What a good AI audit actually delivers
The audit report named one recommendation specific enough to check. What the Build engagement that followed looked like, shown through one real engagement, generalised.
A coach we worked with used to spend the evening before every session with a paper form. One form per client, filled in from memory and old notes: what they’d committed to last time, what had actually happened, where their numbers were tracking, what to raise this time and what to leave alone. Multiply that by a full group of clients, every few weeks, for years. None of it was wasted effort. It was the job. But it was also the exact place an evening disappeared that should have gone to the coaching itself.
That is the pattern most audits are actually looking for. Not a gap in the business. A place where a real skill is being spent on admin instead of on the thing only that person can do. (If you have not read the companion post on why operations-first analysis finds opportunities a technology-first lens misses, start there.)
The audit does not start with a tool
We didn’t open the engagement asking what AI could do. We asked how the practice actually ran: where the hours went, what got written down and where, what was carried in the coach’s head because no system held it, and what happened to that knowledge once a session ended. The paper form turned out to be the whole story. It was a good form. Years of refinement had gone into the fields on it. It just lived on paper, so nothing it captured could be reused, searched, or compared across sessions or across clients. Every insight it held got read once and filed.
That is the finding a technology-first review misses, because it isn’t technology-shaped. Nobody would describe “a well-designed paper form” as an AI problem. It only becomes one once you see what the form is actually for: it is a structured record, sitting one step away from being useful data.
The report has to survive being read alone
The report we wrote didn’t pitch a platform. It named one thing: digitise the form, and use what it captures to do the two hours of preparation the coach was already doing by hand. Specific enough to picture, cost, and check afterwards, not a slide that says “AI could help with client prep.” A recommendation that can’t survive being read on its own, with nobody in the room to talk it up, isn’t a recommendation. It’s a hint.
The report also said what not to build. A client-facing chatbot was on the table early and came off it: the coaching relationship is the product, and nothing about the practice suggested clients wanted a bot standing in for their coach between sessions. Naming the thing that doesn’t belong, before it costs the business money, is worth as much as naming the thing that does, and it’s the same judgement that shows up when the people who’d need to use a tool clearly won’t.
Then a Build engagement tested the report
The report answered what to do. Whether it actually worked only got answered by building it — and that’s a separate engagement, the Build that followed once the coach accepted the report’s recommendation. For this practice, that meant a small working system, not the full practice run on new software overnight. The paper form became a dashboard. The two hours of manual prep became a generated brief, built from the client’s own session history, not a generic coaching template. Patterns that used to live only in the coach’s memory, who’s stalled, whose numbers are drifting, what’s come up in three sessions running, became visible across the whole group at a glance, because the data finally sat somewhere it could be compared.
None of a client’s personal detail ever reaches an AI model in the clear. A coaching record carries exactly the kind of personal information that shouldn’t leave the room ungoverned, so names and identifying detail are stripped before anything is sent, and restored only in the coach’s own browser once the result comes back. That’s not a policy on a page. It’s how the system is built, and it’s the same standard we hold any confidential document work to.
What the coach actually walked away with
Not a faster form. A system that gets more useful the longer it runs, because every session adds to a record the coach now owns, sitting in their own systems, not locked into whichever AI model happened to write it. The model underneath it can change; the system keeps working. That’s what makes it worth calling architecture instead of a tool.
And underneath the time saved, the coach got something they didn’t have before at all: a clear view across the whole group, not just the client in front of them that evening. That’s the part a report alone could never have delivered. It only showed up once the recommendation became something they could actually use.
That’s the whole shape of a good audit: find the place where real skill is being spent on admin, and write down a recommendation specific enough to check on its own. What came next for this coach — a Build engagement that turned the recommendation into a small working system, so the business was deciding on a working thing rather than a promise — is the shape of the Build that can follow, once a client says yes.
Perth AI Consulting delivers AI opportunity analysis for small and medium businesses in Perth. Written report with prioritised recommendations and a clear next step. Start with a conversation.