From reports to decisions: five stages of measurement maturity

Ask a team how mature their marketing measurement is, and the answer is usually a list of tools. Tools matter less than you'd think. What matters is how much of the budget rests on evidence you could defend.

Here's a simple five-stage view. Most teams will recognise themselves in one of the first two.

Stage 1: Reports

Decisions rest on platform dashboards and channel reports. Each platform reports its own success; nobody reconciles the totals.

Next step: agree a single source of truth for spend and outcomes, and stop adding up platform-reported conversions.

Stage 2: Attribution

A shared attribution model assigns credit across channels. Better than stage 1, but it measures where conversions were recorded, not what caused them, and privacy changes keep eroding it.

Next step: build the weekly dataset (spend, outcomes, price, promotions, events) that causal methods need.

Stage 3: Foundations

A clean, weekly, documented dataset exists and is maintained. It's often invisible work, and it's what makes everything after it possible.

Next step: run a first mix model, with ranges and diagnostics, and check its readiness first.

Stage 4: Models and tests

A mix model estimates what each channel contributed, with ranges. Geo tests are run on the channels the model is least sure about.

Next step: connect them: feed test results into the model, and track forecasts against actuals.

Stage 5: Calibrated decisions

Models and tests check each other in a regular cycle. Budgets are planned with ranges and compared by the chance each plan beats the current one. Forecasts are tracked, and finance trusts the numbers because they've seen them checked.

Next step: keep the cycle running. Maturity is a habit, not a destination.

How to use this

Be honest about where most of your budget decisions actually sit, not where your most advanced project sits. A team with one impressive model but budgets still set from platform reports is at stage 1 or 2 for the decisions that matter.

Then aim one stage up. Each step pays for itself, and skipping stages (a sophisticated model on messy data, say) is the most common way measurement projects disappoint.