Turn a GA4 funnel export into three ranked conversion tests with Claude, then lower its confidence scores yourself

A founder staring at a GA4 funnel does not know which drop-off to fix first or how to turn it into a testable idea.

Export the full GA4 funnel exploration with the rate at every step, split by device, plus the traffic sources of the weakest sessions and key-page behaviour. Paste it as a clean table, not a screenshot. Tell Claude the product, price, audience, funnel goal, past tests and constraints. Ask for the three biggest leak points, each with an observation, hypothesis, proposed change, target metric and how to measure it, and tell it to say when data is insufficient. Have it score the ideas with ICE or PIE and return a sorted table. Then lower its Confidence scores, correct Effort to what your developer can do, and choose the test order yourself.

No test results are reported. The author says the process turns raw data into a prioritised list in minutes instead of an afternoon, and that figure is unmeasured. The worked example uses invented numbers, and the post says so.

The author sells CRO and PPC consulting and links to a booking page throughout. He says Claude can invent benchmarks that look like industry figures, overrates its own confidence, remembers no past failed tests, and cannot separate cause from correlation without help. Low-traffic sites may never reach significance on an A/B test. Output quality depends on clean data and specific context.

A GA4 property with a configured funnel exploration and enough traffic to test, a Claude account, and a developer or tool to build the test variants.

• Lionel Fenestraz (freelance PPC and CRO consultant) ↗ Tutorial, published August 5, 2026

Source published: August 5, 2026. Last reviewed here: October 11, 2026. Spot a mistake? Tell us.