What to Fix Before You Add AI to Your Business

Bizer · 2026-10-01

Before you add AI to your business, write down the problem you want it to solve in one sentence, with a number in it. "We take two days to answer quote requests and lose about a third of them." "I spend six hours every Sunday on invoices." If you cannot write that sentence, you are not ready to buy anything, and AI will not tell you what the sentence is.

This is the second of four pieces on bringing AI into a small business. The first is on finding your first use case. The next two cover organizing your data and choosing a tool.

Why does the problem come before the tool?

Because the tool will always look like an answer. Every AI product demo is a solution in search of your problem, and it is built to look useful to everyone. Without a named problem, you buy on the demo, use it for three weeks, and quietly stop. The money is the small loss. The bigger one is that your team decides "AI doesn't work for us" from an experiment that was never set up to answer anything.

A named problem with a number does three jobs. It tells you what to look for, it tells a vendor what to show you, and it tells you a month later whether it worked.

How do I find the real problem?

Ask where time, money or customers are leaking, then keep asking why until you hit something concrete.

Take "marketing isn't working". Why? Leads come in but few book. Why? We answer slowly. Why? Quote requests land in three inboxes and nobody owns them. Now you have a real problem, and it may not need AI at all. It may need one inbox and one person responsible. That is a cheaper fix and you should take it.

Some prompts that help:

• What task makes people groan on Monday morning?
• Where do customers wait on us?
• What do we do by copying and pasting between two programs?
• What do we redo because it was wrong the first time?
• What do I, the owner, do that someone else could do with clear instructions?

Which problems is AI suited to?

As of 2026, general AI tools are good at work that is mostly language: drafting, summarizing, sorting, classifying and answering from a body of text you give them. That covers a lot of small business friction, like slow replies, repetitive writing, long documents nobody reads, and notes that never get cleaned up.

They are a poor fit, on their own, for:

• Problems that are really about people or process. If nobody owns the quote inbox, an AI drafting replies into that inbox still has nobody to send them.
• Problems that need exact numbers without checking. AI tools can produce figures that look right and are not. Your books, payroll and taxes need software built for them and a person who checks.
• Problems that are really about demand. If the phone does not ring, faster replies will not fix it.

Be honest about which kind you have. A lot of "we need AI" turns out to be "we need a checklist".

How do I write the problem down?

Use this shape, and keep it to one page:

1. The problem in one sentence, with a number. "Quote replies take two days; we win about 30% of quotes."
2. Who it affects. The office manager, the customers, the owner's evenings.
3. What it costs, roughly. Hours a week, or jobs lost a month times your average job.
4. What good looks like. "Same day replies; win 40% of quotes."
5. How you will measure it. Where the number comes from, and who checks it monthly.

That page is the most useful thing you can bring to any vendor conversation. A good vendor will engage with it. One who skips past it to the demo is telling you something.

What does this cost me?

An afternoon, and some humility. Writing down the problem often shows that the fix is a process change, not a purchase. That can feel like a step backward if you were excited to try AI. It is not: you saved the money, and you now know which problem to point AI at when the fit is right.

What is uncertain

What AI tools can reliably do is changing quickly, and a problem that is a poor fit in October 2026 may be a good one in a year. The method does not depend on that. A named problem with a number in it is how you judge any tool, whether it uses AI or not. What we cannot tell you is what your number should be. Only your own records can.