The book that walks through the discipline of testing rather than the tooling. Why most A/B tests fail to produce decisions (poorly defined hypotheses, premature stops, small effect sizes treated as wins), how to set up a test that actually runs to significance, and the cultural question — what tests do you run when, and how do you keep the testing system from being captured by HiPPO opinion. Worked examples across landing pages, ad creative, email subject lines, and pricing pages. Aimed at the marketer or growth lead who's been running tests for a while and noticed that most of them aren't producing the decisions they were supposed to.

The walkthrough for getting from 'we should test this' to a clean live test inside a week. Covers hypothesis statement (the structural difference between a hypothesis and a guess), audience definition, sample size calculation (the part most teams skip and pay for later), the variant design rules that protect interpretability, and the go-live checklist that catches the implementation issues — analytics tracking, edge cases, the not-rare-enough scenarios where the test infrastructure breaks the page. Worked example all the way through one full test. Pair with the testing-element-selection guide for the upstream question of what to test in the first place.

The framework for choosing which tests to run when you have more ideas than capacity. Covers the impact-versus-effort grid (with the calibration most teams get wrong), the leverage hierarchy (why headline tests usually beat button-color tests, and when they don't), the dependency check that prevents running tests that contaminate each other, and the test backlog rhythm that prevents the team from accumulating untested 'someday' ideas. Built for the team that's already running tests and trying to be more deliberate about which tests; less useful for an org that hasn't started yet.

Seven daily emails that take an ad campaign from running to optimised — using A/B testing as the structural lever rather than the topic. Day 1: audit current campaign performance with the four-metric read that surfaces actual problems. Day 2: prioritise the test queue. Days 3-5: run three concurrent tests on creative, audience, and offer with non-overlapping samples. Day 6: read the results without the most common interpretation errors. Day 7: scale, iterate, queue the next round. Designed for paid social and search; the structure transfers to native and display with minor adaptation.

The prompts that turn AI into a useful collaborator on the analysis side of paid advertising — not the creative-generation side, where most prompt packs concentrate. Audience-segment hypothesis generation from raw performance data, attribution model sanity-check, channel-mix optimisation against constraints, the post-mortem structure for campaigns that failed (so the lessons survive). Each prompt comes with the input format and the output format expected. Tested across Claude and ChatGPT with notes on which model handles which kind of analysis better. Pair with the testing-element-selection guide for the upstream prioritisation.

The current state of the testing-tool landscape for paid advertising — what each platform actually offers (versus what its marketing claims), where the testing infrastructure is built into the ad platform (Meta, Google) and where you need a separate layer (web tests, server-side experimentation). Names specific tools at specific spend levels: when Optimizely is overkill, when Google Optimize-class free tools are enough, when you need server-side. Includes the integration patterns that hold up at scale and the ones that drift when the team grows. Updated against current pricing and feature sets, not the version that was true two years ago.