth engine simultaneously failed at both when it fell behind.

The more dangerous case is where previously independent engines are being deliberately coupled by a new layer. Microsoft wiring Copilot to ride M365 distribution while consuming Azure capacity is a current example: powerful if it works, structurally more fragile than the engines were when they operated independently.

The disruption question

Among the filters that have proven most discriminating in practice, one stands out for its simplicity and its predictive sharpness: does the disrupting technology route through this business, or around it?

Phrased differently: is the company a toll road that the new technology must drive on, or a road that the new technology builds a bypass around?

The distinction is not always visible in advance. Two software businesses can look nearly identical from a quality metrics perspective — recurring revenue, high gross margins, embedded enterprise relationships — and yet sit on completely opposite sides of this question. We saw this clearly in contrasting ServiceNow and Adobe over the period when AI-native tools began to reshape the creative and enterprise software markets.

ServiceNow is where companies put AI to work. Its position is as a workflow layer — the connective tissue between enterprise systems. As AI adoption accelerated, that workflow layer became more valuable, not less. The technology routed through it.

Adobe’s position, though historically strong, faced a more uncomfortable question. If AI-native tools allow users to create professional-quality content without professional software, the bypass risk is real. The same divergence signal — beaten-down price against apparently intact fundamentals — reads very differently through this lens than through a pure value lens. In the Adobe case, the substrate question overpowered the bargain signal.

This is the hierarchy principle that matters most: the substrate and disruption checks must rank above the divergence filter. "Cheap on fundamentals, narrative beaten down" is a buy signal in many situations. It is not a buy signal when the underlying game is changing in ways that have not yet reached the income statement.

The correlated bet problem

Quality investors tend to focus on individual positions. The portfolio-level question — are multiple holdings actually the same underlying bet held several ways? — receives less attention than it deserves.

A compounder sleeve can look diversified by company while being highly concentrated by bet. Owning Microsoft, Alphabet, Meta, and Nvidia in the same portfolio is not four separate positions on four separate theses. It is, in meaningful part, a single position on AI capital expenditure eventually generating returns — held four ways. The four positions will be right or wrong together on that specific question.

We are not arguing against owning businesses with overlapping tailwinds. We are arguing for naming the shared bet explicitly, sizing it deliberately, and recognizing what the compounder sleeve actually needs from the convexity sleeve to remain balanced at the portfolio level.

On calibrated confidence

The output of a rigorous evaluation process should not be a binary verdict. Pass/fail, buy/don’t buy — these formats are cleaner to communicate and easier to act on, but they systematically misrepresent what the analysis actually produces.

What the analysis produces, when done honestly, is something more like a probability distribution. Some names present as high-confidence holds: multiple independent filters agree, and the disagreements are minor. Others present as high-confidence passes: the filters converge on a clear problem. These cases are tractable.

The genuinely hard cases — and the most important ones to handle correctly — are the low-confidence splits: situations where strong filters contradict each other. Adobe, at the time we were analyzing it, was precisely this. Filters looking at narrative-versus-fundamentals divergence argued it was cheap and attractive. Filters looking at substrate integrity and disruption vector argued it was a value trap in formation. Those filters were not both right, but we could not resolve which was right from the quantitative data alone.

The right answer to a genuine split is not to pick a side and round up to conviction. It is to name the split explicitly, hold it in that ambiguous state, and wait for the adjudicating information to arrive. Position sizes should reflect calibration, not projected confidence.

On graduation

One final distinction worth naming: not every exceptional holding begins its life as a compounder. Some businesses start as cyclical opportunities — early in a structural upgrade that is not yet legible in the quality metrics. Micron is a recent test case: a business with cyclical characteristics that, in our view, is structurally upgrading toward compounder-like behavior as memory becomes increasingly central to AI infrastructure.

The question of whether a cyclical position has earned the right to graduate into a core holding is a different question from the one the standard quality battery asks. The compounder battery is, by design, structurally blind to cyclical inflections — it filters them out as noise. The work of identifying when a cyclical company has crossed a structural threshold requires a different lens, and a deliberate handoff between the two modes of evaluation.

We think this graduation logic matters for a particular reason: the best long-duration compounders are not always obvious at the moment they begin compounding. Some are legible early. Others are not. A framework that only recognizes quality after it has already been priced in is a framework that systematically misses the best entry points.

What we are actually doing

The argument running through all of this is not that the moat concept is wrong. It is that a static moat reading on a dynamic system is not enough. The evaluation frameworks that hold up over long periods are the ones that ask not just whether a business has a competitive position today, but whether the system it operates in is growing or shrinking, whether the new technology routes through it or around it, whether the moat is genuinely compounding or gradually hollowing out.

The goal is businesses that do not merely defend what they have, but structurally increase their surface area with each passing year. That is a dynamic question. It requires a dynamic answer.

Related reading