There’s a scene early in Saving Private Ryan that sets up the rest of the movie.
A woman working in a government office notices something. Three telegrams. Same last name. Same address in Iowa. She wasn’t looking for it. Nobody asked her to look for it. She just saw it, and she stopped.
That moment changed everything that followed. If you haven’t seen the film, I’ll leave the rest for you to discover. But sometimes I think about that powerful scene.
Because what she had isn’t in any job description. It’s the awareness that comes from a human being embedded in a process, paying attention.
When You Automate, You Don’t Just Speed Things Up
The dominant model for AI adoption right now is addition. A chatbot here. An agent handling a narrow process there. AI tools bolted onto existing workflows to reduce friction at specific points.
These are not bad moves. They deliver real wins, and they are a reasonable place to start.
The problem is that workflows were not just designed to move information from A to B. They moved through human hands. And those hands were doing more than the official task required.
People hesitate. They notice when something looks off. They ask a question that wasn’t on the form. They remember the last time a similar situation caused a problem. None of that is documented. None of it appears on a process map.
When you automate the workflow, you automate the handoffs. What you don’t automate - because you can’t see it - is the judgment layer embedded in them. The checkpoint your clerk was running without knowing she was running it.
If you don’t build that back in intentionally, you end up with a faster process and a blind spot you didn’t know you had.
Faster Is Not the Same as Better
There’s a related problem that comes up in almost every automation conversation I have.
Organizations focus on what feels slow. They pick the part of the workflow that seems like a bottleneck and they make it faster. This feels like progress. Sometimes it is. But if you speed up a step that is not actually limiting the whole process, total throughput doesn’t change. You’ve improved one part and left the constraint untouched.
Worse, you’ve now shifted attention away from the real problem.
A marketing team pushes more campaigns, drives more traffic, and generates more leads. The business looks busy. But if inventory planning wasn’t looped in, you get a surge of orders against depleted stock. Frustrated customers. An operational scramble. The opposite of what you were trying to achieve.
Eliyahu Goldratt made this point decades ago with the Theory of Constraints: the system is limited by its weakest link. Making anything else faster doesn’t fix that. It just makes the gap more obvious.
The Worst Version: Doing the Wrong Thing, Faster
The most dangerous version of this is automating a process that shouldn’t exist in its current form.
I’ve seen this more than once. Some companies started rewarding employees for AI token usage. High usage signals adoption, so the logic made sense on paper. The problem is that it disconnected activity from output. Teams generated token volume. Costs went up. Results didn’t move.
A more familiar version of this plays out with Marketing Qualified Leads. Push the MQL metric without a quality floor and your marketing team will deliver exactly what you measured: volume. Your business development team now spends most of its day sifting through low-grade leads looking for the few that are real. Quality candidates get buried. Some get missed entirely.
The system is “working.” It’s just working in entirely the wrong direction. Automation doesn’t fix that problem. It runs the wrong direction faster.
The Question That Changes the Starting Point
The question to ask before any automation project isn’t “can AI do this?” It’s “should this process work this way at all?”
That reframe changes the starting point entirely. Instead of mapping the existing workflow and layering AI on top of it, you step back and ask what the process is actually trying to accomplish. What decisions are being made? Where does judgment come in? What would a capable person notice that a script wouldn’t?
Then you design the automation to handle the routine work. And you build the checkpoints back in on purpose, not as an afterthought.
This is also where change management stops being a soft-skills exercise. When people understand where they still add value, the transition isn’t about replacement - it’s about repositioning. That conversation, done well, is what makes AI adoption stick.
The clerk in the film wasn’t doing her job when she noticed those telegrams. She was doing more than her job. That’s the layer you can’t afford to automate away.
Before You Automate
Every workflow contains informal intelligence that was never written down.
Before you automate, find it. Decide what to keep. Build the checkpoints back in, intentionally and by design.
Speed is not the goal. A process that consistently gets the right outcome is the goal. Those are not always the same thing.