Anatomy of a one-day conversion audit (seven fixes, zero guesses)

We woke up to a familiar founder ache: paid clicks arriving, scans completing, zero sales. The lazy diagnosis writes itself — the price must be wrong, maybe try subscriptions? We almost wrote it. Instead we audited the funnel end to end, and what we found should permanently retire "change the price" as anyone's first hypothesis.

Start with the denominator. Thirty days of data held only 13 customer scans — and of those, exactly one person had clicked toward checkout and zero had clicked the pay button. At a healthy 3–5% conversion, 13 prospects predicts zero or one sale. The price hadn't failed; the price had barely been witnessed. Every downstream conclusion changes once you know your sample size is thirteen.

Then look at pixels, not just metrics. The audit turned up seven independent problems in one day. The worst was invisible in every dashboard: our primary sticky checkout button — the one reading "Unlock fix order – $19" — had white text on a white background for about 22 hours, thanks to a CSS selector that colored every span in its bar, including the one inside the button. Every conversion metric from that window was faithfully measuring an invisible button. Two more sat upstream: a category bug confidently mislabeling businesses (the word "menu" in a nav bar made everything food service), and screenshot proof that arrived up to a day after the visitor had already left, because a once-daily worker processed 4 jobs while each scan queued 10.

Then interrogate the traffic. Our ads ran on Google Performance Max with roughly two soft conversions a day of training signal — statistical noise. PMax responded rationally, for a machine: it shifted the entire budget to cheap display inventory. One afternoon: 229 display impressions, 0 clicks, while every customer we'd ever had came from search. We paused it and replaced it with a plain search campaign we could read and steer.

Then check whether anyone could even find you. A per-keyword report delivered the day's cleanest lesson: all 14 of our hand-written "pain point" keywords — phrases like "ads clicks but no calls" — had zero impressions ever. Not low. Zero. Nobody types the sentences marketers compose. Google's own autocomplete showed what people actually search, and it reshaped our keywords, our landing page, and eventually the product's naming.

The remaining fixes rounded out the day: replacing template analysis with an LLM pass so results are site-specific, honest failure states for unreadable sites, and per-minute screenshot processing so proof exists while the visitor is still reading.

Seven problems. Zero of them were the price. The discipline that found them fits in three sentences: When conversion is zero, distrust your funnel before your economics. Look at rendered pixels, not just event counts. And write everything to a log, because the log is what turns a bad day into an audit trail.

Our scan applies the same discipline to your website — evidence first, free, in about a minute: agentsitescan.com/scan.