Back to Blog

    Agentic AI

    AI chatbot for financial data: a controlled pilot

    An AI chatbot can explain approved financial reports. Here is how to test source data, permissions, answer quality, and human review.

    April 4, 20252 min read
    Marcel Haas

    Marcel Haas

    Solution Architect, CEO

    marcel.haas@cnext.ch
    CNEXT
    In this article

    A chatbot can explain figures. It should neither approve journal entries nor make investment decisions just because its answer sounds convincing. Start with one bounded question, such as: why does revenue in the monthly report differ from the previous year?

    What the pilot should answer

    Give the chatbot only approved reports and definitions. It should name the period, data source and metric used in its answer. A good test question is: “How did gross margin change in the second quarter, and which line items explain the difference?” The answer is only useful once a finance owner can check the calculation and the supporting record in the source system.

    A Power BI report can visualise metrics. A conversation about them also needs a suitable query or knowledge interface, access rules and testing. Check refresh intervals and show in the interface how current the figures are.

    Four controls before rollout

    1. 1Data: Define the calculation, owner, source and refresh time for each metric. Test missing, late and conflicting data.
    2. 2Access: Share only the data a user role is allowed to see. Power BI row-level security filters rows, but it does not automatically protect columns or every connected system. Test roles with real test accounts.
    3. 3Answer quality: Build a set of typical and difficult questions with verified answers. Measure wrong figures, unsupported statements and cases where the chatbot should rather not answer. Microsoft points out that generated answers are not automatically checked for accuracy.
    4. 4Accountability: Have a responsible person review balance-sheet interpretations, recommendations and external statements. If a supervised financial institution uses the service, AI governance and risk management belong in its existing controls.

    A sensible next step

    Choose one internal monthly report and five recurring questions. Check permissions and data quality first, then the answers against the report. Record errors and stop an answer when its source or calculation cannot be traced. Only after this pilot can you judge whether the conversational approach saves time.

    Sources

    Agentic AI
    Teilen:

    AI tools may have supported the creation of this article. CNEXT is responsible for the published content. Product features and prices can change; check the linked primary sources for current details.

    Marcel Haas

    Marcel Haas

    Solution Architect, CEO

    6x Microsoft Applied Skills

    Discuss a finance chatbot pilot

    We plan a pilot with you based on one monthly report, five questions and clear controls.