1. Observe real work
Start by listing the tasks that recur each week: copying information from an email to the CRM, reconciling documents, preparing a quote, following up on a file, or consolidating a report. Note who is involved, which tools are open, and where a human decision is made.
Follow some cases from start to finish. The usual process counts, but incomplete cases, duplicates, and urgent requests often determine the difficulty. A task that seems simple in a demonstration can become fragile if its data changes format or if the rules are not shared.
2. Compare five criteria
For each candidate task, document the following with the people who execute it:
- Frequency: how many times does the task recur, and at what intervals?
- Stability: can the steps and rules be clearly written?
- Data: are the input data available, reliable, and authorised for this use?
- Business impact: what delay, error, or friction will be genuinely reduced?
- Risk: what happens if the solution makes a mistake, and who validates before a binding action?
A first automation is easier to manage when the volume is regular, rules are clear, and errors are reversible. If rules are still contested, clarifying the process may be the best first action. The guide automate, acquire or build helps in choosing the right solution.
3. Test a complete workflow
Let’s take an illustrative example: a commercial request arrives via email. Useful information is extracted, a record is proposed in the CRM, and a draft response is prepared. A person checks the record and approves its sending. Ambiguous requests are returned to a manual processing queue.
This first journey must cover entry, controls, exceptions, and exit. Automating only the copy in the CRM may shift the work to a more costly correction. The pilot must therefore demonstrate the complete process, including what remains intentionally under human control.
4. Measure before expanding
Before the pilot, record a starting point: time per file, percentage of files reviewed, response time, and data quality. After the pilot, compare these same metrics on comparable cases. Also document exceptions and verification time to avoid confusing an accelerated step with a genuinely improved process.
To estimate the recoverable capacity and costs, use our method for calculating the return on an automation. If you wish to prioritise several ideas, theMellion Opportunity Scan is used to scope processes, data, and the next decision. Our page Automation & AI outlines the possible solutions.