AI governance in finance
How can you authorise, control and test AI use cases in finance without delegating sensitive decisions?
3500 € per session, excl. VAT
In short
“Managing AI governance in finance” is a private in-person session of two days for finance leadership teams and management control managers or risk, compliance and internal audit managers. It helps you map financial uses and their risks, assign responsibilities for authorisation and control. Groups of up to 12 participants, in French, with exercises on your own work cases.
Why this training?
In finance, a plausible but unfounded explanation can distort an assessment of the cash position or a period-end closing commentary. Governance must link each use to its data, risks and the person responsible for validation. A charter alone is not enough: decisions are needed on who authorises use, what evidence to retain and when to suspend use. The workshops turn these requirements into testable controls and documented judgements. Running case study: Orvalis Robinetterie, a fictitious organisation, wants to use generative AI to comment on its budget variances and prepare its treasury notes. A summary invents an explanation and exposes confidential banking terms: the finance leadership requires a control framework before any pilot. Transfer to the workplace: Present the dossier to the finance sponsor, the security lead and the legal department; adapt the controls, then request authorisation for a limited pilot, with no connection to payment systems.
Objectives of this training
- Map financial uses and their risks
- Assign responsibilities for authorisation and control
- Design a traceable human validation workflow
- Test controls against financial anomalies
- Define indicators and stopping conditions for the pilot
Who is it for and what are the prerequisites?
Roles and tasks concerned
- Finance leadership teams and management control managers
- Risk, compliance and internal audit managers
- AI project managers for the finance function
Prior skills
- Have a sound command of a financial process and its internal controls.
- Use generative AI and understand its reliability limitations.
What you will work on
Define the boundaries of use
Objective: Establish the financial scope. Writing assistance, analysis and sensitive decisions.
Assess the risks
Objective: Prioritise risk exposures. Numerical errors, confidentiality, bias and dependencies.
Set rules for data use
Objective: Limit the information transmitted. Access, minimisation, retention; applicable requirements to be checked with your legal department.
Assign responsibilities
Objective: Separate production from approval. Finance sponsor, business owner, security and independent control.
Develop the controls
Objective: Make the checks executable. Accounting reconciliation, justification of variances and traceability.
Prototype the authorisation process
Objective: Formalise the transition to the pilot. Use case dossier, reservations and signatures.
Prepare the tests
Objective: Define a verifiable reference. Standard cases, edge cases and expected results.
Put the outputs to the test
Objective: Measure their reliability. Accuracy of amounts and source-backed explanations.
Simulate an incident
Objective: Organise suspension and escalation. Suspected leak, preservation of evidence and resumption.
Measure quality
Objective: Base acceptance on objective criteria. Error rate, source coverage and human corrections.
Plan the pilot
Objective: Define the boundaries of the experiment. Scope, resources, milestones and reversion to the manual process.
Defend the framework
Objective: Justify the trade-offs. Residual risks and required professional validation.
What you take away
- Register of financial uses and risks
- Responsibility matrix and authorisation workflow
- Prototype control dossier with evidence
- Pilot protocol, indicator dashboard and test report
Method and assessment
- Short instructional sessions followed by immediate practice on the case study used throughout.
- Individual work followed by work in pairs, with peer review of the outputs.
- Facilitated group discussion: what works, what needs correcting and what remains a human decision.
- Preparation: Describe a proposed financial use case without confidential information. Identify the tools authorised by your organisation and prepare only fictitious or anonymised examples.
Observable criteria: Each use case has an owner, a risk and an associated control. ; Each amount tested is reconciled to its source. ; The handling of injected anomalies is documented. ; Acceptance and stopping thresholds are set before testing. ; Every financial output requires validation by a qualified professional..
Frequently asked questions
Is an actual deployment included?
No, the 14 hours excluding breaks produce a prototype and a pilot plan.
Can we use real financial accounts?
The exercises use only fictitious or anonymised cases.
Does AI validate the analyses?
No, a qualified professional validates the outputs; no sensitive decisions are automated.
Are programming skills required?
No, the prototype is based on registers, validation workflows and documented tests.
Last updated: