What to automate first (and what to leave alone)
Most teams automate the wrong thing first — the flashy thing, not the expensive thing. A simple way to find the workflow that actually pays for itself.
When a team decides to "add AI," they usually reach for the most visible task — the demo-friendly one. That's almost never where the money is. The highest-return automation is usually boring: the hand-off nobody owns, repeated dozens of times a day, that quietly costs you more than anything on the roadmap.
Find the expensive workflow, not the annoying one
Annoying and expensive aren't the same thing. The task everyone complains about might run twice a week. The task nobody mentions — copying data between two systems — might run two hundred times a week. Multiply frequency by minutes by people, and the real cost surfaces fast.
A quick triage
- High frequency + rules-based → automate first. This is where automation compounds.
- High frequency + judgment-heavy → augment, don't replace. Put AI in the loop, keep a human at the decision.
- Low frequency + complex → leave it alone. The build cost won't pay back.
- Low frequency + simple → a checklist beats a system.
The trap: automating a broken process
Automating a bad process just produces bad output faster. Before wiring anything together, map the flow and fix the obvious breakage. Half the value in our engagements comes from the mapping step alone — clients see costs they'd stopped noticing.
Don't automate what you should delete. Map first, cut second, automate third.
Start with one workflow that's high-frequency and rules-based. Ship it, measure the hours it returns, and let that fund the next one. Automation that compounds beats a big-bang rollout that stalls.