How to run an AI audit on your own business (the framework we use)
You don't need a consultancy to start. This is the framework we use on client engagements, written so you can run a version of it yourself in a week.
An AI audit is not a technology exercise. It's an accounting exercise that happens to end in software. Here's the version you can run yourself.
Step 1: Follow five pieces of real work
Pick five recent, ordinary items: an order, an enquiry, an invoice, a new client, a complaint. For each, walk its actual path: who touched it, what they did, how long each step took, and how long it waited between steps. The waiting usually dwarfs the doing. Write it down exactly as it happened, not as the process document says it should.
Step 2: Find the repeated, rule-bearing work
Go through your trace and mark every step that is (a) done more than ten times a week, (b) governed by rules a competent new starter could learn in a day, and (c) currently done by a person. Copying data between systems, checking documents, drafting routine messages, compiling reports, chasing updates. This is your automation candidate list.
Step 3: Price each candidate
For each: hours per week × fully-loaded hourly cost × 52. Add the error cost where relevant: a mis-keyed invoice or a missed follow-up has a price too. Be conservative; the numbers are usually startling enough without optimism. A task taking two people three hours a week at £30 an hour is £9,360 a year, before errors.
Step 4: Score buildability
Rate each candidate 1 to 5 on: rule clarity (could you write the rules down?), data access (is the information in systems, or in heads?), and failure tolerance (what happens if it's wrong: is there a natural checkpoint for a human?). High rules, accessible data, safe failure = build. Anything needing judgement at every step stays human.
Step 5: Rank and be honest
Plot value against buildability. The top-right corner is your build list, in order. Then apply the hard test to each: if you can't name the number this automation exists to move, don't build it.
Where does a DIY AI audit fall short?
This framework finds the obvious candidates. What it can't easily do is spot cross-system opportunities (where the win spans three tools), price the build side accurately, or design the human-approval points. That's the gap a professional audit closes, and it's also where a genuinely useful audit differs from a slide deck: see what an AI audit should give you for the deliverable to insist on. But even if you never hire anyone, steps 1 to 5 will change how you see your own operation.
How does M22 help with this?
This is the framework M22 Consultancy runs professionally, under the method Understand, Quantify, Build, Improve, with the extra steps a DIY pass cannot easily do: spotting opportunities that span several systems, pricing the build side accurately, and designing where a human should approve. An M22 audit follows your real work, prices what it costs, and ranks opportunities by expected return and delivery cost, leaving you with a process map and a costed build plan you keep whether or not M22 builds anything afterwards. M22 will not recommend a build until the numbers prove it is worth building. The AI Audit starts at £1,500, a fixed fee agreed before day one. Book a thirty-minute call, or see what a UK AI audit costs.