About

A number you can check, or nothing at all.

A homeowner with a burst pipe used to search and scroll. Now she asks an assistant and calls whoever it names. That list of names is the whole market for a local trade, and almost nobody who depends on it can see it.

That is what guidepost reads out loud. It asks the questions your customers ask, records what comes back word for word, and shows which pages the assistant was reading when it decided. Nothing about your business, your books, or your software is involved. It reads the same public answers any homeowner sees.

Why the restraint matters

I built this. Data science mostly taught me when a number proves something and when it only hints. That distinction is the product. Every AI tool will happily hand you a confident answer. The hard part is knowing which answers deserve confidence.

So there are rules:

• Every number shows its work. The figure, the count it rests on, and the questions that produced it. A report you cannot check is a report you should not trust.
• Answers move, so the spread gets reported. Ask an assistant the same question three times and you can get three different lists. Anyone selling you a single confident score is hiding that from you.
• No promises about outcomes. Nobody controls what these models say. guidepost can tell you which sources they read, which is the part you can actually change.
• When the data is thin, it says so. Not sure yet is a real answer and it gets used.

Where it is at

Early. Readings are run by hand right now, one company at a time, which is deliberate: it is the only way to find out what the report needs to say before building the machine that says it. If you want one, request a reading. You'll hear from our founder. Nothing starts until he does.