Attribution is theatre. Use a KPI tree instead.

A marketing KPI tree is the tool most teams should be running instead of a multi-touch attribution model. Attribution promises to tell you which touch caused the sale. It cannot. It never could. What it produces is a confident-looking number built on cookies that no longer persist, identities that no longer resolve, and modeled guesses dressed up as measurement.

I have sat in too many reviews where a team celebrated a channel because the last-click dashboard rewarded it, then watched revenue stay flat when they doubled the budget. The dashboard was not measuring cause. It was assigning credit. Those are different jobs, and only one of them helps you decide anything.

So retire the attribution debate. Build a tree of metrics you can actually influence, then test the links with real experiments. Here is how that works.

Why attribution stopped working

Attribution was always a shaky idea, but three shifts turned it into theatre.

• ▸Cookie deprecation and tracking prevention gutted the signal. Safari and Firefox block third-party cookies outright, and iOS App Tracking Transparency opt-in rates sit near 25 percent. A large share of your journeys are invisible before the model even runs.
• ▸The gaps get filled by modeling. Platforms now stitch missing touches with statistical guesses, then report the guess as if it were an observed fact. You are auditing a black box you did not build.
• ▸Last-click still dominates in practice. Even teams that bought a multi-touch tool default to last-click when the numbers need to add up, because it is the only view everyone trusts, which tells you how much they trust the rest.

Attribution answers a question you did not ask: who to thank. The question you asked was what to do next.

Reza Hadid, Analytics Lead

What a marketing KPI tree actually is

A marketing KPI tree is a decomposition. You take the one metric the business truly cares about, the North Star, and you break it down into the inputs that mathematically produce it. You keep breaking each input down until you reach a metric a specific team can move with a specific action.

The logic is simple arithmetic, not correlation. Revenue equals traffic times conversion rate times average order value times purchase frequency. Each of those splits again. Traffic splits into sessions by channel. Conversion rate splits by device, by landing page, by segment. The tree makes the relationships explicit, so an argument about priorities becomes an argument about numbers.

Inputs, not outcomes

The point of the tree is to push your attention down to metrics you own. Revenue is an outcome you cannot touch directly. Landing page conversion rate is an input you can change on Tuesday. Attribution obsesses over the top of that chain. The tree forces you to the bottom, where the levers are.

How to build the KPI tree, step by step

1. 01Name one North Star metric. Not four. Pick the number that best reflects durable value delivered, for example net revenue or activated accounts. If two candidates compete, the business has a strategy problem, not a measurement problem.
2. 02Decompose it into a driver equation. Write the North Star as a product or sum of inputs. Revenue equals sessions times conversion rate times average order value. Resist the urge to skip this; the arithmetic is the whole point.
3. 03Expand each driver one more level. Break sessions into paid, organic, direct, and referral. Break conversion rate by the two or three segments that behave differently. Stop when a named team can act on the leaf.
4. 04Attach an owner and a baseline to every leaf. A metric with no owner is a chart nobody defends. Record the current 28-day value so you can see movement, not just level.
5. 05Tag each link with a confidence level. Some parent-child relationships are proven, some are assumed. Mark the assumed ones. Those are your experiment backlog.
6. 06Rank the leaves by expected impact times feasibility. This gives you an ordered list of what to test first, replacing the last-click dashboard as your prioritisation tool.

Prove the links with incrementality, not credit

A tree tells you what should drive the North Star. It does not prove that spending more on a given leaf moves it. For that you need incrementality testing, which measures the counterfactual: what would have happened without the spend. Three methods cover most cases.

• ▸Geo holdouts. Turn a channel off in a set of matched regions, keep it on elsewhere, and compare. This is the cleanest read on incremental effect you can get without a lab, and it needs no user-level tracking at all.
• ▸Marketing mix modeling. Regress the North Star against spend and external factors across time. MMM is coarse and slow, but it is privacy-proof and it sees channels attribution cannot, including offline and brand.
• ▸Conversion lift tests. Randomise exposure at the user or audience level where a platform still supports it, then measure the difference in conversions. Treat the platform's own lift number with the same suspicion you would any self-graded exam.

None of these tells you which of five touches deserves the credit. That is the point. They tell you whether the money did anything, which is the only measurement a budget decision requires.

Running decisions off the tree

The weekly review changes shape. You stop opening a last-click dashboard and start at the tree. Which leaves moved against baseline. Which assumed links are still unproven. Which experiment reported a clean result you can now bake in as fact.

Budget follows tested links, not credited channels. When a geo holdout shows a channel drives a 9 percent incremental lift, you fund it. When the same channel that attribution loved shows no lift in a holdout, you cut it, and you stop being fooled by the credit it was hoarding.

The cultural shift matters as much as the method. A tree makes disagreements concrete. Nobody argues about whether the model is fair. They argue about which leaf to test next, which is a far more useful fight.

At Aivento we build these trees with clients in the first fortnight, then wire every experiment back to a named leaf, so the roadmap and the measurement are the same document. Stop grading touches. Start moving inputs.