Skip to content

RPA automation and AI agents

Bordeaux and Nouvelle-Aquitaine, remote across France

Your teams spend hours re-keying, copying, checking, chasing. Part of that work can be handed to a software robot, part to an AI agent, and part should never be automated at all. I step in to sort the three: identify candidate tasks, check feasibility, build a proof of concept, quantify the days saved, then frame the rollout and supervision. The goal is not to automate for its own sake, but to give time back where it is missing.

Book a free call

The problem I solve

RPA has a poor reputation in many organisations: robots launched on unstable processes, breaking at every screen change, that nobody dares touch any more. The problem is not the tool, it is the choice of tasks. A rules-based robot excels on a stable, high-volume process with few exceptions. It fails as soon as special cases multiply. AI agents open a second family of cases, more tolerant of variation, but they need a supervision framework and human control. My job: qualify each task before writing a line, and say no when that is the right answer.

How I work

Candidate task inventory

Short workshops with the teams who do the work. We list repetitive tasks, their volume, frequency and exceptions. Each task gets a stability score and a value score: that is what decides what happens next.

Feasibility and proof of concept

For one or two well-chosen tasks, I build a robot (UiPath or Selenium) or an AI agent that handles real cases. We measure the time saved and the exception rate on real data, not on a demonstration.

Quantifying the days saved

From the proof of concept measurements, I calculate the days of manual work given back per year, the build cost and the maintenance cost. You decide on a figure with its assumptions written down, not on a promise.

Rollout and supervision

Rollout plan in waves, access and secrets management, logging, alerts on failure, and a named owner on the business side. A robot without supervision always ends up running on empty without anyone noticing.

Real example

Where this comes from: on assignments at La Banque Postale and EDF, I put in place RPA automations with UiPath and Selenium that amounted to between 100 and 200 days of manual work saved. At EDF, I was also steering a demand management application handling around 20,000 tickets a year.

Read the case study

Who it is for

  • Business units whose teams spend their days re-keying data between two tools
  • IT departments that want to frame an RPA approach before every team launches its own robot
  • PMOs looking to free up days on reporting or control processes
  • Organisations that already have fragile robots and want to decide which ones to keep

What you get

  • An inventory of candidate tasks, scored by stability, volume and value
  • A feasibility study per task: suitable tool, risks, access prerequisites
  • A working proof of concept on real cases
  • A quantified estimate of days saved per year, with its assumptions written down
  • A rollout plan and a supervision setup with a named owner

Frequently asked questions

When is RPA the right tool, and when is it not?+

RPA suits a stable, documented, high-volume process with few exceptions: file extraction, re-keying between two applications, consistency checks. It does not suit a process that changes every month, relies on one person's judgement, or has more exceptions than normal cases. In that case, I say so at the inventory stage.

When does an AI agent replace a rules-based robot?+

When the task means reading an unstructured document, classifying a request or writing a variable reply, a rules-based robot stalls. An AI agent tolerates that variation, provided there is a strict framework: authorised data, human validation on sensitive cases, logged decisions. I always start with the rules-based robot when it is enough: it is cheaper and more predictable.

Who maintains the robots once they are deployed?+

That is the question to settle before rollout, not after. The supervision plan names a business owner and a technical contact, defines who is alerted on failure and who fixes things when an application changes its screens. I can train your teams in that maintenance or provide occasional follow-up, on a quote basis.

How much does it cost?+

On a quote basis, after a free first call. The inventory and feasibility study are priced in days, the proof of concept as a fixed fee once the tasks are chosen, the rollout according to the number of robots or agents. You can stop after each stage: each one produces a deliverable that stands on its own.

The programmes

Automation covers business processes. It includes neither a technical security audit of your applications nor legal advice on data use.

Hours to give back to your teams

Book a free first call. Bring one or two tasks that annoy you: we will see whether they deserve a robot, an agent, or nothing at all.