Case study · Client · Consulting
A consultancy now sees every legal deadline 60 days ahead, instead of relying on one person's memory.
An innovation-funding consultancy runs missions that last years, with legal deadlines that cannot be missed. Its data was spread across six tools and its steering relied on the coordinator's memory. We built a working layer on top of its database, one view per job, then a clean data model the team can query from its AI assistant.
- Company
- Innovation-funding consultancy, about 12 people
- Function
- Operations and mission management
- In production
- Since July 2026
- Users
- Five teams, 19 accounts
From the scoping audit and our delivery logs, July to October 2026.
Everything depended on one person's memory.
Each mission involves a coordinator, a scientific lead and a tax lead, sometimes over several years, and ends in a file that must stay defensible long after. The firm's data lived in a shared database, plus drives, meeting recordings, spreadsheets and email: about 235 missions, 250 companies, 350 contracts and 520 invoices.
The database showed everything to everyone, and nobody had the screen their job needed. Who has to act now, where a file is blocked, which deadline is coming: the answers lived in the coordinator's head.
A working layer first, then a clean foundation.
First, we placed an application on top of the existing database, without migrating anything: one space for coordination, one for the scientific team, one for the tax team, each seeing only what concerns it. Meeting recordings appear on each mission, assigned in one click.
Then we rebuilt the data model properly, reconciled every record field by field against the old base, and connected it to the team's AI assistants. Staff can now query their data from Claude or ChatGPT, with their own permissions. The old invoicing automation is being rebuilt as rules inside the system.
Map
Process workshops with each team: coordination, scientific, tax.
Layer
An app on top of the existing database, with one view per role and a full activity log.
Rebuild
A clean data model, imported read-only, with bad records quarantined and parity checked.
Connect
An AI connector, so staff can query and act on their data from their assistant, with their own rights.
Before and after.
Before How it worked
- Data spread across six tools
- One shared grid where everyone saw everything
- Deadlines and blockers held in one person's memory
- Key pages taking up to 20 seconds to load
After How it works now
- One system, with a view per team
- Each role sees and edits only what concerns it
- Dashboards for missions at risk, clients to chase and deadlines 60 days ahead
- Dashboards that open in under a second
Steering stopped relying on someone remembering. The system remembers, and shows each person what to do next.
The business impact.
- 60 daysof key filing dates visible ahead, for the whole firm
- 30 dayswithout contact: the point at which a client is flagged to follow up
- 20 s → 3 sto open the meetings page, then under a second for dashboards
- 74steps of an invoicing automation being rebuilt inside the system, one fragile tool fewer
The effect on revenue per employee has not been measured yet. Speed and parity figures come from our delivery logs; parity sets aside errors already present in the old database.
More case studies.
See all 19You can keep adding tools.
Or rebuild how your company works.
Not ready yet? Join the free live on how to create your company brain, every Thursday at 12pm, Paris time.
Sign up for the live