The AI Revenue Measurement Gap
Two Revenue at Risk exercises, run this year in banking and travel, turned up the same structural blind spot. In each case, a large company already had direct evidence that AI systems make concrete, decision-stage competitive choices in one part of its business. In each case, an adjacent, comparably sized part of the same business had never been measured at all. Neither company had asked the second question, because nothing had forced them to.
The Banking Case
The first case involved a global money-center bank. AIVO's existing reputation work with the bank, run through a consumer-facing AI RQ instrument, had already shown that AI systems do not simply cite the bank when asked about retail banking options. They make a final recommendation, and in testing that recommendation defaulted to two named national competitors ahead of the bank at the actual decision turn. The mechanism was real and it was measured, but only for consumer banking. The bank's middle market and small business banking lines, a commercial segment worth several billion dollars in annual revenue, had no equivalent evidence either way. Generative AI adoption among business owners is now widespread, with multiple 2026 surveys reporting usage in the 75 to 87 percent range across marketing, sales, and operations. No published survey yet isolates how many of those businesses are using AI specifically to choose a banking provider. That gap sat directly beneath a live, proven mechanism next door.
The Travel Case
The second case involved a global accommodation marketplace. Here the exposure was not an awareness problem. The brand is among the most recognized in its category, and the natural objection, that a marketplace with an enormous inventory footprint should be structurally hard to displace, is intuitive but not obviously true. Inventory breadth is not the same as recommendation preference. AI assistants increasingly act as choice compression systems, reducing large option sets into a single ranked recommendation, and whether inventory scale actually translates into recommendation preference, or whether the model instead favors institutional trust signals like loyalty programs, expense policy integration, and cancellation guarantees, remains an open empirical question rather than a settled one. The company's broader consumer business is well studied and well understood. Its business travel line, a segment plausibly worth several hundred million dollars in annual revenue by the company's own last disclosed program share, several years old and never updated, had never been tested against the specific mechanics of AI mediated recommendation at all.
Where Measurement Stops
Laid side by side, the pattern is not really about banking or travel. It is about where measurement stops. Both companies had strong evidence of AI mediated competitive displacement somewhere in their business, evidence detailed enough to include which specific competitors won and at which point in a multi turn conversation the losing brand was displaced. Both companies had, immediately adjacent to that evidence, a comparably large revenue line where the same dynamic had simply never been tested. Neither absence was a company specific failure. It reflects a general property of how this kind of measurement work has been sequenced across every industry AIVO has looked at so far: proof of mechanism arrives first, usually in whichever category is easiest to instrument, and the adjacent categories wait.
The Funnel AI Broke
Part of why the adjacent categories wait is that most companies are still measuring the funnel that existed before AI entered the decision. That funnel ran from brand awareness through search visibility, consideration, and conversion, and it is still the funnel most measurement budgets are built around. AI inserts a new stage into that sequence rather than replacing it. A user states an intent, the AI system evaluates a candidate set against that intent, the system recommends one option over the others, and only then does a transaction happen or not. The measurement gap identified in both cases above is not a gap in brand awareness or search performance. It is a gap in visibility into that middle stage, the evaluation and recommendation step that now sits between a customer's intent and their purchase, and that stage does not show up in any of the instruments built for the funnel that came before it.
Two Metrics, Not One
The working framework AIVO uses to size this gap, published under the Revenue at Risk methodology, breaks the question into two distinct metrics rather than one blended figure. Observed AI Displaced Revenue equals Category relevant Annual Revenue multiplied by the AI mediated Purchase Influence Rate multiplied by the Measured AI Displacement Rate. That number is observational. It describes how much revenue sits inside decision journeys where AI assistance is meaningfully present and where the brand in question is not the final recommendation. It does not assume that revenue would otherwise have gone to the brand being studied. A second figure, AI Attributable Revenue at Risk, multiplies that observed figure by an attribution factor grounded in study design rather than a direct causal estimate. The first metric is observational. The second is inferential. The methodology insists on keeping them separate, because collapsing them into a single number is how sizing exercises quietly turn into loss estimates they were never designed to be.
What the Numbers Do Not Yet Say
In both the banking case and the travel case, the actual dollar figures published so far are illustrative rather than measured. A five to twenty percent range for the AI mediated Purchase Influence Rate produces a wide spread of possible exposure in each case, precisely because no study has yet narrowed that range with real data. That is the point of publishing the framework ahead of the numbers. The purpose of this kind of paper is not to tell a company how much revenue it is losing. It is to specify exactly what would need to be measured before that number could withstand analytical scrutiny, and to note, plainly, that the adjacent segment sitting next to a proven mechanism is usually the first place worth looking.
What Measurement Requires
Both studies point toward the same next step: multi turn probing across the categories in question, run both with the brand named directly and with it withheld so the model has to surface its own candidate set, tracking the specific turn at which a brand first loses its position and whether it holds that position through to the point where the conversation would actually hand off to a purchase decision. Visibility tells a company whether a brand appears. Multi turn measurement tells it whether the brand survives.
AIVO Meridian (aivomeridian.com)
Comments ()