Case Study: a public pension fund brings AI into its Finance Division
Pain-point audit, Copilot training and one-to-one support, role by role
Situation
This Geneva public pension fund manages the occupational pension scheme for the State's employees. Its Finance Division deals every day with financial flows, property data and benefit files, across several specialised business applications. The organisation has Microsoft 365 Copilot. The question from the administrative and finance management was not "do we need AI?" but "where does it save us time, in concrete terms, and how do we adopt it without stepping outside our framework?".
Objective
Move from a general-purpose tool to the repetitive tasks of each role, despite three obstacles: pain points that were scattered and rarely documented (Excel lists, reconciliations, bank advices, withholding tax, PDF renaming), a secure environment in which Copilot has a restricted scope, and business applications that are closed off from one another.
Third-party account check (versus 6 to 9 h)
1 h/week
Reusable Copilot agents
2
Reconciliation time
1/10
Duplicates found at year-end close
400
Process
Since November 2024: finance management first
Three half-day training sessions for the finance management, then in-person coaching in December 2025.
Process audit (April 2026)
Interviews with the accounting and back-office team, the wealth management team and management, to list the pain points and repetitive tasks.
Copilot training (April 2026)
Hands-on introduction to the tool already available, to start from a common base.
Summary and prioritisation (July 2026)
Audit report and scoring of each use case with the Maijin Score, according to its feasibility in the client's environment.
Accounting team workshops (September and October 2026)
Three sessions: one collective case for the whole team, then one-to-one support, with each case starting from a real file of the employee.
What comes next
The support is ongoing, and the approach is meant to extend to the wealth management team, already mapped during the audit.
Results
Checking a third-party property bank account, the number one priority of the audit, drops from 6 to 9 hours a week to about 1 hour.
Checking the supporting documents for one building in the portfolio used to take about two weeks of work per quarter. With Copilot and the procedures put in place, it now takes about two days. That is close to 21 hours saved per month, or about 256 hours a year, for a single person.
Seven real cases from the accounting team handled. Two became reusable Copilot agents: the clean-up of the incoming flows export, and a monthly pension summary of about 25,000 lines.
Reconciliation of receipts now takes a tenth of the time, according to the employees at the end of the workshops.
An exhaustive check became possible at year-end close: by cross-checking the actuary commitments and the resigning members to be paid, more than 400 duplicates appeared across more than 4,000 lines, a check that cannot be done by hand.
Not every case saves time. On the reconciliation of property manager statements, the AI does the job well without going faster, as long as the files are not organised for it.
What remains difficult: no direct connection to the business applications (everything goes through exports), Copilot only sees OneDrive, SharePoint and Teams, and reading scanned PDFs remains limited.
Much more could be automated with an agentic tool such as Copilot Cowork, or with MCP connectors to the business applications. For now, the compliance framework does not allow it.
One lesson: by explaining their process to the AI, an employee discovered that they were not always working on documents frozen at the same date as the actuary.
Testimonials
- Very good training. At last, I have a fairly clear view of AI and of the future that awaits us. (a member of the finance management)
Contact us
- Want to upskill your teams?
Let's schedule a meeting to assess your training needs
- Want to identify high-ROI use cases for generative AI?
Let's schedule a strategic call