Boris Kulakhmetov is Head of the AI Incubator at AI Digital, building tools that measure how AI models represent brands.
getty
I’ve spent most of my career trying to answer one question: What actually makes someone buy? I’ve been wrong about the answer more than once, and I believed each wrong answer completely.
Sales began as a craft. The process was whatever your best closer said it was. Then came the funnel—dashboards, stages, conversion rates. Real progress was made, but it bred an obsession that has wasted more marketing budget than any bad campaign I’ve ever seen: attribution.
Which touchpoint gets the credit? The first answer was last-click. Whatever the buyer touched right before purchase took all of the credit. It was comforting, but it was also nonsense. Nobody watches one ad and buys on impulse.
I learned that the expensive way. I advised a healthcare brand whose buyer journey went like this: A parent sees an ad, consults a clinician and, only then (sometimes months later), arrives at a specialized clinic. Three marketing systems (digital, clinician outreach and below-the-line) fed that journey. Each was measured in its own silo, with no attempt to attribute them together. Last-click handed all of the credit to whatever the parent tapped last. We were measuring the doorknob, not the house.
Last-click never measured what caused the sale. It measured what happened to be standing closest to the finish line.
Then, the industry got sophisticated with black-box attribution—multitouch models from the big platforms that promised to weigh every touchpoint with algorithmic precision. I trusted them. My clients trusted them completely. We moved real money on what they told us.
In 2021, I contributed to a McKinsey article urging marketers to deactivate their digital marketing autopilot and take back control of their own numbers. I underestimated how literally the industry would need that advice. It took me too long to see that every walled garden grades its own homework.
During the pandemic, I led full-funnel marketing work for a major European airline at a moment when every euro of spend was contested. Each platform’s dashboard told the same flattering story: Its own inventory was essential, and every rival channel was waste. We stopped arguing with black boxes and rebuilt around numbers we actually owned, starting with a checkout flow that bled conversion at every step, on the order of 10 million euros moved across channels, tactics and keywords. The verdicts came from our data, not theirs.
When the same company sells you the ad space and scores which ad space worked, you already know how the story ends. That’s not a conspiracy. It’s an incentive. You can’t audit a model you’re not allowed to open.
I don’t say that to bash any one platform. I say it because we built an entire discipline, with budgets, teams and careers attached, on numbers produced by the companies being graded. Now, the whole thing is being deleted underneath us.
Here’s the shift most leaders haven’t priced in: AI is absorbing the entire discovery funnel. According to Adobe’s 2026 global consumer research report, 25% of surveyed customers named AI platforms such as ChatGPT as their top research tool, ahead of brand websites and online reviews.
When a buyer asks an assistant to compare options, the model becomes aware of your brand, weighs it against alternatives and narrows the choice before a human sees a single result. Awareness, consideration and decision are all inside the machine. There’s no click to attribute. Often, there’s no visit at all.
A July 2025 Pew Research Center report tracking real browsing behavior found that when an AI summary appeared in search results, users clicked a traditional link just 8% of the time, roughly half the rate as without a summary. Links inside the summaries themselves got clicked 1% of the time.
Sit with what that does to every attribution debate we’ve ever had. You can’t split credit across touchpoints when the touchpoints have collapsed into one opaque exchange between a person and a model. The funnel didn’t get more measurable. It moved somewhere you can’t put a pixel.
I’ve stopped asking which channel gets the credit. That was always the wrong question in some sense; AI just made that undeniable. The right question is simpler and more uncomfortable: When the AI does the choosing, does it surface you at all?
That’s measurable, not with attribution but with how often, and how favorably, the models represent your brand when they answer a buyer’s question. It’s early and unstable. Ask the same question twice, and you’ll get a different set of brands. Most companies aren’t even looking, and frankly, that’s the best news in this article. The scoreboard that will decide the next decade of marketing is sitting in public view, and your competitors haven’t found it yet.
I spent 20 years being wrong about attribution in increasingly sophisticated ways. This time, I’m early. I’d recommend it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?