Case study

How Doctolib made its commercial organisation self-sufficient on AI agents

An agent platform already deployed at scale, but usage concentrated among a few pioneers. Three months on the ground so that every member builds their own agents, then so that the coaching role moves in-house.

Published 30 July 2026

Industry

HealthtechMedical appointment booking

Scope

Commercial and operations organisation5 business teams

Tandem service

Adoption and upskillingScoping, training, co-build, handover

Starting point

Platform already deployedUsage carried by a few pioneers

The challenge

Deploying a platform is not enough to create usage

In most companies, the level of fluency with generative AI is very uneven from one team to the next. A few pioneers seize it and pull far ahead, while the majority stays on the sidelines. Rarely out of disinterest: people are caught up in day-to-day operations, and learning a new tool is never the priority of the day. Doctolib had already deployed an AI agent platform at scale internally, but its commercial and operations organisation showed exactly that pattern. The challenge was not the tool, it was already there. It was making every member self-sufficient at building their own useful agents, then moving from individual use to deeper automation, team by team.

Uneven

fluency from one team to the next

A few very advanced pioneers, a majority left on the sidelines, and no intermediate step to progress.

business teams with different needs

Five jobs, therefore five different relationships to the tool, which ruled out any uniform training.

Operations

come before upskilling

People caught up in their daily workload, for whom learning a new tool is never the priority of the day.

The company

Doctolib, France and Germany

Doctolib

Doctolib is Europe's leading medical appointment booking platform, operating in France and Germany.

The group employs around 3,000 people and equips close to 400,000 healthcare professionals. Tandem's engagement covered its commercial and operations organisation, structured into five business teams: sales enablement, process, organisation and planning, automation, analytics. Five different jobs, therefore five different relationships to the tool, which ruled out any uniform training.

Employees

~3,000

Healthcare professionals equipped

~400,000

Markets

France and Germany

Engagement scope

5 business teams

Our approach

Prove value one case at a time, then hand it over

An iterative, pragmatic approach: prove value on one use case, measure, then move to the next, rather than a single mass rollout. And an exit prepared from day one, so that coaching does not depend on Tandem.

Scoping before any training

Several dozen internal interviews, complemented by external benchmarks of market best practices, to target the highest-value scope before training anyone.

A founding all-hands

Generative AI fundamentals, prompting, then a workshop where every member builds their first agent. The link to action is direct: 75 % of the agents attributable to the engagement were created within the following two months.

Team sessions every two weeks

Mapping use cases and debugging with each owner, team by team, to move from the personal agent to business automation.

Co-building agents wired into business tools

Billing anomaly detection with an alert in the team messaging tool, real-time bonus calculation through a dedicated MCP connector, to-do automation, quarterly performance review preparation. Each integration produced a reusable standard for permission inheritance and data partitioning.

Results

From a handful of pioneers to 20 agent builders across 5 teams

In three months, the organisation went from usage carried by a few pioneers to broad, self-sufficient adoption. 90 % of members now build their own agents, and around fifty use cases were identified, about ten of them already in production. Beyond the numbers, the skill did not concentrate in a few experts: it spread across the five business teams, each making the tool its own. Natural AI champions emerged, ready to carry that autonomy over time, and the closing session was run by the internal team: the coaching role was already in-house by the time Tandem left.

of members now build their own agents

agents attributable to the engagement, built by 20 people

Hundreds

of users on the most widely adopted agent

About ten

use cases already in production