What AI automation actually costs a small business

Nobody publishes prices, so let us be more useful than average about the shape of the spend — even where the exact number depends on your business.

The three costs people quote for

The audit. A fixed fee for mapping your workflows and telling you what is worth doing. This should be a flat number, agreed up front, and the output should be yours whether or not you continue.

The build. Scoped from the audit. It varies enormously, and anyone who quotes a build before understanding your process is guessing.

Running it. Monthly. Covers monitoring, fixes, and the API costs of any AI models involved.

The costs nobody quotes for

Your team's time during discovery. Someone has to explain how the work actually happens. Budget several hours per process, from the person who really does it — not their manager's description of it.

The cleanup you will discover. Automation exposes the mess it runs on. Duplicate records, three spellings of the same supplier, a spreadsheet that one person maintains by hand. This is genuinely useful to find, and it is work.

Change, not software. The system will be ignored unless someone owns making people use it. This is the single most common reason automation fails, and it costs time rather than money.

How to sanity-check any quote

Ask what happens if it stops working at 2am on a Sunday. If the answer is vague, you are buying a prototype. Ask who owns the account and the credentials — the answer should be you. Ask what it costs to leave.

The honest arithmetic

Take the people doing the repetitive work, times their hours per week, times a rough hourly cost, times about 48 working weeks. That is your annual bill for doing it by hand. Somewhere between 45% and 70% of genuinely repetitive, rules-based work can come off that number. Judgement work will not.

If the recoverable slice does not comfortably exceed the build cost inside a year, the honest answer is that you should not automate that process yet.