The Attribution Trap: Why Your Marketing Metrics Lie
Why marketing's most trusted number is its least reliable - and the humbler one that actually tells the truth.
Every marketer has a number they live by. Cost per sale. Return on ad spend. Conversions by channel. It feels like bedrock — the closest thing to the money, therefore the closest thing to the truth.
We recently worked through a client's data and found that this bedrock number was quietly lying. Worse: the more confident and precise it looked, the less it could be trusted.
Here's what we learned, and the paradox at the centre of it.
The setup
The client sells a considered purchase online - the kind with a multi-step journey. A prospect gets a quote, completes an application to generate a draft, then comes back and pays to activate. A clean, familiar funnel.
They ran the usual mix: paid search, paid social, and an automated "performance" campaign type, alongside email, organic, brand search and direct. And every month the report led with the number everyone wanted: sales, by channel, at last click.
By that number, the story was clear. Email and brand search were the heroes. Paid social looked like a laggard barely worth the spend.
That story was wrong. Not slightly - backwards.
The problem: harvesting isn't winning
Email and brand search closed most of the sales. But they weren't winning those customers. They were harvesting them.
Picture the real journey. Someone discovers the brand through a paid social ad. They don't buy. They get a draft. A week later an email nudges them back, or they Google the brand name directly, and then they complete. Last-click attribution hands that sale to email or brand search. The channel that actually created the customer - the paid ad that started it all - gets nothing.
Once you see this "harvest dynamic," every downstream metric collapses for the paid channels. Cost per sale is overstated. Channel-level ROAS is fiction. The channels doing the hardest, most valuable job - creating demand from scratch - were being robbed of the credit, while the channels that simply mopped up existing intent took the applause.
The fix: measure the honest step
So we changed what we measure. We stopped judging paid channels on sales, and started judging them on cost per draft - the last step in the journey we could still reliably attribute to the channel that won the customer. Everything past that point is contaminated by the harvest.
The moment we did, the picture inverted. The "laggard" paid channel turned out to be the most efficient source of new customers. Budget, it emerged, had been allocated almost exactly opposite to real performance - because the reporting metric had been pointing the wrong way the whole time.
The paradox
Here it is, plainly:
The numbers that look most like performance - closed sales, by channel, at last click - are the ones you can trust least and control least. The humble number, cost per draft, is the one that tells the truth.
Why are the confident numbers the treacherous ones?
They're the most granular - and granularity feels like rigor. A channel-level, bottom-of-funnel number looks like precision. But that's exactly where the harvest noise is loudest. Zooming in didn't increase accuracy. It reduced it.
They're the outcome everyone cares about - so nobody questions them. Sales is the point of the business, so the sales number gets a free pass. The emotional weight of a metric turns out to be inversely related to its reliability.
They flatter the wrong channels and starve the right ones. A wrong number doesn't just misinform. It actively steers. Ours had been quietly funnelling budget toward the least efficient channels for months.
Three smaller paradoxes that fall out of it
Your best-looking channels depend on your worst-looking ones. Email can only harvest drafts that acquisition created. Cut the "loser" and you starve the "winner."
The better your harvest engine performs, the worse acquisition looks. Every sale email closes is a sale stripped from its true source at last click. Success in one place manufactures apparent failure in another - inside the same funnel.
The harder you chase clean attribution, the more it proves clean attribution is impossible. The deeper we dug, the clearer it became that a meaningful share of customers trace back to an untrackable email/offline bucket. Perfect attribution is a mirage. The move isn't to chase it - it's to manage to an honest number instead.
What to do about it
Separate what you can buy from what you can't. Paid search, paid social and automated campaigns are dials you turn. Email, brand, organic and direct are byproducts - dividends of awareness and owned assets. Don't budget them; read them as signals.
Judge each layer on its own job. Acquisition on cost per draft (or your last cleanly-attributable step). Harvest channels on conversion effectiveness. Sales as a whole-system outcome - reported, not carved up by channel.
Set targets on what teams actually control. Holding a team to a sales figure that email quietly harvests is both unfair and misleading. Give them a draft-volume and cost-per-draft target they can own.
Treat the satisfying number with suspicion. If a metric is granular, bottom-of-funnel and flattering, question it hardest. That combination is where the illusions live.
The takeaway
The instinct in marketing is to trust the number closest to the money. But the closer a number sits to the sale, the more hands have touched it - and the less any single channel can honestly claim it.
Sometimes the truest signal is the boring one further up the funnel. Trust the humble number.
Written by the team at Unfold. If your channel reporting rewards the channels that harvest and punishes the ones that acquire, you may be optimising your budget in exactly the wrong direction. We're happy to talk it through.
