™Affaires
AI automation for business processes
The hard part is not automating a process. It is choosing the right one and knowing what it currently costs.
Par Tomryn ·
Cet article n'est disponible qu'en anglais pour le moment.
Automation projects fail for boring reasons. The process was never written down, so the automation encoded a version of it that nobody actually follows. Or the automated step was not the bottleneck, so nothing got faster. Or it worked, and nobody used it.
Choosing what to automate first
- It happens often, several times a week at least, or the savings never accumulate.
- It follows a recognisable shape, even if the details vary.
- It currently takes time from someone whose time is expensive or scarce.
- A mistake is visible and recoverable rather than silent and costly.
- You can state today's cost in hours, even roughly.
Map it before you automate it
Write down what actually happens, including the informal steps and the exceptions. This is uncomfortable, because it usually reveals that two people do the same job differently and neither knew.
That discovery is valuable on its own. A surprising number of process engagements end with a simplification rather than a system, and that is a legitimate outcome.
Where AI fits inside a process
- Reading unstructured input (emails, forms and PDFs) and turning it into structured fields.
- Classifying or routing work that does not fit clean rules.
- Drafting routine text for a person to check and send.
- Summarising long material so a decision can be made faster.
What to measure
Pick the measure before building: hours spent on the step, time from request to response, error rate, or backlog size. Measure it for a fortnight first. Without a baseline you will have an opinion about whether it worked, and opinions about your own project are not reliable.