Method
Six steps, and two of them are the reason to trust the rest.
Most things that grade a business run once, produce a number, and cannot tell you where the number came from. This keeps the evidence and counts in code, so every finding points back at something you can open.
They asked for my card just to hold a table, and nobody could tell me what it was for.
Tripadvisor · one of three reviews saying it · 14 August
A quote is printed verbatim or it is not printed. Under 12 characters it is dropped rather than tidied up, because a paraphrase you cannot check is an opinion with quotation marks around it. This one is Marisol’s, the invented business on the sample.
Two views of the same business
Well-run businesses develop blind spots for a structural reason rather than a careless one: management sees the business from the inside, and customers only ever see it from the outside. Neither view is wrong. Only one of them is what your next customer will meet.
From the inside
What the business can see
- What you intended to deliver
- What the process says should happen
- What staff are trained to do
- What the customers who complained to you said
- What you would notice if you walked in as yourself
From the outside
What customers describe
- What actually happened, on the day
- Which step they gave up at
- What they expected, and where that came from
- What the ones who never complained decided instead
- What a stranger concludes in the first ten seconds
The method Six steps
Research
We identify the public sources where this particular business is actually discussed — which differ by trade far more than most people expect — and read them. Its own site as a first-time customer would meet it, its listings, its reviews, the directories and industry platforms its buyers use. Nothing behind a login, no form ever submitted, and robots.txt obeyed.
Record
Every page read is kept as a dated snapshot with a content hash. That is what allows a finding written four months ago to still point at the thing it was drawn from — and it is the difference between a conclusion you can interrogate and one you have to take on faith.
Identify
Recurring complaints, inconsistencies between what is promised and what is described, and signals that appear in more than one place. This is the step where a language model earns its place: reading several hundred pieces of customer text is work no owner has time for and a model does well.
Validate
Repeated evidence is separated from isolated comment. One articulate complaint is an anecdote. The same theme raised independently on unrelated platforms is a property of the business, because independent sources cannot coordinate and a pile-on can.
Quantify
How often, across how many sources, and whether it is becoming more or less prominent. All of this is arithmetic performed in our own code over what was extracted. A model is never asked how often something happens — it would answer, and the answer would not be a count.
Explain and recommend
Customer language is translated into an operational problem, with what the evidence suggests it may be costing and what we would investigate first. Stated as findings that deserve attention, not as certainties the evidence cannot carry.
The model is given the measurements and cannot add one. Everything counted, timed or quoted came from the evidence.
How often a thing is said is arithmetic · and the arithmetic is ours
What we refuse to tell you
Most of this category tries to sound omniscient. The list below is the opposite, and it is the part worth reading twice — a firm that will not say where its evidence stops has said something about the rest of its findings.
We do not score what we could not read
If a site is behind a bot challenge, or is assembled in the browser in a way our crawler cannot see, you are told that — not given a low score. A grade built on a page nobody read is worse than no grade.
We do not assert anything we cannot source
Every material finding names a page, a review or a listing, and the date it was seen. If you disagree with one, you can open the same thing we opened. That is the difference between a finding and an opinion.
We do not print a number nobody measured
How often, across how many sources, and whether it is worsening — all of it is arithmetic performed in our own code. The model that writes the opening paragraph is given the measurements and cannot add one.
We do not let the model decide what is serious
AI reads and extracts, at a volume no person has time for. Frequency, corroboration and severity are computed afterwards. A model asked how often something happens will answer, and the answer will not be a count — which is the single rule that makes this worth arguing with.
We do not treat a pile-on as a pattern
Three complaints in one place can be a bad fortnight, a competitor, or one person with several accounts. The same theme raised independently on unrelated platforms is a fact about the business. Only the second one moves a finding up.
We do not use this as a route to something else
No upsell, no audit call, no agency waiting at the end of it. The research is the work, and the work is the product.
We do not go inside your business. We do not log in as you, connect to your booking system, or ask for access to anything. Everything in a report is something any customer could have seen — which is the point. If a problem is only visible from inside your business, your customers are not hitting it.
Not every customer writes something. The public record is a sample. A business with eleven reviews does not have a measurable reputation problem, and if that is what we find, that is what the issue says.
Some problems are invisible from outside. If a failure never reaches a customer, no amount of public reading will find it. Read our findings alongside what you already know internally, not instead of it.
Commercial impact is an estimate. It is reasoned from observed frequency and stated as an estimate every time. It is not accounting, and we will not dress it up as one.
We do not resell any of it. The underlying data is not for sale, and one client's evidence is never used to write another's.
See what six steps would find in your business
The first report is researched and written the day you start.
Half of an issue is read from a site of your own. If your business does not have one, leave this empty — you get the half read from your listings and reviews, and here is what that leaves out.