Competitive Moat Analysis: How to Test Whether Your Advantage Is Real

Why most moat analysis proves nothing

A company earning good returns always has an explanation for them. Turning that explanation into a claim about the future requires a different document, and the usual taxonomy does not supply it. Calling an advantage a network effect or a switching cost says how protection is produced. It does not say whether the protection is still being produced, or for how much longer.

The question a board actually needs answered is narrower and testable: what would it cost a challenger to take one of your customers, and is the customer worth that much to them? If the answer is yes, the moat has already been crossed, whatever the taxonomy says.

The four sources of a gap, and what erodes each

Every durable gap comes from one of four sources. Each has its own way of eroding, and each shows the erosion somewhere specific before it reaches the income statement.

Naming the source matters because the defense differs. A switching-friction moat is defended by making migration expensive; a scale moat by staying ahead of what suppliers will rent; a demand-side moat by making multi-homing costly; an access moat by holding the channel or the rule. A company that defends the wrong one spends money and keeps the exposure.

Switching friction: watch for the migration path

Nvidia's protection is less the chip than what has been written for it: over 5.9 million developers used CUDA and Nvidia's other software tools as of its fiscal 2025 filing, and in AI training its share exceeds 90%.

The erosion arrives the way the table predicts. AMD funded ZLUDA, a compatibility layer that runs unmodified CUDA binaries on AMD GPUs, and the project was open-sourced in February 2024. A migration path built by someone else is the first sign that a switching-friction moat is being priced, even while the share still looks safe.

Unity shows the opposite mistake: charging as if the friction were higher than it was. In September 2023 it announced a per-install runtime fee aimed at the successful games it most needed to keep. It canceled the fee entirely a year later and returned to seat pricing. By 2024 Unreal held 31% of Steam game sales by revenue against Unity's 26%, passing it for the first time since 2018.

Scale economics: watch for what suppliers will rent

Coca-Cola's wall is not the formula, which it kept a trade secret from 1891 rather than patent. It is a network of 225 bottlers and about $5 billion a year of marketing upkeep, assembled over a century.

Scale has a ceiling, and it is usually a customer. PepsiCo's Frito-Lay runs nearly 15,000 routes, the largest direct-store-delivery system in North America, built because owning the route controls placement and freshness. In December 2025 a federal judge unsealed the FTC's complaint alleging PepsiCo gave Walmart preferential pricing not offered to smaller retailers. The scale moat held against rivals and bent in front of the largest buyer.

Demand-side returns: watch for multi-homing

Visa's advantage traces to a governance choice in 1970: reincorporating the card program as a company owned by the issuing banks, which made rival banks co-owners rather than competitors. In fiscal 2024 Visa processed 234 billion transactions on its own network.

Airbnb is the case to watch. With 28% of its listings cross-listed elsewhere, its hosts already multi-home. From October 2025 it moved hosts connected through property-management software to a mandatory 15.5% host-only fee, converging with Booking.com while Vrbo stayed cheaper for hosts. That is a bet that the demand advantage is strong enough to charge for while hosts can list on both, and no dated outcome yet says whether it holds.

Access: watch the channel and the rule

Palantir's position rests partly on a court ruling. In 2016 the Court of Federal Claims found the Army had ignored a 1994 procurement law by never checking whether a commercial product existed, and in 2018 an appeals court agreed the Army had acted arbitrarily and capriciously.

American Express defended its access the same way: when Visa and Mastercard settled the government's challenge to anti-steering rules, Amex litigated, and in 2018 the Supreme Court upheld the rules 5-4. Those rules are what let it hold its higher merchant price.

An access moat erodes when the middle changes hands. Roku's value was its neutrality between Netflix, Disney+ and every other service. In June 2026 Fox agreed to acquire it at about $22 billion, putting a content owner in control of the platform its rivals depend on.

An access rule can also bind the holder. Diageo sits behind a UK-built wall around 40% of all Scotch whisky, and in 2018 the Scotch Whisky Association declined Diageo's own request to finish single malts in tequila barrels. The rule that keeps rivals out keeps the incumbent in.

Compute the gap in the challenger's spreadsheet

Once the source is named, put a number on it, and do the arithmetic from the challenger's side. The gap is what it costs a challenger to win one of your customers, less what that customer is then worth to them.

The three answers usually differ by years. In the worked case used in the Moat Anatomy Strategy Pack, the gap holds for all ten years against a startup, for five against the neighbor, and for one against an entrant earning its return elsewhere. Nothing about the company changes between those three readings; only whose spreadsheet the arithmetic is done in.

An incumbent's own behavior is often the clearest reading. Adobe agreed to buy Figma for $20 billion and, with no clear path to approval from the European Commission and the UK CMA, paid a $1 billion fee in December 2023 to walk away.

What these cases show

Each case is drawn from the Stratrix case library and checked against the filings and reporting listed under Sources. The last column says what each one does not establish.

Running the analysis

What this method cannot tell you

Moats are identified by their returns, so any collection of them is selected on the outcome it is being used to explain. The businesses with the same mechanism and no returns are invisible. The cases above support recognizing a mechanism. They cannot support a survival rate, and this page does not claim one.

The method also does not say how wide a moat should be. It replaces that question with a narrower one that has an answer: whether, and until when, the arithmetic still favors you in a challenger's spreadsheet.

Questions