Most businesses are carrying a few hours a week of copying data between systems.
Most businesses are carrying a few hours a week of copying data between systems.
Look for anything where a person reads one screen and types into another. That is not work, that is an integration that has not been built yet.
The second signal is a spreadsheet that exists purely to reconcile two systems that should already agree.
Time is the obvious saving, but consistency matters more. Manual transfer produces errors at a steady rate, and the errors are usually discovered by a customer.
Automate in one direction first, log everything, and keep the manual path available until the automated one has run clean for a while. Reversibility is worth more than elegance in the first month.
Ask people to note every time they copy something from one screen into another for a week. The resulting list is usually longer than management expects and shorter than the team fears.
Rank it by frequency multiplied by tedium, not by how technically interesting each item is. The most valuable automation is often the dullest one on the list.
The source system changes a field name. A record arrives malformed. The API is down for an hour. Somebody edits a record halfway through a sync. Decide in advance whether the automation retries, queues, skips or alerts a human.
Silent failure is the worst outcome, because the process appears to be working while data quietly diverges. Anything automated needs to be able to say loudly that it has stopped.
For the first month, keep the manual route available and compare the two outputs. Reversibility is worth far more than elegance while trust is still being established.
Once it has run clean for a few weeks, retire the manual path deliberately rather than by neglect, and make sure somebody still knows how it worked.
Look for copy and paste between two screens. Every instance is an integration that has not been built, and someone is being paid to be the API.
Look for spreadsheets that exist to reconcile two systems. The spreadsheet is a map of the missing connection, and it usually documents the business rules better than anything official does.
Look for anything that happens at the end of the month and takes a full day. Batch work is nearly always automatable, and it is the easiest to justify because everyone can already state what it costs.
Look for the tasks people quietly do at home in the evening. Those are the ones that never appear in a process document and are frequently the worst.
The build is the visible part and usually the smaller one. The real commitment is ownership: someone has to notice when it stops, understand what it does, and update it when the upstream system changes its export format without telling anyone.
Budget for the unhappy paths, which are most of the work. A script that handles the expected case is a demo. Handling a malformed file, a timeout, a duplicate record and a partial run is what makes it a system.
And budget for the handover. An automation only one person understands is a liability wearing the costume of an asset, and it becomes a crisis the week that person leaves.
Never automate a process you have not first written down and simplified. Half the steps in most manual processes exist because of a constraint that disappeared years ago, and automating them faithfully preserves the constraint forever in code.
Write the process down, delete what is no longer needed, agree the simplified version with the people who do it, and only then automate. Frequently the simplification alone recovers most of the time, and the automation becomes optional.
The worst outcome in this work is not a failed automation. It is a successful one that makes a bad process permanent.