The CRO loop

When do you reach for The CRO loop?

Built for whoever owns conversion on a small team and has no research ops function behind them. It assumes one person, a testing tool, and a repo they can put a markdown file in.

What problem does the loop solve?

Most conversion programmes do not fail on statistics, they fail on memory. Tests get run, results get discussed in a thread, the person who ran them leaves, and eighteen months later somebody proposes the same test again with a different button colour. The loop puts one file at the centre and makes every step read from it or write to it, so the programme compounds instead of restarting.

How do the six steps run?

In order, one skill per step. Each hands the next one its input, and two of them touch the learnings file: step two reads it before anything gets built, step six writes to it once the test is done. Nothing else edits that file by hand.

• Notice. A metric drop, a support thread, a session replay. How Might We generator turns the complaint into questions and the questions into ideas.
• Check. Learnings check searches every finished test for that idea, weighted by age and confidence, and returns proceed, proceed with changes, or reconsider.
• Hypothesise. Test hypothesis writes who you are helping, what changes and what you will learn, carrying the tests the check surfaced as its evidence.
• Build. QA pass runs the built variant at seven widths before traffic reaches it, because a variant broken on a phone produces noise rather than a result.
• Decide. Test iteration establishes what actually happened, then generates ten next moves across eight angles rather than one obvious follow-up.
• Record. Learnings log writes the outcome in four fixed sections, then promotes only what is proven into the digest, tier and date attached.

What does a week actually look like?

One test in flight, one being written up, one being drafted. Monday, read the running test against its guardrails and do nothing else to it. Midweek, write the next hypothesis, which starts with a learnings check and quite often ends there because the answer already exists. Friday, conclude whatever reached its end date and log it the same day, while the context is still in your head and not in a dashboard you will misread in a month.

Where does the loop live in a repo?

In one file of its own, docs/cro-loop.md, sitting next to the learnings digest it points at. Not in CLAUDE.md. That file is read at the start of every session in the repo, so a six-step conversion programme parked there gets loaded for work that will never use it, and pasting it in wipes whatever the repo already had. Save the block below as the playbook, then add one line to CLAUDE.md so a session knows where to look: "Conversion work: read docs/cro-loop.md first." The rules then arrive when the work does. Change the paths to match your repo and leave the rest alone until you have a reason from your own data.

Which skills does The CRO loop run on?

6 skills, in the order the playbook runs them. Each one has its own page with the block to copy.

1. 1How Might We generatorTurn a customer problem into three to five How Might We statements, then into specific website ideas for each one, with where it goes and how you would test it.
2. 2Learnings checkBefore you build a test, find out whether your team already ran it. It searches the finished tests, weighs each one by how recent and how proven it is, and tells you to proceed, change it, or think again.
3. 3Test hypothesisWrite an A/B test up properly before it gets built: who you are helping, what they are trying to do, what you are changing, why you think it will work, and what you will learn either way.
4. 4QA passPut a page through the same checks every time: how it holds up at seven screen widths, what the markup says, whether anyone on a keyboard can get through it, and whether the words are right.
5. 5Test iterationA test finished. Get ten specific things to try next, each with what changes, why it might work, and what success would look like, grounded in what the last one showed.
6. 6Learnings logWrite a finished test up in four fixed sections, then decide what earns a place in the file everyone searches before proposing the next one.

What you copy

One block, ready to paste. Nothing else to install unless the block says so.

1# The CRO loop
2
3The process every test in this repo goes through, and the rules it has to hold
4to. Read this file before any conversion work and follow it in order.
5
6## The loop
7Every test goes through the same six steps, in order. Skipping one is how a
8programme ends up re-running its own experiments.
9
101. `hmw-generator` - a complaint or a metric drop becomes questions, then ideas
112. `learnings-check` - the idea is searched against every finished test before
12   anyone writes it up. Returns proceed, proceed with changes, or reconsider
133. `test-hypothesis` - the idea becomes a written hypothesis, carrying the
14   tests the check surfaced as its evidence
154. `qa-pass` - the built variant gets checked at seven widths before traffic
16   reaches it. A variant broken on a phone produces noise, not a result
175. `test-iteration` - the finished test becomes ten next moves, worst case
18   included
196. `learnings-log` - the outcome becomes an entry, and only what is proven
20   reaches the digest
21
22## The memory
23- Raw record: wherever tests are tracked. That is the source of truth.
24- Digest: `docs/learnings.md`. Curated, not a dump. Every entry carries a
25  confidence tier (proven / provisional / weak) and a date.
26- Step 2 reads it, step 6 writes it. Nothing else edits it by hand.
27- Never edit a past entry to match a newer result. Add a line under it naming
28  the test that revised it.
29
30## House rules
31- No test ships without an observation and a stated mechanism.
32- The decision rule is written before the test runs, never after.
33- Minimum runtime two full weeks, whole weeks only.
34- A win that was never re-validated is provisional, not proven. Say so.
35- A segment result found after the fact is the next hypothesis, not a finding.
36- Losses are written up in the same detail as wins.
37- Never invent a metric, a sample size or a customer quote. Missing numbers get
38  written as [need: ...] and stay that way until somebody fills them in.
39

Free to use in your own work, paid work included, no attribution required. Not for repackaging into a product you sell. Full terms.

More in playbooks

The launch checklist15 Sept

Back to every playbooks entry.