The question at this stage Why do customers stay or leave?
Record now, or lose it: a daily snapshot of your records. Without one, you cannot see how things looked on a past day.
The five steps at this stage
- Step 1Choose the problem
Pick the questions for one weekly review, where each number answers a question and leads to an action.
If you skip it: The review becomes a tour of charts with no decisions.
- Step 2Collect what you need
Bring product, payments, sales, support and ad spend together under one ID. Turn any spreadsheet that acts as a record into a table. Keep one central copy refreshed nightly, daily snapshots and the raw records, with every amount stored with its currency. Alert a named person when data arrives late.
If you skip it: You cannot connect a cancelled payment to that customer's usage, and a stalled feed goes unnoticed.
- Step 3Choose the number
Agree on a short list of key numbers, one definition each, written as metric files.
If you skip it: Two people bring different numbers with the same name to the same meeting.
- Step 4Count it right
Leave test accounts out by default. Make every import safe to rerun. Give people read-only access, and let the agent write only in a scratch area, never to the tables people rely on.
If you skip it: A rerun import counts a day twice and nobody sees it.
- Step 5Read it right
Review funnel and retention weekly. Run experiments only when you have enough traffic to read them. With few accounts, review them one by one.
If you skip it: You act on a difference that is only noise.
What can wait
- You can skip new dashboards until the question and its decision are written down.
- Until growth is repeatable you can skip a shared layer that every tool reads its definitions from. Start with the few numbers your question needs, each with a written meaning.
- Until growth is repeatable you can skip new monitoring tools. A copy of the data, a test flag and one second source to compare against come first.
- Safe by default
Write down how long you keep each personal detail, and make sure you can remove a person's details on request.
- Who runs it
The founder runs it, with an AI agent.
A part-time data lead, working a few hours a week, owns the list of key numbers.