Where AI Automation Pays Off (and Where It Doesn't Yet)
Automation is oversold in both directions — pitched as magic by vendors and dismissed as hype by skeptics. The truth is narrower and more useful: some workflows are excellent candidates, and some will quietly waste a quarter of engineering time. The skill is telling them apart before you build.
Good candidates share three traits
- High volume. The task happens dozens or hundreds of times a day. Automating something that runs twice a month rarely earns back the build.
- Low judgment. The rules are stable and mostly explicit. "Route this invoice to the right approver" automates well; "decide whether this contract is worth signing" does not.
- A clear success signal. You can tell, cheaply and quickly, whether the automation did the right thing — so errors surface instead of compounding silently.
Traps that look like opportunities
The seductive projects are the ones that touch a painful, visible process but violate one of the traits above. A workflow with a dozen edge cases and no clean way to verify the output will consume more time in exception-handling than it ever saved. When a human still has to check every result, you've added a step, not removed one.
Keep a human at the boundary, not in the loop
The reliable pattern isn't full autonomy — it's automation that handles the confident 90% end to end and escalates the uncertain 10% to a person, with the context attached. That keeps throughput high while capping the blast radius of a mistake. Design the escalation path first; it's the part that makes automation safe to trust.
Start with one workflow, measured
Resist the platform. Pick the single highest-volume, lowest-judgment task you have, automate just that, and measure hours saved against hours spent for a month. A working automation on one real workflow teaches you more than a six-month "automation strategy" ever will — and it's reversible if the numbers disappoint.
Automation pays off when the work is frequent, the rules are stable, and mistakes are visible. Match the project to those conditions and it's one of the highest-leverage things a team can do. Ignore them and it's an expensive way to move a problem around.