1. Assess the initial situation
Select a precise process: follow up on an invoice, produce a document, retype a request, or consolidate a report. Over a representative period, record the volume processed, time per file, rework, and errors. Distinguish active time from wait time.
Interview the people who perform the task and observe a few cases. A memory estimate may underestimate exceptions or overestimate the time that can be eliminated. If volumes vary significantly by season, keep this variation in your annual calculation.
2. Calculate recoverable capacity
For a daily task, a starting formula is: calculated as: workforce involved × minutes per day ÷ 60 × working days × actual realizable reduction. The last variable must incorporate the time control that will remain necessary.
Purely illustrative example: 8 people each dedicate 30 minutes per day to a task over 220 days. This amounts to 880 hours per year. If a solution genuinely halves this time, it frees up 440 hours. Assuming a fully loaded labour cost of 40 € per hour, the corresponding capacity is valued at 17,600 € per year.
This amount is neither a guaranteed profit nor an automatic reduction in costs. It may result in more cases processed, fewer overtime hours, or more time for clients. The scenario should specify which of these effects is expected.
3. Include all costs
- Implementation: scoping, development or parameterization, data recovery, testing and training.
- Operation: licenses, hosting, consumption of AI services and supervision.
- Follow-up: maintenance, residual human oversight, and evolution of connected tools.
- Internal mobilization: time of pilot users and the project manager.
Avoid double-counting the same position: if human controls have already been subtracted from the recoverable capacity, do not add them again to the cost. Establish a conservative scenario and a central scenario with explicit assumptions.
4. Decide through a measured pilot
An indicator of return on the first year compares the realized benefit to the total costs of that year: (net profit − total costs) ÷ total costs. This calculation makes sense only if the valuation chosen corresponds to a useful and observable effect for the company.
A pilot verifies hypotheses before expanding the scope. Track time per dossier, the proportion of exceptions, errors, and adoption. A decrease in average time can mask an increase in verification work: look at the full process.
If the data is unstable or the task is too infrequent, a process simplification may suffice. The guide automate, acquire or build helps compare options. The Mellion diagnostic enables prioritisation before aautomation and AI.
Frequently Asked Questions
Should the longest task be automated first?
Not always. Frequency, data quality, exceptions, and risk also matter. A shorter task that is very frequent and stable may constitute a better first pilot.
How can you apply this method with Insight Group?
Gather volumes, durations, and an example process. Mellion supports the diagnosis and implementation across France, Brussels, and Madrid, on-site or remotely depending on the project.
Mellion supports your projects across France, Brussels, and Madrid with the expertise of Insight Group collective. Also consult our intervention terms in the Hauts-de-France and in Île-de-France.