What an AI audit should actually tell you
A ranked list of AI opportunities is easy to agree with and impossible to act on.
That is the output of most AI audits sold to firms of twenty to a hundred people. Twelve to twenty ideas, sorted by somebody's judgement into high, medium and low, benchmarked against an industry average. Everyone in the room nods. Nothing happens, because nothing in the document tells you what any of it is worth to you specifically, and nobody signs off a six-figure change programme against a colour-coded grid.
The one question the document has to answer
What does this specific piece of work cost us every month, in our own numbers?
Not the industry average. Not a percentage improvement. A number from your payroll, attached to one named process, that the person who owns that process would recognise as roughly true.
Once that number exists, the decision stops being about AI. It becomes an ordinary business decision: this costs us that much a month, fixing it costs this much once, the payback is that many months. Firm owners make that decision every week about vehicles, software and hires. They are not slow at it. They are only slow when the number is missing.
Why so many audits skip it
Because producing it is the hard part, and the list is the easy part.
Getting to a real monthly cost means sitting with the people doing the work rather than the people who own it, asking how long things actually take rather than how long they are supposed to take, and then putting those hours against what those people actually cost. It is unglamorous, it takes days rather than hours, and it produces one paragraph rather than twenty slides.
A list of opportunities can be assembled from a survey and a template. It looks like more work and is much less.
Four things to ask for before you sign
Ask who they talk to. If the answer is the leadership team, the audit will describe the business as leadership believes it runs. The gap between that and how it runs is usually where the money is. Somebody has to talk to the people doing the work.
Ask what the deliverable names. One process, or a list? A list means the prioritisation problem has been handed back to you, which is the problem you were paying to have solved.
Ask where the numbers come from. Your payroll and your timesheets, or a benchmark report? A benchmark tells you what firms like yours supposedly spend. It cannot tell you what your quoting process costs, because it has never seen your quoting process.
Ask what happens if you stop there. If the roadmap is only useful when you also buy the build, it is a sales document. The test of a real audit is whether it is worth something to you when you take it to somebody else, or do it yourself.
What it should cost, roughly
An audit of this kind sits somewhere between S$2,500 and S$8,000 for a firm of this size. Below that, nobody is spending days inside the business. Well above it and you are buying a strategy deck, which is a different product that has its own uses and is not this one.
It should also be credited against whatever you buy next. An audit that is priced as a profit centre has an incentive to recommend a large project, and you can feel that incentive in the recommendations.
The uncomfortable version
Sometimes the honest output is that the expensive work is not the work anyone complains about, and the thing everyone wanted fixed is not worth fixing. That is a good audit. It saves you from spending three months on the loudest problem instead of the costliest one.
We ran one at a thirty-person agency where revenue was growing and margin was not. Everybody had a theory about where the hours went. The survey went to all thirty-one staff, we interviewed five departments, and put the answers against their own salary costs. The expensive work turned out to be something nobody had mentioned.
That is the point of measuring rather than asking.