The measurement problem nobody wants to solve.

I sat in a quarterly business review last year where four different channel leads presented four different dashboards, each showing their channel as the primary growth driver. Paid search claimed credit for sixty percent of conversions. Paid social said it was responsible for fifty-five percent. Email insisted its nurture flows drove forty percent of closed deals. And organic content took credit for thirty percent of first touches. If you added it all up, the company had driven roughly one hundred and eighty-five percent of its actual revenue. Nobody blinked.

The CMO nodded through all four presentations and approved budgets that looked almost identical to the previous quarter. Everyone left feeling validated. Nothing changed. And that, in miniature, is the measurement problem that marketing has been refusing to solve for the better part of a decade.

The comfortable fiction

Marketing attribution, as practiced by most organizations, is a consensus fiction. Not because the data is fabricated, but because every measurement framework carries assumptions that flatter the person who chose it. Last-click attribution gives all the credit to the final touchpoint, which means whoever owns the bottom of the funnel always looks indispensable. First-click does the opposite, making awareness channels look like they carry the whole operation. And the various multi-touch models that were supposed to split the difference just introduced new biases that are harder to see.

The dirty secret is that everyone in the room knows this. Channel leads know their dashboards are generous. Marketing ops knows the models are imperfect. The CMO knows the numbers don't add up. But the fiction is useful. It lets every team point to a dashboard that justifies its existence. It lets the CMO avoid making hard calls about where to cut. It lets the CFO see numbers that look rigorous even when they aren't. Everyone participates because the broken version of measurement is more politically convenient than the honest one.

Why multi-touch didn't fix it

Multi-touch attribution was marketed as the solution. Instead of giving all credit to one touchpoint, you'd distribute it across the journey. Except the distribution model itself encodes assumptions about what matters. Linear attribution assumes every touchpoint is equally important. Time-decay just makes last-click slightly less aggressive. Position-based models give forty percent to first and last touch and split twenty across the middle. Why those numbers? Because someone had to pick, and those felt reasonable.

I've worked with brands that switched multi-touch models three times in two years. Each time, the team whose channel looked worse lobbied for different weighting. The model became a political negotiation dressed up as a technical decision.

Algorithmic attribution promised to take the politics out of it. But algorithmic models are trained on conversion data, which means they can only measure what already happened. They're excellent at telling you which paths led to conversion in the past. They're terrible at telling you what would have happened if you'd taken a channel away entirely. And that counterfactual is the one that actually matters for budget decisions.

The experiment nobody wants to run

The most reliable way to understand whether a channel is actually working is to turn it off and see what happens. Incrementality testing. Holdout experiments. Geographic lift studies. These methods aren't new, and they aren't technically difficult. They are, however, terrifying.

I worked with a consumer goods brand that had been spending heavily on a particular paid social platform for three years. Their attribution dashboard said it was driving twenty-two percent of online revenue. When I suggested a holdout test, pausing spend in a few markets to measure the actual incremental impact, the room went quiet. The channel lead said the timing wasn't right. The agency said it would disrupt their optimization algorithms. The VP of e-commerce worried about leaving revenue on the table during a key quarter.

Eventually they ran the test. The incremental lift was closer to four percent. Not zero, but nowhere near twenty-two. They reallocated and saw overall performance improve. But it took six months of internal advocacy just to get permission. If you're only willing to test the channels you're already skeptical about, you're not doing measurement. You're doing confirmation bias with a spreadsheet.

It's an org design problem, not a data problem

The measurement problem is typically framed as a technology challenge. We need better tools, better data, better models. And yes, the technical infrastructure matters. But I've seen companies with world-class analytics platforms produce the same politically convenient nonsense as companies running everything through basic UTM parameters. The technology isn't the bottleneck.

The bottleneck is organizational. In most marketing departments, channel leads are evaluated on channel performance. The paid search manager's job security depends on paid search looking good. The social media director's budget depends on social metrics going up. When you build an org where every person's incentive is to prove their channel works, you've built an org that will never honestly evaluate whether any channel works.

The brands I've seen actually solve this problem did something counterintuitive. They stopped evaluating people on channel metrics and started evaluating them on portfolio contribution. They built testing cadences into the annual plan, so incrementality experiments were just how the team operated. One outdoor brand I worked with required every major channel to prove its incrementality annually through actual holdout tests. The first year was brutal. Two channels considered core performers turned out to be largely duplicating conversions that would have happened anyway. But because the testing was systemic, it didn't feel like an indictment. It felt like learning.

What honest measurement actually looks like

Honest measurement starts with admitting what you don't know. Most marketing organizations claim far more certainty about what's working than the data supports. The first step is acknowledging that your attribution model is a useful heuristic, not a source of truth. That your dashboards show correlation, not causation.

From there, you build a testing culture. Not a testing tool, not a testing team. A culture where running experiments is normal, where results that challenge assumptions are celebrated rather than buried, and where the person who discovers a channel isn't working gets promoted rather than blamed.

This is harder than it sounds. It requires leadership willing to say "we're not sure" to a board that wants certainty. It requires channel managers secure enough to let their numbers be questioned. But the brands that do this outperform over time, because they're making decisions based on what's actually true rather than what's politically comfortable. They reallocate faster. They waste less. And everyone else is still arguing about attribution windows.

The measurement problem isn't a mystery. The methods exist. The math works. The reason it doesn't get solved is that solving it means someone has to be willing to learn that their favorite channel, their biggest budget line, their career-defining initiative might not be doing what they think it's doing. Until that's a safe thing to discover, every dashboard will keep telling a flattering story. And marketing will keep buying it.