CRO Audit With a Backlog Sized to Your Traffic

An audit that returns forty ideas has not done the arithmetic. On a site sending 40,000 monthly sessions to the funnel step in question, with a 3 percent baseline, you can resolve roughly one test at a time at a minimum detectable effect worth shipping — eight to twelve decided tests a year. This audit produces a backlog sized to that number, in order, with the ideas your traffic will never decide separated out and marked build or drop. Before any opinions: The first days go to the measurement layer. The conversion event must fire once and only once. GA4 or Amplitude and the experiment tool must agree on the denominator within 2 percent. Consent-mode gating must not be quietly removing a third of the sample, and bot and internal traffic must be out. On a pet nutrition subscription site this step found a duplicate purchase event that had been inflating reported checkout conversion since a tag manager migration eleven months earlier. Nothing later in the audit is worth anything if this part fails, and if it fails badly I say so and we stop until your team has fixed it. How a hypothesis earns its rank: Each candidate carries four things: the observed behaviour it is built on and where I saw it, the mechanism it assumes, the primary metric with a direction committed in advance, and the runtime at your current traffic to detect the smallest effect that would justify shipping. Anything whose runtime runs past ten weeks does not enter the test queue. It moves to a separate list with a recommendation to ship on judgement, ship behind a holdout, or leave alone. That separation is the part most backlogs skip, and it is why most backlogs stall around test four with three inconclusive results and a team that has lost interest. What you receive: A funnel model showing where sessions and revenue leave, by step and device, with the patterns specific to your catalogue named rather than generic ones — freight quoting and lead-time display, subscription cadence choice, the handoff from web to app store. A ranked queue of hypotheses, each with its runtime and detectable effect attached. A build-without-testing list. One analysis plan template your team keeps and reuses. A ninety-minute session where I defend the ranking and you argue with it. Not included: No implementation, no test builds, no design, no copy. This engagement ends at the decision of what to test and in what order. Execution is a separate arrangement, and I would rather you hand the backlog to your own team than assume you need me to run it. Who should not buy this: Sites below roughly 8,000 monthly sessions to the step being optimised. At that volume experimentation is theatre, and five moderated sessions and one decision will serve you better for less money. Teams who want an audit to ratify a redesign already sitting in Figma; I have been hired for that twice and refunded both. Anyone who needs the deliverable to promise an expected uplift figure. I will tell you the size of effect your traffic can detect. Nobody can tell you the size of effect the test will find, and a consultant who puts that number in a proposal is describing their sales process rather than your site.