Before you automate: five questions
Automating a messy process just breaks it faster. Five questions to tell whether a process is ready to be automated with AI.
- Published
- November 13, 2026
- Reading time
- 2 minutes
- Written by
- The Roshal Lab team

Automation isn't always the answer
Today you can automate almost any task with AI. The question isn't whether you can, but whether you should. Before we build anything, we go through these five questions with you.
The five questions
- 01
Is the process clear and stable?
If everyone does it differently, or it changes every week, it needs to be sorted out first. Automating chaos makes it faster, not better.
- 02
What does it cost today?
Hours per week, errors, delays, lost customers. Without that number, there's no way to know whether automation is worth it.
- 03
What happens when it fails?
Every automation fails at some point. You need to define how failures get detected, who gets notified and what the manual fallback is.
- 04
Who maintains it?
Tools change, data changes and the business changes. Someone has to take ownership when that happens.
- 05
How will we know it worked?
We agree on the metric before we start: time saved, errors avoided, faster response times.
When not to automate
- When the task happens rarely and takes only a few minutes.
- When it requires judgment that changes case by case, and a mistake is costly.
- When the real problem lies elsewhere, for example, a business rule nobody has reviewed.

