Do I need to sort out my spreadsheets and data before I can use AI?

You are not alone in this. Half the business lives in your head, the other half lives in a dozen spreadsheets with names like "Customers FINAL v3", and the assumption is that AI needs all of it clean and organised before it can do anything useful. So the whole thing gets parked, indefinitely.

That assumption is wrong, and it is worth being precise about why.

How clean does the data really need to be?

AI does not need your business tidy. It needs the inputs for one specific job to be reliable. Those are very different sizes of problem.

If you want AI to chase unpaid invoices, it needs your invoice list: who owes what, since when, and a contact email. It does not care that your customer list has duplicates, that your job log from 2021 is a mess, or that your pricing logic lives entirely in your head. None of that touches invoice chasing, so none of it needs fixing first.

The work is scoped to the process, not the business. Pick the job you want off your plate, look at the one or two spreadsheets that feed it, and tidy those: consistent column headings, no duplicate rows, no blanks in the fields that matter. For a typical small business spreadsheet that is an afternoon, not a project.

What about the stuff that only exists in my head?

This is the part people worry about most, and it is the easiest to deal with. The rules you carry around ("this client always pays late but always pays", "never quote that job without checking access first") get written down as part of building the system, not before it. A good builder asks you questions, turns your judgement into explicit rules, and shows you the result so you can correct it. You do not need to document your business before you start. Documenting it is what starting looks like.

Waiting until your data is tidy is the most respectable way of never starting. I would rather take your worst spreadsheet and get one process running on it than watch you spend another year planning the clean-up.

Where people genuinely come unstuck

Two traps worth naming.

First, the grand clean-up. Someone decides to sort the data across the whole business before automating anything. That project has no finish line, produces nothing anyone can use, and dies quietly around week six. Clean data is a by-product of a working system, not a prerequisite for one. Once a process runs through a system, the system forces structure on the data: every new record arrives in the right shape because there is no other way in.

Second, the opposite failure: automating on top of inputs that are wrong for the job in hand. If your invoice list has out-of-date email addresses, AI will chase the wrong people, faster. Messy is fine. Wrong is not. The check is narrow: are the specific fields this one process reads accurate? You verify that before switching anything on, and it takes an hour, not a month.

So no, do not spend three months organising spreadsheets. Spend a week getting one process live, let the system tidy the data it touches, and move to the next. If you are weighing up which process to pick first, our guide on when your business has outgrown manual processes walks through the symptoms and the usual first moves.