AI won’t save a broken operating model
The AI demo never shows you the broken process underneath.
Point AI at a process and it’ll run it faster. That’s the promise, and it’s real. Here’s the catch nobody puts on a slide: if the process is broken, AI doesn’t fix it. It industrialises it. You don’t get a better way of working. You get the wrong way of working, automated, at scale, and harder to see.
Slow and painful is often a symptom
This is the trap underneath most AI business cases. The instinct is to find the slow, manual, painful processes and automate them. But “slow and painful” is often a symptom, not the problem. The process is slow because it has six approval steps that exist for reasons nobody remembers. It’s manual because two systems were never properly joined. Automate that and you’ve just spent money making a bad design permanent, and faster.
The dysfunction goes quiet
Worse, you’ve hidden it. A clunky manual process at least announces its own dysfunction; people complain, workarounds surface, the pain is visible. Wrap automation around it and the dysfunction goes quiet. The work still happens the wrong way. It just happens invisibly now, which means it stops getting fixed.
An operating-model decision
The uncomfortable reframe is this: AI in HR and finance isn’t a technology decision. It’s an operating-model decision wearing a technology costume. The real question isn’t “what can we automate?” It’s “what should this process look like, and is it worth automating at all, or worth removing?”
Redesign first, then automate
That ordering matters enormously. Redesign first, then automate, and AI compounds a good decision. Automate first and redesign never comes, because once it’s running, it’s “done”, and the appetite to revisit it evaporates. The sequence isn’t a nicety. It’s the whole game.
This is the least glamorous truth in the AI conversation, which is exactly why it’s worth saying. Everyone wants to talk about agents and capability. Far fewer want to talk about whether the thing being automated deserves to exist. But that’s where the value is won or lost, long before any platform is switched on.
Frequently asked questions
Why doesn’t AI fix a broken process?
It runs the process faster without changing its design. A broken process automated becomes the wrong way of working, at scale, and harder to see.
Why is a slow, manual process not always a good automation candidate?
Slow and painful is often a symptom. A process can be slow because of approval steps that exist for reasons nobody remembers, or manual because two systems were never properly joined. Automating it makes a bad design permanent.
Is AI in HR and finance a technology decision?
It is an operating-model decision wearing a technology costume. The real question is what the process should look like, and whether it is worth automating at all, or worth removing.
What is the right order: automate or redesign?
Redesign first, then automate. AI then compounds a good decision. Automate first and redesign rarely comes, because once the process is running it is considered done.
So before you automate: are you fixing the work, or just speeding it up?