Honest limits
Where AI won't help your business
Part of my job as an AI consultant is saying where AI isn't the answer. These are real cases from client scoping, my own sites and tools I use every day. Each one shows what worked better, or where I keep AI on a short leash.
Where AI wasn't the answer
When the spreadsheet already does the maths
An electrical contractor's quoting workbook was mature, formula-driven and version-tracked for years, with its own checks. There was no pricing engine to build. The time went on the work around the sheet, and half of the possible fixes needed no AI at all.
When the spreadsheet just keeps breaking
A holiday-let business's linen spreadsheet could break the moment someone typed over a formula. The fix was the same maths fixed in code, with totals nobody can overwrite. No AI in it, and the weekly prep went from about two hours to about 20 minutes.
When the information already has a shape
In a library of past work, AI earns its place once: reading each document and pulling out the same few fields. After that, finding "everything we've done in hospitality" is a plain filter. No AI needed.
When a simple rule will do
Filing the emails for a long personal case works on an approved list of senders and a subject tag. Anything unclear comes to me. Rules are easier to trust than a model's guess, and easier to fix.
When it's really a short script
On a 400-page directory site I run, an AI tool for internal links looked like the answer until the gap was measured. What it needed was a short script and a few editorial rules.
Where AI stays on a short leash
AI does the reading, checking and dull work. A person keeps the decisions where a mistake would cost something.
Decisions about people
Hiring, credit, eligibility, anything that affects someone's access to a service. I don't build AI for these.
Money
An AI advisor can read the figures and explain them. It doesn't move money or take actions.
Speaking as you
AI can research, check, compare and draft. It doesn't write as you to your customers.
Changing live records
AI can prepare the update. A person approves it before anything changes.
Facts and figures
In a market report I automated, the model added two real numbers into a total that no source ever printed. So there it must quote its sources word for word, and software checks every quote.
Pick the dull option
When I wanted a short video for this site, the obvious choice was an AI avatar. I wrote the video as code instead, so nothing in it was generated and every frame matches the site. The narrower, duller option you fully control is usually the right one. The right kind of AI video is the boring kind.
Questions people ask
- When should a small business not use AI?
- When the job is already handled by a spreadsheet that works, when a plain rule or a short script would do, and when the decision is about people or money. In those places AI adds cost and a new way to go wrong. The Diagnostic says where AI would pay and where it won't, and what to leave alone.
- Is a broken spreadsheet an AI problem?
- Usually not. If people keep breaking the formulas, the fix is to lock the maths away so it can't be overwritten. That is ordinary code. AI helps with the messy work around a spreadsheet, like reading documents or notes, not with arithmetic that already works.
- Will you tell me if AI isn't worth it for my business?
- Yes. Saying where AI won't help is part of every Diagnostic. If the honest answer is a better spreadsheet, a simple rule or leaving things alone, that's what the plan says.
Find out where it would help
The Diagnostic looks at how the work actually gets done and comes back with a short plan: where AI would pay, where it won't, and what to leave alone. A first conversation costs nothing.