Where do I even start with AI in my business?

Everyone tells you that you should be using AI. Almost nobody tells you what for. So Monday morning comes round, you open ChatGPT, type something vague, get something vague back, and close the tab feeling like you missed a meeting the rest of the economy attended.

You did not miss anything. When the Department for Science, Innovation and Technology surveyed 3,500 UK businesses, the most cited barrier to adopting AI was not cost, regulation or risk. 71% of businesses said they had not identified a use for AI, and 60% said they had limited AI skills or knowledge in-house. Not knowing where to start is the national default, not a personal failing.

The fix is to flip the question. Stop asking "how do I use AI?" and start asking "which job do I want rid of?"

What does the first step look like in practice?

Write down the tasks in your business that are repetitive, follow a pattern, and quietly eat hours. Chasing unpaid invoices. Answering the same five enquiry emails. Turning site notes into quotes. Writing up meetings. Copying details from forms into a spreadsheet.

Pick one using three tests:

• It happens at least weekly.
• It follows rules you could explain to a temp in ten minutes.
• You would be mildly embarrassed to admit how long it takes.

If a task passes all three, AI can almost certainly do most of it today.

Then fix that one job in the smallest way that works. For most businesses the first version is not software, it is a reusable prompt. Paste three examples of the task done well into ChatGPT or Claude, describe the rules you follow, and ask it to do the next one. Review the output, correct it, save the prompt that worked. That is the whole first step. No platform, no pilot programme, no strategy deck.

Where do people go wrong?

Three traps catch nearly everyone.

Buying tools before naming a problem. Subscriptions feel like progress and mostly are not. If your first AI conversation is about which tool to buy rather than which task to kill, you are in the wrong conversation.

Going too big. "Use AI across customer service" stalls in committee. "Draft replies for the info@ inbox" can be working by Friday, and a working thing is the best argument for doing the next one.

Waiting until you understand AI properly. You learn what AI is good at in your business by giving it a job and marking its homework, not by reading another report. The understanding gap closes through use, nowhere else.

What happens after the first job?

Repeat the loop. Once one task is handled you have real evidence: what AI does well in your business, where it fails, how much checking it needs. That makes the second pick obvious and the third easy. Businesses that look "good at AI" are not smarter, they are just further round the same loop.

If you are not sure your business has reached the point where any of this matters, From spreadsheet to system walks through the symptoms that a manual process is quietly costing you money, and which ones to fix first.