Automation projects usually begin with someone naming a task that is obviously tedious. That is a reasonable starting point and a poor stopping point, because the tedious task is often a symptom of a process that should not exist in its current form at all.
Map it before you script it
Follow one real case end to end, with the people who actually do the work rather than the people who designed it. The documented process and the practised one diverge in almost every organisation, and the divergence is where the useful information is: the manual check somebody added after an incident, the spreadsheet that reconciles two systems that should have been integrated.
You will usually find steps that exist only to compensate for an earlier step. Automating those cements the compensation permanently.
Delete before you automate
The cheapest automation is removing the step. Approvals that have never been refused, reports nobody opens, duplicate data entry between two systems that could exchange records directly - these do not need a bot, they need to stop.
- Steps with no recorded instance of changing the outcome
- Reports produced on a schedule that nobody reads
- Data typed into a second system that could be integrated
- Checks that duplicate a validation performed upstream
Decide what happens when it fails
Manual processes degrade visibly - work piles up and somebody notices. Automated ones fail silently and confidently, and the queue looks empty because nothing is being added to it. Before going live, decide how a failure surfaces, who it reaches, and what the manual fallback is.
An automation with no monitored failure path is not a time saving, it is a deferred incident.
Keep a human in the consequential loop
Automate the gathering, the routing and the formatting. Be much more careful about automating the decision, particularly where it affects a customer, a payment or an employment matter. The value is usually in removing the twenty minutes of assembly before the judgement, not the judgement itself.
The measure of a good automation project is not how many steps it removed. It is whether anyone would want the old process back, and whether they could tell you within minutes if it stopped working.