64% of marketers can't prove ROI. The other 36% are lying.

So AI Digital published a piece recently noting that 83% of marketing leaders say that proving ROI is their top priority, yet only 36% believe they can actually measure it accurately.

I'll give them credit for one thing: they've correctly identified this as an organizational problem, not a technical one. That's a more honest framing than most.

But then they go and spend the rest of the piece prescribing fifteen KPIs and suggesting AI-assisted media mix modelling as the fix, which is somewhat the equivalent of diagnosing someone with a broken leg and handing them a fancier walking stick.

Let me tell you what's actually going on here. The measurement problem in digital marketing isn't that we lack KPIs. We have too many.

I've sat in rooms — boardrooms, conference rooms, the kind of rooms where some AVP insists on dimming the lights for the deck — and watched grown adults present seventeen slides of engagement metrics without once mentioning revenue.

Bounce rate. Session duration. Impressions. Cost per click. Cost Per View. Cost Per Completed View. Cost Per Like. Cost Per Comment. These numbers have the comfortable quality of being easy to generate and impossible to argue with. No CFO knows enough to push back on a 4.2% click-through rate. Because she doesn't know what it means. Like a football coach reporting to the team owner that the ball was in the air 74.28% of the time, which sounds impressive but has no bearing whatsoever on who scored how many goals and whether the team won. (There, had to get my World Cup reference in!)

So it sits in the deck, looking busy, meaning nothing. The deeper problem — and this is the bit that the fifteen-KPI crowd conveniently ignores — is that most digital attribution is fiction dressed up as science.

Last-click attribution, which still runs more campaigns than anyone wants to admit in 2026, is basically astrology with a dashboard.

You're not measuring what caused the sale. You're measuring who happened to be standing nearest to the POS machine when the credit card was swiped. It's quite meaningless - though Google and Meta would not have you believe so.

First-click has the opposite problem. Multi-touch models distribute credit with the confident randomness of a committee deciding who gets the window office based on alphabetically arranging everyone's middle names.

We've been at this long enough at Pinstorm to have watched the full arc: from the early days when digital was going to solve measurement forever — finally, accountability! — to the current moment where the industry has produced more measurement tools than at any point in history, and yet only 36% believe they are measuring it. And because most of them are doing the same Google - Meta last click attribution fiction, they're probably wrong as well.

Sure, they're measuring something, but it's not ROI.

If anything, I'd argue that 36% number is generous. It assumes those who say they can measure ROI accurately are correct about that. Confidence in your measurement methodology is not the same thing as having a valid one. I think a vanishingly small number - perhaps 1/10th of that 36% are anywhere close to getting it right.

Here's what I think is happening in those organizations that do get this right — and I'll be honest, they're a small club. They've made a decision, probably a painful one, to stop treating brand and performance as separate functions with separate budgets and separate KPIs. After all are you going to have a separate budget and department for making your brand stand out among the others out there - and a separate one to stand by the cash register while it's being bought?

Because the moment you split them, you get exactly the dysfunction that produces the problem. The performance team optimizes for short-term conversions and shows beautiful ROAS numbers. The brand team talks about awareness and consideration and shows beautiful awareness lift numbers.

Neither number connects to the other, and neither team has any incentive to make the connection. The CFO sits in the middle wondering why, despite both teams claiming to be winning, revenue growth is underwhelming.

Binet and Field documented this exhaustively — their work on the IPA databank, looking at hundreds of campaigns over decades, is probably one of the more rigorous things the industry has produced. Though, truth be told, it's not rigorous enough. It's a tiny sample of the real world campaigns out there - and these samples are already pre-selected for their successes. Add all campaigns, including the ones that failed and the ones that didn't apply for the awards, and the data might be entirely different.

But even within this small and unrepresentative sample, it shows that 60% of the budget is better spent working on long-term brand, and 40% on short-term activation.

Not because it's a magic ratio, but because the data shows that's approximately where you stop leaving money on the table.

And critically — the metrics for each have completely different time horizons. Judging a brand campaign on six-week ROAS is like judging a forest by how many trees you can see in a photograph taken in week one. Absurd, but ubiquitous.

The media mix modelling push — which the AI Digital piece is quite enthusiastic about — is somewhat useful, I'll grant that. It's a step up from last-click.

But I'd caution against the idea that better modelling solves the problem. The model is only as good as the data going in, and the data going in is almost always gibberish. so garbage in, garbage out.

Dark social, word of mouth, the person who saw your outdoor campaign on the way to the office or the retail lightbox while shopping and typed your brand name directly into Google three weeks later — none of that shows up cleanly in a media mix model.

What you get is a partial picture presented with full confidence, which is usually much worse than a partial picture presented with appropriate humility.

One victim of this is often the OOH - Out Of Home - industry. They've not invested as much in justifying their existence - and hence they don't often get taken into modelled mixes and hence the output spit out doesn't tend to favor them much. But most street-savvy businessmen swear by the medium - and all of retail does - so be careful what your modelled media mix says.

The other thing nobody wants to say plainly: most brands are chasing the wrong buyers.

There is this fallacious loyalty hypothesis — that your best customers, your heavy buyers, are where the growth comes from — is seductive and wrong.

Byron Sharp and the Ehrenberg-Bass Institute have been making this point for fifteen years. Growth comes from light buyers and non-buyers. The people who buy your category occasionally, who barely think about your brand, who don't follow you on anything and couldn't name your tagline. Those are the people who move the revenue needle when you reach them, because there are so many of them.

Your loyal core were going to buy you anyway. All your targeting precision, all your CRM segmentation, all your lookalike audiences optimised on your best customers — you're spending money talking to people who were already sold. Congratulations, you've preached to the choir and they're all believers now.

I've watched clients spend serious money building loyalty programs that were simply discounting to people who'd have otherwise paid full price. The program didn't create loyalty. It rewarded existing behaviour at a cost.

Points and tiers and 'exclusive member benefits' — the unit economics on most of these things are terrible, and the attribution is circular. Of course your loyalty programme members have higher Life Time Value: they were already your best customers before they joined, silly. You enrolled them because of their high lifetime value behaviour, then measured their high lifetime value behaviour and then declared victory.

So what should you actually track? Here are some thoughts: Revenue. Market share. Penetration — how many people in your category bought you at least once this period versus last.

Mental availability — harder to measure, but trackable through brand health surveys done properly, not the nonsense vanity ones. And you need honest, uncomfortable conversations about time horizons.

A campaign that looks flat at ninety days might look very different at eighteen months.

Most organizations aren't structured to have that conversation, which is why they reach for fifteen KPIs and call it rigor.

The measurement problem is real. It's just not what people think it is.

The problem isn't that we can't measure. The problem is that we're measuring the wrong things, on the wrong timescales, in an attempt to look busy and cover-your-ass while doing so.

No amount of AI-assisted dashboarding fixes that.

It's a thinking problem pretending to be a tooling problem. And thinking problems are considerably harder for Google and Meta to solve on your advertising front.