The Full-Day Audit: Evidence, Recordings and the Funnel File

The full-day audit produces one artefact: the funnel file. Its contents: the baseline and current rates by segment; for each weak stage, a recording sample (twenty to thirty sessions), the visitor-voice evidence (survey answers, ticket tallies), and findings written as stage, cause, evidence, repair, read rule. The day runs in four blocks: data (morning one), recordings and voice (morning two into afternoon), findings written cold (afternoon), and the fix list drafted against the prioritisation method (last hour). No repairs are made during the audit day; mixing diagnosis and treatment is how audits turn into unrecorded redesigns.

Assemble the numbers before judging any of them: funnel steps by device, source and new/returning; landing page rates; the search log's top queries and zero-result queries; form completion where forms matter; post-sale cohort rates. Note data gaps as findings in their own right. The block ends with the stage ranking: the two or three stages where unintended loss is largest, which ration the rest of the day. An audit that tries to deep-dive every stage finishes none; the ranking is the day's first real decision and it is made from numbers, not from the loudest page.

For each nominated stage, work the diagnosis method from the leak bench: a recording sample split between leavers and passers, watched for behaviour; the visitor's own words from the survey question, the tickets, the chat logs, tallied by theme; and the stage's pages read against their doubt lists. Keep evidence verbatim - quotes and timestamps, not paraphrases - because the findings meeting (even if it is you, later, sceptical) should be able to re-try each finding against its exhibits. Evidence discipline is what separates the deep audit from the hour audit multiplied by eight.

Write the findings away from the pages, in the standard format, each with a confidence note (proven by recordings and voice; indicated by rates, awaiting recordings; a lead, not yet a finding). Resist repair language creeping into causes: 'visitors cannot find the delivery date' is a cause; 'add a delivery banner' is a repair, and it belongs in the next block where alternatives can compete. Findings written cold, in prose, are the audit's intellectual product - the part that stays true when the fix list is reprioritised by cost or by season.

Convert findings to repairs, score them with the prioritisation method (leak size, evidence strength, cost), and sequence the quarter: what ships, in what order, with what read rules, and what is deliberately deferred with its reason. The funnel file closes with the verification calendar - the dates each repair will be read - so the audit schedules its own accountability. File the whole thing with the change log; next year's deep audit starts by re-trying this year's findings against what the repairs actually did, which is how an audit practice learns.

The deep audit refuses three things: competitor page tours (their evidence cannot enter your findings, however inspiring), redesign sketching (the audit's output is findings, not layouts), and mid-audit fixes beyond genuine emergencies (a live defect taxing orders today is fixed and logged as an incident, not smuggled in as audit work). These refusals keep the day diagnostic. The business's impatience to fix is honoured by the calendar: the fix list ships the following week, in order, verified - a short delay that buys the difference between treatment and guesswork.

The deep audit needs the people who hold the evidence: whoever reads support tickets, whoever runs the packing or delivery end, whoever owns the numbers - in a small business often one person wearing the hats in sequence, which is fine, provided the hats are actually changed. The support view and the fulfilment view regularly contradict the dashboard's story, and their exhibits (the ticket thread, the returned item) belong in the funnel file beside the recordings. If the audit is outsourced, these are the interviews to insist on; an audit conducted from analytics access alone has declined the business's best evidence before starting.