A client comes in and says: "We want to optimize this process with AI." I ask: "How does this process work today?" Silence. Then someone says: "Well... it depends. Depends who's doing it. Depends on the day."
That's the moment I know AI should be the last thing we discuss.
Processes That "Somehow Work"
There's a pattern I see repeatedly in companies. A process was designed at some point — maybe 5 years ago, maybe 10. Since then, the people have changed, the systems have changed, the business requirements have changed. But nobody changed the process. People developed workarounds, informal paths, personal solutions. Everything "somehow works" because people compensate for the process's shortcomings with their experience and goodwill.
And then someone says: "let's add AI to optimize this."
AI Is an Accelerator, Not a Fixer
This is a fundamental misunderstanding I see in every other project. AI doesn't fix processes. AI accelerates what already exists. If what exists is a well-functioning, repeatable, documented process — AI can significantly improve it. If what exists is chaos — AI accelerates chaos.
Imagine a production line where every third product is defective because the machine is poorly calibrated. Adding a faster conveyor belt won't solve the defective product problem. It will simply produce more defective products in less time.
That's exactly how AI applied to a broken process works. It generates incorrect results faster. It replicates bad decisions faster. It escalates problems that people used to patch manually — faster.
Where This Temptation Comes From
Companies reach for AI as a solution to process problems for several reasons:
It's easier to buy technology than to change the organization. Deploying an AI tool is a project with a budget, a timeline, and a vendor. Changing a process means working with people, habits, organizational politics, resistance. One is "sexy" and conference-worthy. The other is tedious and thankless.
Market pressure. The board hears that competitors are "deploying AI." Can't fall behind. Nobody mentions that competitors are probably making the same mistakes.
Vendor promises. Tech companies sell a vision: "our AI will optimize your processes." They don't say: "but first you need to have processes worth optimizing." Because that would complicate the sale.
What Actually Happens
I've seen this many times. A company deploys AI for its customer service process. The customer service process goes like this: customer calls, reaches a consultant, consultant searches for information in three systems, doesn't find it, asks a colleague, colleague doesn't know, customer waits. AI was supposed to "optimize" this.
What happened after deployment? AI searched the three systems faster. It still didn't find the information because the data was incomplete (sound familiar?). The customer still waited — except now they were dealing with a bot that elegantly informed them it was "processing the request." Frustration increased because expectations were higher.
The problem wasn't search speed. The problem was that information was scattered, incomplete, and nobody knew who was responsible for it. That's a process and organizational problem, not a technological one.
Fix First, Then Automate
The principle is simple but uncomfortable: before you deploy AI to a process, make sure that process works without AI.
What does that mean in practice?
-
Map the process as it actually looks — not as drawn on a 2019 diagram. Talk to the people who work in it. Ask about workarounds, informal paths, moments when "you have to call Kasia because only she knows."
-
Identify what's broken — where the process stalls, where people waste time, where errors occur. This doesn't require AI. It requires a sheet of paper and an honest conversation.
-
Fix what can be fixed without technology — simplify steps, eliminate unnecessary handoffs, establish clear responsibilities, standardize data. It can be tedious, but it's the foundation.
-
Only then consider AI — as a way to accelerate and scale a process that already works. A process audit for AI readiness will show which elements of the fixed process will genuinely benefit from automation.
Does This Mean AI Is Useless?
Absolutely not. AI is extremely useful — where it has something to work with. Repeatable, well-defined tasks with clear input data and expected outputs — that's the terrain where AI shines.
But AI is not a magic wand that turns mess into order. It turns mess into faster mess. And that's worse than the starting point, because now you have the mess plus the AI budget on your back and a disillusioned board that has "lost faith in AI."
The Question Worth Starting With
Instead of asking "how can AI optimize our process?" start with: "does our process work when everything goes well? Do people know what to do at every step? Is the data where it should be?"
If the answer is "yes" — great, let's talk about AI. If the answer is "well... it depends" — let's talk about the process. Because fixing the process is cheaper than deploying AI to a broken process and then fixing both at once.
If you want an honest assessment of which processes in your company are ready for AI and which need fixing first, I invite you to a conversation — Leszek Giza.