Standing up the stack — event collection, product analytics, a warehouse, reporting on top — is a well-understood project. Weeks, not years. Any competent team can do it.
What is hard is the part nobody scopes: changing what happens in the room where priorities are set.
Why data usually loses the argument
It arrives after the decision. The roadmap is agreed in a planning cycle; the data lands three weeks later. By then the decision is committed and the analysis becomes retrospective justification, which nobody reads.
Nobody trusts the number. Two dashboards disagree, someone notices, and the meeting relitigates the definition rather than the question. Once trust breaks, it does not partially recover — every future number is discounted.
It has no owner in the room. If no one is accountable for bringing the evidence, the loudest opinion wins by default. That is not a failure of the loud person; it is a vacancy.
An organization does not become evidence-led by acquiring evidence. It becomes evidence-led when the cost of asserting something unsupported goes up.
What actually moved it
One definition per metric, written down, with an owner. Unglamorous and the highest-leverage thing we did. "Active customer" meant four things across four teams; agreeing one definition made every subsequent conversation cheaper.
Instrument the funnel before the features. Complete, trustworthy coverage of the core commerce journey mattered more than rich instrumentation on any individual feature — because the funnel is what people argue about.
Put the data in the planning meeting, not in a dashboard. Dashboards are pull; planning is push. The behaviour changed when the funnel numbers were on the first slide of the prioritisation session, every cycle, whether or not they were flattering.
Let it kill something visible. The turning point was a feature we had built and were fond of, which the data showed almost nobody used. Retiring it publicly, on the evidence, did more for the credibility of the practice than a year of advocacy.
The test
Can you name a decision in the last quarter that went differently because of data? If not, you have reporting, not analytics — and the difference is entirely in whether anyone is willing to be contradicted.
The B2B caveat
Consumer analytics practice assumes high volume and short cycles. Business buying is neither. Purchase cycles run weeks, the same account has multiple users with different roles, and the meaningful unit is often the account rather than the session.
A/B testing that assumes consumer traffic volumes will mislead you at B2B sample sizes, and cohort analysis anchored on individual users misses that the buying decision is frequently made by someone who never logs in. Getting the unit of analysis right matters more than the sophistication of the technique applied to the wrong one.