The request usually arrives in the same form: someone on the team is doing something repetitive, everyone can see it, and the ask is to automate that. It is a reasonable instinct and often the wrong starting point. Visibility and cost are different measures. The task that generates the most complaints is frequently a five-minute daily irritation, while the expensive process is the one nobody mentions because it has always been done that way.
Four questions before anything is built
Before scoping a workflow, we work through four questions with whoever owns the process. They take an hour and remove most bad candidates.
Frequency beats complexity
A complicated process performed twice a month is a poor first candidate. A simple one performed forty times a day is an excellent one, even if each instance takes only a couple of minutes. Automation returns compound value on frequency, and simple high-frequency work is also where the rules are clearest, which makes the build faster and the failure modes easier to predict.
This is why lead handling, document intake and internal approvals are so often the first projects. They are high volume, rule-based, and the current cost is spread thinly enough across the week that nobody has added it up.
Decide where the human stays
Every workflow needs a defined point at which a person is still responsible. In lead qualification that might be the moment before a price is quoted. In document processing it is usually the exception queue, where anything the extraction step flagged waits for review. Deciding this early is what separates automation people trust from automation they quietly work around.
The failure mode worth guarding against is not the model producing a wrong answer. It is the wrong answer being written into a system of record where nobody notices for three weeks. Logging, reversibility and a visible exception path matter more than model choice.
Start narrow enough to finish
A first automation should be small enough to deliver inside a few weeks and complete enough to run without supervision. Half-automated processes create their own overhead: a step that runs automatically but requires someone to check whether it ran is worse than the manual version.
Once one workflow is genuinely running, the second is easier to specify, because the organisation has learned what it actually wants from automation — and that is usually not what it asked for in the first meeting.
If you can name a process that runs daily and involves moving the same information between two systems, that is usually where to start. Tell us what it is and we will tell you whether it is worth automating.
Discuss Your Business Challenge