Why I'm writing this
Every month I get calls from companies that have already tried to bring in AI and it went wrong. They tell it with a mix of frustration and embarrassment, as if it were their fault. It isn't. Nobody explained the mistakes to them before they made them.
These are the eight I see most, with what they cost and how to avoid them. If you're going to invest in AI this year, read them first.
Mistake 1: buying the tool before defining the problem
What it looks like: "We've bought AI licenses for the whole team." Three months later, five people use it to draft emails and nobody knows what has changed.
What it costs: the licenses, which isn't the serious part. The serious part is the conclusion: "AI doesn't work for us". And with that conclusion the door closes on what would have worked.
How to avoid it: start with a sentence that begins "we lose time/money/customers when…". That sentence defines the problem. The tool comes later, and often it's not the one you thought.
Mistake 2: automating a broken process
What it looks like: quotes take four days because they pass through three people who don't coordinate. AI is brought in to draft them faster. Now they take three and a half days.
What it costs: the whole project, because the bottleneck wasn't in the drafting.
How to avoid it: draw the full process before touching it. Remove steps. Simplify. Then automate what's left. In many cases, simplifying solves half the problem without technology.
Mistake 3: letting AI decide about money without supervision
What it looks like: a system applies "smart" discounts automatically. Or approves supplier payments. Or sends invoices without review.
What it costs: the day it gets it wrong, it gets it wrong in real money. I've seen duplicate invoices for thousands of euros and discounts applied to customers who shouldn't have had them.
How to avoid it: a fixed rule I apply in every project: AI proposes, a person approves. In everything touching money, always. One click is enough, but a person makes the click.
Mistake 4: not counting the cost of maintaining
What it looks like: the implementation is budgeted and the rest is forgotten. Six months later, a supplier changes their API, a document format is different, a case appears nobody anticipated. Nobody knows how to fix it. The automation stops and everyone goes back to manual.
What it costs: the entire initial investment, plus the rollback.
How to avoid it: set aside 10 to 15 percent of the implementation cost per year for maintenance, or get a support plan that includes it. An automation without maintenance is an automation with an expiry date.
Mandatory question for any AI provider: "What happens when this stops working and who fixes it?". If they don't have a clear answer, don't sign.
Mistake 5: ignoring the team until launch day
What it looks like: management decides, the provider builds, and the team finds out when it's done. The tool doesn't fit how they work. They use it halfway or not at all.
What it costs: adoption, which is where everything is won or lost. A tool nobody uses has a return of zero.
How to avoid it: involve two or three people from the team from the diagnosis stage. Their specific complaints are the best specification. And their support on launch day is worth more than any order. I wrote about this in how to roll out a new tool without your team hating it.
Mistake 6: feeding AI with dirty data
What it looks like: an assistant is connected to the product catalog, but the catalog has prices from two years ago, discontinued products and incomplete descriptions. The assistant answers with that information. Confidently. To real customers.
What it costs: credibility. A customer who gets a wrong price from your "intelligent assistant" doesn't trust it again.
How to avoid it: before connecting anything, clean the source. It's boring, but it's half the project. And assign someone responsible for keeping it up to date.
Mistake 7: not measuring anything
What it looks like: it's implemented, launched, and three months later nobody knows whether it worked. "It seems to be going well." "I think it saves time." No numbers.
What it costs: the ability to decide. Without data, you don't know whether to expand, adjust or cancel. And without knowing, the most comfortable decision gets made, which is usually to do nothing.
How to avoid it: before starting, write down the three figures you want to move (response time, closed quotes, admin hours, whatever) and their current value. At four weeks, compare. Everything gets decided from that comparison. On how to calculate the return I wrote a full guide: how to measure the ROI of automation.
Mistake 8: trying to do everything at once
What it looks like: customer service, invoicing, marketing, reporting and sales, all automated in the same quarter. Five open projects, none closed, the team overwhelmed.
What it costs: all the projects. When five things are half done, none works well, and the sense of chaos makes everyone abandon everything.
How to avoid it: one process. The one that hurts most. Until it runs a month without incidents. Then the next. It's slower on paper and much faster in reality.
A bonus mistake: choosing on price
I see this especially in companies that come to me after a bad experience. They hired the cheapest provider, or a nephew who "knows about this", and now they have something that half works and nobody can fix.
Badly implemented AI isn't cheaper. It's more expensive, because you pay for it twice: the first time and the time to fix it.
The right way, summarized
- Define the problem in one sentence with a cost.
- Simplify the process before automating it.
- Start with a single case, the one that hurts most.
- Clean the data you're going to use.
- Involve the team from the start.
- Human supervision on everything touching money or customers.
- Measure before and after.
- Budget for maintenance.
It's not complicated. It just takes discipline and someone who has been through this before.
If you've already made one
That's fine. Most of my monthly partnership clients come after a failed attempt. What we do is simple: review what was done, rescue what works, fix what doesn't, and this time do it with method.
If you want a second opinion on what you have in place, or want to avoid these mistakes before starting, book a call. Fifteen minutes, no commitment, and you leave knowing where you stand.
