Writing
Five Things Owners Get Wrong About AI
Every one of these has cost a real company real money. None of them are about the technology being bad.
Published · Updated · 6 min read
The short version
The five expensive beliefs: that the tool is the decision, that you need to start small to be safe, that your business is too specific, that a pilot proves anything, and that adoption takes care of itself.
None of these are about AI being oversold. They are about where the money actually goes wrong, which is almost always upstream of anything technical.
1. That choosing the tool is the important decision
Owners spend weeks comparing platforms and about forty minutes deciding what to automate. That is exactly backwards.
Any competent tool can build the thing. Whether the thing is attached to your largest cost is what decides the return, and no comparison table answers that.
2. That starting small is the safe option
Starting small sounds prudent and usually means picking something low-stakes, which by definition means picking something that was not costing you much.
Six weeks later there is a working system, a real invoice, and no visible change to the business. Then the conclusion is that AI does not work here.
Small in scope is right. Small in importance is not. Pick a narrow slice of the expensive operation rather than the whole of a cheap one.
3. That your business is too specific for this
Every owner believes this and it is the least true of the five. Specificity is the asset here, not the obstacle.
A generic model knows nothing about your pricing. Trained on your quotes, your replies, and your rules, specificity is exactly what makes it useful. The companies that struggle are the ones with no history to train on, not the ones with unusual businesses.
4. That a pilot proves something
A pilot proves the technology can do the task, which was rarely in doubt. It does not prove your team will use it, which is the only question that matters.
A system that works in a demo and is ignored on a Thursday afternoon has failed, and it fails the same way whether the pilot went well or badly.
5. That adoption takes care of itself
This is the one that kills the most projects. The build finishes, a video gets recorded, and everyone goes back to the old way because the old way is known and the new way is not yet trusted.
Trust is built by sitting with the person who will run it, watching them use it on real work, and fixing what makes them hesitate. That is not a nice extra at the end. It is the difference between a system that is alive in a year and one that is not.
If a provider cannot name the person on your team who will be trained, adoption is not in the plan, whatever the proposal says.
The pattern underneath all five
Every one of them is a way of avoiding the uncomfortable part: looking honestly at where the money goes and admitting that the expensive operation is the one nobody wants to touch, because it is held together by two long-serving people and a spreadsheet.
That is the one worth replacing. It is also the reason the assessment comes before any building.
Questions
Is it true that most AI projects fail?
Most fail at adoption rather than at build. The system usually works; the team keeps working around it because nobody was trained and nothing was made mandatory.
Should we wait until the technology settles?
The models change quickly, but the work of identifying an expensive operation and preparing your history to train on does not. That work holds its value regardless of which model is current.
What is the safest first project?
A narrow slice of an expensive operation, with a number you already track. Same-day invoicing and speed-to-lead both qualify in most companies.
Read next
If you are comparing options
Find out what your most expensive operation is
The assessment covers the whole company: how you get customers, how you deliver, and how the place runs. You get it in writing, and you can build from it with or without us.