From number to decision without going wrong

Measuring isn't enough: you have to decide well

Having data doesn't guarantee good decisions. You can own an impeccable dashboard and draw false conclusions from it. The last skill of the data stack is the subtlest: knowing how to interpret your numbers without falling into the classic reasoning traps. It's what separates the entrepreneur who "has data" from the one who "decides with data".

The A/B test: settle it instead of debating

Should the button be green or red? A price at €29 or €39? A short or long headline? Rather than debating on instinct, you test. The A/B test shows version A to half the visitors, version B to the other, and measures which converts best. It's the most honest way to settle a question: you let users vote with their actions.

Tools like PostHog, alternatives such as VWO, AB Tasty, or the tests built into your emailing tool make it accessible without coding. Three rules to respect:

1. Test one thing at a time, otherwise you don't know what caused the difference.
2. Wait for sufficient volume. Concluding on 20 visitors is worthless — you need enough data for the difference not to be chance (the famous statistical significance).
3. Define the success criterion in advance, so you don't tell yourself stories after the fact.

Cohorts: comparing what's comparable

A cohort analysis groups users by their arrival date (the "January sign-ups" cohort, "February"…) and tracks their behavior over time. It's the tool that reveals whether your product is genuinely improving. If the March cohort is better retained at 30 days than January's, your improvements are paying off. Without cohorts, new and old users mix and mask the real trend.

The interpretation traps to know

The most costly trap is correlation ≠ causation. "Our best customers use feature X, so let's push everyone toward X": maybe, or maybe the good customers use X because they're already engaged. Confusing the two leads to investing in the wrong cause. When in doubt, an A/B test settles it.

Confirmation bias: your worst enemy

The danger isn't only in the numbers, it's in the head of whoever reads them. You have a starting intuition and you look, often unconsciously, for the data that confirms it while ignoring what contradicts it. The entrepreneur in love with their feature will see in any number a reason to keep it.

The antidote is a simple discipline: before looking at the data, write the hypothesis and what would prove it wrong. "I think this feature increases retention; if the cohort that uses it isn't better retained, I'm wrong." Defining in advance what would change your mind is the best protection against self-delusion.

The decision rhythm

Not all decisions have the same cadence. Some are made every week (adjusting a campaign, fixing an underperforming page), others every month (reviewing core KPIs, deciding on an investment), others every quarter (reorienting strategy). Set your data reading to these rhythms: looking at your CAC every day makes you nervous and changes nothing; looking at it every month lets you decide calmly. The right measurement frequency is the one at which a decision can actually be made.

From data to action: the complete loop

graph LR
    A[Clear question] --> B[Reliable measurement]
    B --> C[Bias-free reading]
    C --> D[Decision]
    D --> E[Action]
    E --> F[New measurement]
    F --> A

Data only has value when closed by action. A precise question triggers a reliable measurement, which you read without bias, from which a decision emerges, which produces an action, which you re-measure to know whether it worked. It's this loop — not the accumulation of tools — that makes an entrepreneur someone who truly steers their business.