Your attribution report can't answer the budget question

Every quarter the same meeting happens. Someone puts a channel report on the screen, the numbers look precise, and a budget moves. Nobody in the room can say what would have happened if that money had not been spent.

That isn't a failure of effort. Most marketing teams work hard on their reporting. It's a failure of the question. The report answers one question, and the meeting is asking another.

What attribution actually measures

Attribution answers a bookkeeping question: which touchpoints were present before a conversion was recorded? Last-click gives all the credit to the final one. Data-driven models spread it around using rules learned from the paths. Platform reports do their own version, inside their own walls.

All of them are useful for running campaigns. If you want to know which ad set to pause or which keyword to bid up, they're the right tools.

None of them are evidence of cause. People who were going to buy anyway still click on things. A customer who has already decided searches your brand name, clicks a paid ad and converts. The ad gets the credit for a sale it didn't make. Multiply that across a year and you have a budget built on credit, not contribution.

The question the budget actually needs

Moving money between channels is a bet on a counterfactual: if we spend less here and more there, what changes?

That is a different question. It asks what would have happened in a world we didn't observe, the one where the spend didn't happen. No report built from recorded conversions can answer it, however sophisticated its rules, because it only ever sees the world where the spend did happen.

There are two families of methods that do answer it. Marketing mix models look at all your spend, all your sales and everything else that moves sales (seasonality, price, promotions, competitors) over time, and estimate what each channel contributed. Experiments, such as geo holdouts, switch spend off or on in some places and compare them with places that carried on as normal.

Neither is perfect. Together they're the best evidence most businesses can get.

Three better questions

The next time a channel report lands, try asking these instead.

1. What does each channel return per pound, with the range? A single number hides how much we don't know. "£2.40" sounds certain. "Most likely £2.40, somewhere between £1.80 and £3.10" tells you how much weight it can bear.
2. Where does the next pound go furthest? Channels saturate. The first pound and the millionth are not worth the same, so the average return of a channel says little about what happens if you add more.
3. Which result would you bet your job on without a test? If the honest answer is none of them, that's where the next experiment goes.

What to do on Monday

Keep attribution for what it's good at: optimising inside a channel, day to day. For budget decisions between channels, look at a mix model's ranges, and test the channel you're least sure about.

And when the next report says a channel "drove" a number, ask the only question that matters for the budget: compared with what?