What Are You Really Paying for When You Hire an AI Expert?

There’s an old story, probably made up, about a repairman who fixed a machine with one tap of a hammer. His bill said two cents for the tapping and a thousand dollars for knowing where to tap. AI can now write a lot of the code. An expert knows which question to start with, which kind of AI fits, where it will break, how to test it, and when to stop. You’re paying for someone who has seen those mistakes before and knows how to steer around them. I get paid for my skills, the same way you get paid for yours.

AI writes code, and the code is the easy part

AI really can write code. Ask a chat tool for a script that reads a spreadsheet and adds up a column, and you will have one in seconds. Often it runs first time. That is useful, and I use these tools in my own work.

Code is rarely where an AI project goes wrong, though. As I explain in why an AI project costs real money, most failures start much earlier, with a question nobody pinned down. A tool that writes code fast will write code for the wrong question just as fast. It will also tell you the job went well.

AI does what you ask, so the asking is where the skill sits. When you hire an expert, you pay for the choices made before any code exists, and for the checks made after it runs.

The repairman story is a parable, and the point holds

The tapping story is a parable, and I want to be honest about that. Quote Investigator, a site that tracks where famous sayings come from, traced it back to 1907. The earliest version it found ran in a Yorkshire newspaper as a fable, with the bill in pounds and shillings. Over the years the story got pinned on famous names. A 1965 letter to LIFE magazine told it about the engineer Charles Steinmetz, who fixed a generator at a Ford plant with one chalk mark. A newspaper column in 1977 told it with the two-cent tap. The researchers suspect almost all of the versions are made up.

The story has lasted over a hundred years because people who do skilled work know it is true in spirit. The tap takes a second, and learning where to tap took years. A customer who only sees the tap will think the job was worth two cents.

I saw this when I built an instant costing engine for metal castings. It priced a casting in moments, but that speed came from years of estimators’ know-how written down as rules. The estimators stayed, because they still knew where to tap.

An expert makes five calls a tool can’t make

Every AI build turns on five decisions. A good expert makes each one with you, out loud, before money is spent.

1. Which question to start with. Pick one question that costs your team hours every week, where a wrong answer is easy to spot. A vague goal like “use AI in sales” gives you a demo, and a demo answers nothing.
2. Whether your data can answer it. The answer has to live somewhere: your ERP (the system that runs orders, stock and jobs), a spreadsheet, or an inbox. Someone has to check that the data is there and good enough.
3. Which kind of AI fits. A chat model, a forecasting model, and plain software with no AI at all are very different tools. Plenty of problems need a simple rule or a report, and an expert will say so.
4. How to test it. Check its answers against ones you already know, and count how often it gets them right. Agree what “good enough” means before the test, so nobody moves the goalposts after.
5. When to stop. If the small version can’t beat the way you do it today, stop there and keep what you learned.

A tool can help with each of these. It can’t make the call for you, because it has never seen your business, your customers, or the mistakes you can and can’t live with. Getting one of these wrong is where the real cost hides: a month spent building the wrong thing, or a system your team stops trusting after the first bad answer.

Knowing where it will break is part of the bill

Every AI system breaks somewhere. It meets a record it has never seen, a word your team uses in its own way, or a supplier who changes their file format. An expert expects this. They build the system to show where each answer came from, so a person can check it, and they plan for what happens when it is wrong.

That planning matters most after launch. Your data changes, your other systems get updated, and the AI models change too. I cover that in what happens to an AI system after it goes live. Someone who has built and run these systems before knows which parts tend to drift, and builds them to be easy to check and fix.

If you’re weighing whether to do all this in-house, the toilet seat story is worth a read first. You can build it yourself. Count the hours your best people would spend learning, and compare.

A real expert tells you no sometimes

The surest sign you’re talking to a real expert is that they will tell you when AI is the wrong answer. Some jobs are better done by a person, a simple rule, or a cleaner spreadsheet. I wrote about how to spot those in when AI is the wrong answer.

A few other signs to look for:

• They show their sources. Every answer the system gives should point to the record it came from.
• They prove it small first. A short test on your real data comes before any big build.
• They agree the price up front. One fee, agreed before work starts, so you are never paying for someone else’s learning by the hour.

I set out how I work on the AI Builds page. Use the same test on anyone else you talk to.

An outside expert is not always the right hire. If your team already builds software and data systems, and they have time to spare, you may not need one. If all you want is help drafting emails, a chat tool on its own is fine. You can read more about my background on the About page.

Start with one question

The best way to judge any expert, me included, is to bring them a real problem and listen to what they ask back. A good one asks about your data, your people, and which mistakes cost you most, long before they talk about models.

Test me the same way. If you’re deciding whether to hire help with AI, bring me a real problem through the AI Builds page. I’ll tell you where I’d start, and whether you need outside help at all.