Decomposition tells you where a difference sits. Only an experiment estimates what caused it.
A measured movement is split by the business metrics service into the segments that compose it — channel, campaign, audience, creative, market, device, plan type. The arithmetic is exact, and the contributions sum to the total difference. What it gives you is the segment carrying most of the change, which is not the same thing as the reason for the change. Each decomposition here states which explanations it cannot rule out and names the test that would.
In practice
Demo dataWhy did CAC rise this week?
Blended CAC moved from €136.40 to €144.69 against your €140 target, a rise of €8.29, and Google prospecting carries €6.10 of it. The split is exact: €5.20 is Google's own CAC deteriorating to €184 over 14 days, and €0.90 is mix, because Google took a larger share of spend while Meta paced under budget. Evidence: the 14-day metrics window, the same attribution rule on both sides, and customer counts reconciled against invoices in your finance system. Business impact at this pace is about €2,600 of extra acquisition cost this month. Confidence is high on the decomposition and low on any single cause — the decomposition locates the money, it does not explain it. Alternative explanations remain open: 3 qualified opportunities from 41 Google leads reads as worse traffic, but a slow qualification week in the CRM looks identical in this arithmetic, and so does a competitor bidding up the same terms. What no decomposition can tell you is how many of the customers still recorded after Google clicks would have arrived anyway; a 21-day geo holdout in Spain would estimate that, the way the Netherlands holdout put 29 of 41 Meta-credited customers as incremental. Recommended action is to run that holdout and meanwhile take ACT-2294, whose value is not contingent on the answer. Constitution check: both clear — the signal change meets the 30-outcome threshold at 34, and the holdout suppresses spend rather than adding it.
The arithmetic is exact
A CAC movement of €8.29 splits into a within-channel efficiency part and a mix part, and the parts add back to €8.29. The metrics service computes the split; the model chooses which dimensions to cut by and writes the narration around the result.
A segment is not a cause
Google prospecting carrying €6.10 of an €8.29 rise is a fact about where the money went. Competitor auction pressure, a seasonal shift in plan mix, a new landing page and a slow CRM week can each produce that identical arithmetic. The copy names which of them are still live instead of picking a favourite.
The test is named, not implied
Where a causal claim is worth spending on, the page proposes the design: a geo holdout with suppressed regions matched on prior conversion rate and population, a duration, and the difference it could detect. Attribution credit and incrementality stay in separate columns, and neither is ever restated as the other.
Behind one answer
- Answer
- Meta. Its customers are credited with €14,800 more collected 90-day revenue than Google's on 15% less spend.
- Evidence
- 311 customers, 90-day window, invoiced value from the finance system, last-non-direct attribution.
- Business impact
- Moving €1,200/day would add an estimated 31 customers this month.
- Confidence
- High — 14 days above the minimum evidence threshold.
- Alternative explanation
- Meta's audience skews to annual plans. Some of the gap is plan mix rather than channel quality.
- Recommended action
- Move €1,200/day from Google Prospecting to Meta Prospecting.
- Constitution check
- Passes — within the €1,500/day movement limit.
Intelligence
The rest of the intelligence layer
The Growth Brain
The planner above your ad accounts. It decides which channel gets the next euro, and what that euro is pointed at.
Ask AI
A question resolves into a window, a cohort, an attribution rule and four service calls — and the plan is shown beside the answer.
Actions
Every change to an ad account is an action object: checked against your limits, logged, evidenced and reversible.
Recommendations
A ranked queue of what to do next — each item with its evidence, the spend it affects and the rule it clears.
Forecasts
What a budget is likely to buy over the next 30 days, with the range, the assumptions and the later score.
What Changed
A diff of your advertising since the last read: your changes, the networks' changes, ranked by money at stake.
Why It Changed
Decomposition names the segment carrying a measured difference. It does not prove that segment caused it.
Evidence
Every figure opens into the rows behind it: which service computed it, over which window, under which rule.
Business Objectives
One written objective — outcome, target CAC, maximum CAC, margin floor, markets — that every channel is judged on.
The other intelligence capabilities are listed above.
One goal. Every channel. Real business truth.
Connect your ad accounts, say what a good customer is worth, and let the Growth Brain work the objective inside limits you set.
No ad spend commitment · Connect one channel to start · [email protected]