The Last Click Lie

There’s a Phoebe Gates AI shopping app called Phia that’s been accused of cookie stuffing — claiming affiliate commissions for sales it didn’t actually drive. Bloomberg reported it. LinkedIn is dissecting it. The mechanics are interesting, but they’re not the real story.

The real story was told by Christine Barnett, who pointed out something more uncomfortable than fraud: people don’t interrogate numbers that come from a dashboard. A number on a screen feels objective. The interface confers authority. The methodology behind it — whether it’s sound, incomplete, or actively manipulated — barely gets questioned.

This is the attribution problem. And it’s everywhere.

The Model Masquerading as Truth

Attribution is a model. It’s a set of assumptions about which touchpoints deserve credit for a conversion. Last-click attribution, the default in most marketing organisations, gives all the credit to the final interaction before someone buys. It’s clean. It’s simple. And it’s wrong.

Not wrong in the way that fraud is wrong. Wrong in the way that a map with half the roads missing is wrong. It shows you something real. It just doesn’t show you enough to navigate by.

The problem starts with what last-click actually measures. A customer clicks an ad and converts. That channel gets the credit. Budget shifts accordingly. The dashboard confirms the decision. Everyone feels data-driven. The quarterly review looks good.

But that channel is often doing two completely different jobs in the same journey. It’s frequently the first touchpoint: someone searches a generic term, clicks an ad, browses a site, leaves without converting. That’s a top-of-funnel interaction that started the entire journey. Last-click gives it zero credit for that.

Weeks or months later, after a dozen other influences have done their work, the same person searches a brand name, clicks another ad, and converts. Now that channel gets all the credit. For a conversion it both started and finished, but didn’t carry through the middle.

The model can’t distinguish between these two interactions. It treats a discovery ad and a brand search ad as the same thing. It’s measuring the container, not the contents. And every budget decision made on that basis is built on a foundation that’s incomplete at best and misleading at worst.

The Funnel You Can’t See

The deeper problem isn’t that last-click ignores other channels. It’s that it creates a structural bias toward the bottom of the funnel and against the top.

The channels that fill the top of the funnel — content, community, brand, organic social, word of mouth, PR, and increasingly, AI-mediated discovery — are the hardest to attribute. They don’t produce clean clickstreams. They influence consideration without triggering a trackable action. They change someone’s mind in a way that no pixel can capture.

Because they’re hard to measure, they’re easy to cut. And because the damage from cutting them is delayed — the funnel doesn’t empty immediately, it empties six months later — nobody connects the cause to the effect. The budget gets reallocated to conversion channels. The dashboard looks healthy right up until the point it doesn’t.

I’ve seen this play out in student accommodation, where the booking journey stretches across months. A student might encounter a brand through a parent’s recommendation, a TikTok from a current resident, a ChatGPT recommendation, a friend who lived there the previous year, a review site, and a Google Ad, in that order. Last-click gives all the credit to the ad. The marketing team, following the data, shifts budget toward conversion ads and away from the brand, content, and community work that actually started the journey. Six months later, the pipeline thins. The team is confused. The dashboard says everything is optimised.

This isn’t a sector-specific problem. It happens in e-commerce, where brands cut content marketing budgets because paid search looks more efficient, only to find that branded search volume drops three quarters later. It happens in SaaS, where companies defund community programmes because the attribution model can’t connect a Slack conversation to a signed contract, even though every sales team knows that community is where pipeline starts. It happens in retail, where word of mouth — arguably the most powerful conversion channel in existence — shows up as zero in the attribution model because nobody clicked anything.

The pattern is always the same: what you can’t measure, you cut. What you cut, you lose. What you lose, you don’t notice until it’s too late.

The Same Channel, Two Different Jobs

The misattribution problem isn’t just cross-channel. It exists within individual channels, and in some ways that’s more damaging because it’s harder to spot.

Take paid search. A generic search ad — “best running shoes” or “accounting software for small business” — is a discovery play. It puts a brand in front of someone who didn’t know it existed. The value of that impression extends far beyond the session. Even if the person doesn’t convert, they’ve now got a brand name in their head. They might come back later through a completely different channel. They might mention it to a colleague. They might search for it again in three weeks.

A brand search ad — someone typing your company name into Google and clicking the top result — is a conversion play. That person already knows who you are. They’ve already decided to visit. The ad is catching demand, not creating it.

Last-click treats both of these as “paid search.” The budget model doesn’t know the difference. So when the CFO asks “what’s our cost per acquisition?” the answer is a blended number that tells you almost nothing about which of those two jobs is actually driving growth.

You could be over-investing in brand search ads that are catching people who were already coming, while under-investing in generic discovery ads that are actually filling the top of the funnel. Or you could be doing the opposite. The model won’t tell you. It can’t. It was never designed to.

The AI Blind Spot

The attribution problem is about to get significantly worse. AI assistants — ChatGPT, Gemini, Perplexity, Claude, and others — are increasingly part of the consideration phase across every industry. They make recommendations. They form shortlists. They shape the set of options a person considers before they ever click anything.

None of this is visible to last-click attribution. Someone asks an AI for recommendations, receives a shortlist that includes your brand, and three weeks later searches your name and clicks an ad. The AI did the heavy lifting. The ad caught the overflow. The dashboard gives the credit to the ad.

This isn’t a marginal channel. In some categories, AI-mediated discovery is already the primary way that consumers form their initial consideration set. If your attribution model doesn’t see it, you’re not just missing a channel. You’re missing the fastest-growing influence on the top of the funnel. And you’re making budget decisions as though it doesn’t exist.

The companies that figure this out first will have a significant competitive advantage. Not because they have better AI, but because they understand what their attribution model is actually measuring — and what it isn’t.

The Comfort of False Precision

There’s a reason last-click attribution persists despite its well-documented limitations. It’s comfortable. It gives you a number. That number feels definitive. You can put it in a slide deck and everyone nods.

Multi-touch attribution, incrementality testing, media mix modelling — these are harder. They require more data, more sophisticated analysis, and a willingness to accept ambiguity. They don’t give you a single clean answer. They give you a range of possibilities with different confidence levels.

Most organisations would rather have a wrong answer that feels right than a right answer that feels uncertain. The dashboard rewards this preference. It presents incomplete data with the visual language of certainty: clean lines, precise percentages, definitive rankings. The interface itself is an argument for confidence.

This is Christine Barnett’s point, and it’s the one that matters most. The problem isn’t that last-click attribution exists. It’s that it exists inside an interface designed to make you trust it. The number on the screen comes with an implicit guarantee of objectivity that the methodology behind it doesn’t support.

Fraud exploits this. But so does routine marketing practice. Every time a team shifts budget based on last-click data without asking what that data is missing, they’re making the same error as someone who trusts a dashboard without questioning the methodology. The scale is different. The mechanism is the same.

What to Do About It

The answer isn’t to abandon attribution. It’s to treat it as what it is: one input among many, not the final word.

Start by asking what your attribution model isn’t seeing. If you’re running last-click, it’s not seeing the top of the funnel. It’s not seeing the channels that create demand versus the channels that catch it. It’s not seeing the difference between a discovery interaction and a conversion interaction, even within the same channel.

Then look at the channels that your model consistently undervalues. Brand, content, community, organic social, PR, AI visibility. These are the channels that fill the funnel but don’t get credit for it. If you’re cutting them because the dashboard says they’re inefficient, ask yourself whether the dashboard has the information it needs to make that judgement.

Finally, talk to your customers. Not in a survey. In a conversation. Ask them how they found you. Ask them what influenced their decision. Ask them what almost made them choose someone else. The answers will not fit neatly into an attribution model. But they will tell you things that no dashboard ever will.

The Cost of Not Looking

The Phia story is about fraud. Most attribution isn’t fraudulent. It’s just incomplete, presented with a confidence it hasn’t earned.

The cost isn’t in the obvious cases. It’s in the thousands of routine budget decisions made every day, across every industry, based on a model that measures the last step and calls it the whole journey. It’s in the channels that get defunded because they can’t prove their value in a last-click world. It’s in the funnel that slowly empties while the dashboard says everything is fine.

Attribution isn’t truth. It’s a lens. And if you’ve never questioned what it’s not showing you, that’s not because there’s nothing to see. It’s because the dashboard made it easy not to look.

David Chadderton spent his twenties and thirties teaching people how to make life-or-death decisions at forty thousand feet. He now applies the same principles to consumer psychology, which, depending on the brief, can feel equally high-stakes. He’s the creator of the STAR Framework and the author of The STAR Framework: Rewriting the Rules of Consumer Engagement (NYC Big Book Award 2025), The STAR Operating System: Decode Mindset, Understand Motivation, Transform Human Behaviour, and Dear Algorithm, It’s Not Me, It’s You. By day, a Chief Marketing Officer. By night, a behavioural science obsessive who writes The Unoptimised Human because he can’t stop thinking about why people do what they do.