Pick a process that happens often, takes real hours, follows a recognisable pattern and does limited damage when something goes wrong. Measure it for a week before you decide. Don't start with the most impressive idea; start with the work your team would gladly hand over tomorrow.
How do you score a process?
| Question | Good sign | Warning sign |
|---|---|---|
| How often? | Daily or weekly | A few times a year |
| How much time? | Hours that you can count | "It feels like a lot" |
| How consistent? | A clear pattern | Everyone does it differently |
| What if it goes wrong? | Easy to catch and undo | Direct damage to money or customers |
| Is the data available? | In systems you can connect | In people's heads |
Which processes often come out on top?
- Handling and routing incoming leads
- Sorting and answering a shared inbox
- Keeping a CRM up to date
- Preparing quotes from job details
- Processing invoices and documents
- Building recurring reports
How do you measure before you decide?
Follow the work for a week. Count how often it happens, how long it takes and where it stalls. Those few numbers do two things: they tell you where to start, and they give you a baseline to judge the result against later. Without a baseline nobody can tell whether it worked.
What should you not start with?
Don't start with processes where the rules are still being discussed. Automating an unclear process only makes the confusion faster. Also avoid starting with work that goes straight to customers without a check, or with the most complex process in the company because it's the most exciting one.
What does the first step look like?
One process, a clear definition of done, and a person who checks the output until it has proven itself. An Automation Sprint is built for this. Larger agents start with a technical Discovery, so that the scope is clear before the quote.
