AI strategy and AI governance
Using AI safely with confidential financial data
Three clear rules and well-configured tools protect confidential financial data better than an AI ban.
Walk through any finance department today and you will see people using AI assistants. Someone pastes a variance analysis into a chatbot for a first draft of the commentary, someone else asks it to explain a formula or tidy up forecast text. So finance is using AI. The question is whether it is doing so safely.
The first reflex is often to block it. In my experience that backfires: people switch to their personal account on their phone and you lose all visibility. It is better to start by asking where the real risk lies.
Where the risk lies
The model itself is rarely the problem. The risk sits in three places.
The first is the account. A consumer account may use prompts to train future models and keeps nothing in a company environment. An enterprise licence usually excludes training, keeps data in agreed regions and provides audit logs.
The second is what goes into the prompt. Unpublished results, salary data, customer names and acquisition plans are each sensitive in a different way.
The third is what comes out and where it ends up. If an AI-drafted figure lands unchecked in a board pack, your control has failed, and you can't blame the AI for that.
Three rules that work in practice
1. Use the company tools, never personal accounts. Make the approved enterprise tool the easiest option. If it is harder to reach than the free version, people will pick the free version.
2. Classify before you paste. A traffic-light model is enough. Green is public or already published information, usable anywhere. Amber is internal data such as budgets and forecasts, only in approved enterprise tools. Red is personal data, results not yet published and deal information: only with explicit approval.
3. Every number has a human owner. AI may draft, summarise and suggest. Whoever signs off remains accountable, and every figure must be traceable to the source system.
Make the right behaviour easy
Rules alone don't change behaviour. What helps is a short list of approved use cases with example prompts, for instance for a commentary based on an exported, anonymised table. A small group of champions in finance can experiment first and help colleagues get started. And people need somewhere to go with the question "may I use this data here?", with an answer within a day.
In the end, safe AI use in finance is mostly good governance: the right tools, clear data classes and a person who stays accountable. Then AI saves time, and you don't have to hope nobody notices how it is being used.
Read next
Let's talk
Would you like to exchange ideas on data, analytics, AI or FP&A? I would be glad to connect.