The Grandmother Test
Twelve plain answers to the questions people actually ask about their analytics — each one with the picture, the working, and what to do on Monday.
If I can't explain it so your grandmother follows it, I haven't understood it yet. That's a standard I hold myself to, not a comment on your grandmother — who, in my experience, grasps "count the till, not the door" considerably faster than most steering committees.
Can I trust the numbers?
TrustWhy don't GA4 and Meta numbers match?They never will, and making them match is the wrong goal. What matters is whether the gap is stable.…
PlumbingOur numbers live in eight tools and no two agreeThe fix is never a ninth tool. Pick one system as the truth — almost always billing — and make every other number explain its difference from that one. Reconcile before y…
DefinitionsNobody can agree what "active user" meansThere is no correct definition, and hunting for one is the mistake. A definition is a business judgment call, so write it down with seven fields and test it: hand it to t…
InstrumentationHow many things should we actually track?Fewer than you think — around 30 distinct events is where most teams start to struggle. Past that, analysts lose days choosing between near-duplicates like "subscription …
Where does the money come from?
AttributionWhich channel actually pays?Add up what your ad platforms claim and you will get more sales than you actually made — often 30–60% more. That is not fraud, it is arithmetic: each platform counts ever…
MoneyWe know people churn. We don't know who, when, or what it's worthThree numbers make churn actionable, and an average tenure is none of them: which segment (with its count), when in the lifecycle the risk actually peaks, and what a save…
RetentionWhy is my retention curve lying to me?Almost always because it pools cohorts from different periods. A blended curve mixes people who joined under an old product with people who joined under a new one, and th…
Is this work worth doing at all?
The real oneSomeone asked me for a dashboard and I can't work out what forAsk one question before building anything: what will you do differently depending on the answer? Roughly 60% of analytics requests quietly disappear at that question, bec…
TimeMonth-end reporting eats three days and nobody reads itAlmost always, the fix is not a faster tool — it is deleting most of the report. Name who opens each tile on a Monday and what they change because of it; anything that fa…
ExperimentationDo we have enough traffic to A/B test?Usually less than you hope. To detect a 10% relative lift on a 3% conversion rate at conventional thresholds you need roughly 25,000 visitors per variant — so a site doin…
What does it cost, and can AI do it?
PricingWhat does an analytics consultant actually cost?Published 2026 bands: a one-off tracking fix runs $199–$500; an audit with a roadmap $475–$2,500; a paid assessment credited against a build $3,000–$6,000; a mid-market b…
AIShould I let an AI answer my team's data questions?For curiosity, yes — and it will save your analyst a great deal of time. For anything someone will act on, only with guardrails: written definitions it must follow rather…
Still not sure whether your data can answer your question? That's what the €400 Data Readiness Check is for.
See prices →