What happened when I actually hired AI as a new starter
The situation I walked into
A company I was working for had been through an acquisition. Eighteen months on, the way the business said it worked and the way it actually worked had drifted apart. The written procedures, the documents that tell people how to run a project, still pointed at systems and names from before the deal. Nobody had gone through them. There were hundreds.
Reading every one by hand to find what was out of date would have taken a person weeks, and it was the kind of job that never reaches the top of anyone's list. So it sat there, quietly wrong, while people followed instructions that no longer matched the tools on their desks.
What I did
I treated AI the way you would treat a new starter. I gave it a clear job to do, showed it what "good" looked like, and kept it on a short leash until it earned trust.
I pointed it at the full set of procedures and asked one narrow question at a time: where does this still reference the old business, the old systems, the old way of doing things? Then I had it flag each one and say who owned it, so a real person could make the call and fix it. I did not ask it to change anything on its own. It found the problems; people decided.
The first thing I learned was that the tool matters less than the setup. A basic assistant with no memory kept losing the thread and giving shallow answers. The moment I gave it proper context, a defined output, and guardrails on what it could and could not touch, it became genuinely useful. Same technology, completely different result, purely down to how it was hired and managed.
The outcome
Weeks of tedious checking turned into a structured list of exactly what was out of date and who needed to fix it. The drift got surfaced and routed to owners instead of sitting undiscovered. And I came away with something more valuable than the list itself: a clear, tested view of how to put AI to work safely inside a business, and where it falls over.
What it would have cost the usual way
To put a number on it. Reviewing, double-checking and updating hundreds of procedures by hand is weeks of senior people's time. Reckon on a senior manager working through them properly, plus a second pair of eyes to check the changes before anything goes live, and you are into the region of fifty days of senior time. On a loaded cost of around £500 a day, that is roughly £25,000, and the best part of two months before a single procedure is put right.
The way I did it, with AI doing the finding and people making the decisions, it took about a fifth of my time over four weeks. A few days of work, and the drift came out as a structured list ready to action. That is the difference a well-managed AI new hire makes. Not that it replaces the judgement, but that it clears the grunt work that made the job too expensive to ever start.
What it means for you
Most owners I speak to are either nervous about AI or expecting magic from it. Both miss the point. AI is not a genie and it is not a threat. It is a new hire. It needs a job description, clean information to work from, and someone checking its work until it has earned the right to work unsupervised. Hire it that way and it pays for itself. Skip that and you get confident nonsense.
This is exactly how the PeakRatio AI Apprenticeship works: we start it on supervised, low-risk tasks, and promote it only as it proves it can be trusted, on foundations we have got right first.
The Ratio Check is a free, five-minute read on where your business actually stands, and it is the best place to start.