GBP Audit Engine — what a profile audit can honestly claim
A Google Business Profile audit is an easy thing to sell and a hard thing to be honest about. You tell a local business how its Google listing compares to its rivals, and every line in that report has to survive the owner checking it. So before building the product I built the measurement: a small data puller, a spike that ran it end to end over twenty real UK hair salons across eight towns, and four hand-written customer-facing reports to find out what the data would actually support in front of a reader.
What the measurement said
Cost and speed came out fine. Two API calls per audit, no model spend at all, about five pence an audit at list price and free below a thousand a month. Median wall clock was 1.4 seconds, with the worst of the twenty at 1.76 seconds, which fits inside the budget for rendering a page while someone waits.
The accuracy did not come out fine. Four of the twenty lookups failed, and three of those four failed silently, returning a confident, error-free result about a different company. A salon I searched for in Chester came back as a day spa in Maryland. Another came back as a crazy-golf venue in the right town with a similar name. The most dangerous one returned the sibling beauty listing at the unit next door, same brand, same street, so an owner skimming the report would see their own name on it and only notice further down that the whole comparison was against nail bars. Drop the town from the query and all five deliberately ambiguous names failed.
The mechanism matters more than the rate. The search endpoint is a ranked search, not a lookup. When it cannot match a name it does not fail, it returns the nearest plausible business in the right category and the right town, with no confidence score and no error. Nothing in the response shape distinguishes a correct audit from a confident audit about a rival down the road.
The claims I had to drop
The reports were written by hand precisely so I would feel where the evidence ran out, and the note recording that is more valuable than the reports.
The single most sellable line available is "you need N more five-star reviews to reach 4.5", and it cannot be written. Google returns a rating already rounded to one decimal and never returns the distribution underneath it, so the arithmetic would look rigorous and rest on a guess. Photo freshness, the most natural finding in this category, does not exist either: photo objects carry no date field of any kind. Photo counts cap at ten, and two-thirds of the listings I pulled returned exactly ten, so comparing photo counts is mostly comparing two ceilings. Nearly half the competitors Google returned as rivals were the wrong shape anyway, including chair-renters at the subject's own address and, in one set, the subject's own sibling business.
What survives is seven signals comparable on both sides in most audits. Every finding in those four reports therefore ends in an observation rather than an outcome, because there is no traffic or conversion data anywhere in the pipeline to license a consequence clause. That is the biggest felt difference between these and a typical agency audit, and it is the whole reason to measure first. The version of this product that would have shipped without the spike is the one that confidently tells a stranger something about their own business that they can disprove in thirty seconds.