What we don’t automate
Anything where the answer depends on who’s asking, or where being wrong costs more than being slow.
Half of building automation well is deciding what not to build. We get asked to automate things we won’t, and the reasons are worth writing down, because the same test applies to every workflow we do build.
What shouldn’t a business automate?
Anything where the right answer depends on who’s asking, and anything where being wrong costs more than being slow. That covers most customer-facing judgment calls, most approvals with money attached, and anything a regulator or a relationship would hold a named person accountable for. Everything up to that decision can be drafted; the decision itself stays human.
What are common examples?
Deciding which overdue customers get a reminder and which get a call from the owner. Extending credit. Anything in hiring. Responding to a complaint. Signing off a payment run. Choosing what to say to a client whose project has slipped. In each case the facts can be gathered and the options drafted by an agent; the choice can’t, because the system can’t see the relationship or carry the accountability.
Isn’t that just being cautious?
No. Automation that removes people from decisions fails in a particular way: quietly, then all at once, and then nobody trusts it again. Keeping the decision with a person is what makes the rest of the workflow something a team will keep using a year later. It’s method, not nerves.
What do you do instead?
Everything up to the decision. For the overdue-customer example: pull the ledger, flag who’s overdue and by how much, draft both versions of the message, and put them in front of the person who knows the customers. They pick. The system sends what they picked and records it. The person spends two minutes deciding instead of twenty gathering.
How does this show up in an engagement?
In writing. The checkpoint map that ships with every workflow names the decisions that stay human, and the audit report names the processes we recommend leaving alone entirely. If nothing is worth building, the audit says so, and that list is usually the most useful page in the report.
Does the line move over time?
The drafting expands as the workflow earns trust; the decisions don’t move unless the client chooses to move them. A rule that was right at launch can be revisited at the monthly review of the record, with evidence. What never changes is that nothing we build runs entirely on its own.
The test is short enough to remember and strict enough to be useful. We apply it to our own operations first.
If you’re not sure whether something should be automated, ask who would be accountable if it went wrong. If the answer is a person, it’s a decision.
Start with the work that keeps getting in the way.