Don't automate yesterday's process

Most AI projects start with a workflow that already exists and ask a narrow question: what here can we automate? It's the wrong first question, and it quietly sets the ceiling on everything that follows.

Every process was designed around constraints — the tools available, the data you could get, the people you had, the speed the business moved at. Many of those constraints are gone. But the process still carries their shape. Automating it faithfully means encoding decade-old assumptions into software and running them faster. You don't fix the problem. You make it permanent, and you make it quick.

Reporting is where we see it most. A team spends two days a month assembling a report because, years ago, the data lived in five systems and someone had to reconcile it by hand. Automate the pulls and you've saved two days — while preserving a report that exists only because the reconciliation used to be hard. The question was never how to produce that report faster. It was whether it should exist in its current form at all.

So the first move isn't automation. It's subtraction. Before you ask what to automate, ask what the process was built to work around, and whether that thing is still true. Often the highest-return AI work isn't acceleration — it's elimination: removing steps that only existed to compensate for a limit you no longer have.

This is uncomfortable, because acceleration is legible and elimination is disruptive. “We made the existing process 40% faster” is an easy thing to report. “We removed the process and replaced the decision it fed” is a harder thing to sell and a better thing to do.

We learned this building ClarityOps. The instinct in diligence software is to make document review faster — the same review, more speed. What actually mattered was redesigning the work around evidence and provenance first, so the review that remained was the review that had to happen. Faster was a byproduct. The redesign was the point.

Earn the right to automate something by first asking whether it should exist. The best process improvement is often the one that leaves less behind.