There's a stat that should make every consultant uncomfortable: 72% of companies worldwide now use AI in at least one business function. And yet most of them will tell you their AI projects haven't delivered what they expected.
The tools work. The models are capable. The vendors are selling. So what's going wrong?
After working inside operations at companies ranging from regional logistics carriers to healthcare groups, the answer is almost always the same: they automated a broken process.
The most expensive mistake in AI adoption
When a company decides to "add AI," the natural instinct is to look for tasks that feel manual and repetitive. That's correct. But the next step is where things go wrong.
Instead of stepping back and asking why the task is manual in the first place, most companies just wrap an AI model around the existing process and expect the results to improve.
Sometimes they do — a little. But the fundamental inefficiency remains. You've just made it faster.
Here's a real example. A logistics company was manually processing customer status requests — a team of two people answering the same questions via email all day. So they built a chatbot.
The chatbot could answer common questions. Response time improved. But the system it was pulling data from was still being updated manually by a dispatcher four times a day. So the chatbot was fast — and frequently wrong.
Customers who got a wrong answer called anyway. Now there were two broken processes instead of one.
Fix the system, then add the AI
The right order is: understand the workflow, fix the workflow, then automate the workflow.
Before any AI gets involved, you need to answer these questions:
- Why does this process require human intervention at the frequency it does?
- What information does a person need to complete this task, and where does that information live?
- What are the exceptions — the situations that don't follow the normal pattern?
- What would have to be true for this process to run without a person touching it?
The answers to these questions often reveal that the real problem isn't the task — it's the three systems that don't talk to each other upstream, or the data that gets entered in three different places, or the approval step that exists because someone didn't trust the original process.
Fix those things first. Then the AI has something real to work with.
Why this matters for your business
Every company right now is feeling pressure to "do something with AI." That pressure is real and the window for competitive advantage is genuine.
But moving fast by automating your existing broken processes doesn't create competitive advantage. It locks in your inefficiencies at higher speed.
The companies that will pull away from their competitors over the next two years aren't the ones that deployed AI the fastest. They're the ones that used the AI moment as an excuse to fix the underlying systems they'd been ignoring for years — and then built the AI into the better version.
That's harder. It takes longer to scope. It requires asking uncomfortable questions about why things work the way they do.
It also delivers 10x the ROI.
What to do instead
Before your next AI project, run a simple audit:
- Map the current process end-to-end. Every step. Every person. Every system touched.
- For each step, ask: what's the input, what's the output, and where do errors happen?
- Find the three steps with the most errors or the most human intervention.
- Fix those steps first — with or without AI.
- Now automate.
The AI project you end up building after this process will look completely different from the one you would have built without it. And it will actually work.