AskMIS — Analytics Request Concierge
Submitted by Nader Mahmoud Ghloum Abdullah
Pain point
Ad-hoc data requests arrive as one-line emails or Teams messages ("need merchant numbers for the report, asap") with no timeframe, metric definition, or stated purpose. The first half of the job is negotiating what was actually wanted — typically two or three email round trips before work can start. Sometimes a request duplicates one answered months earlier. The queue itself is invisible — nobody, including the team lead, can say how many open requests exist or which are at risk.
Proposed solution
A Power Automate flow triggers on submission to a SharePoint request list. An AI Builder prompt scores the request 0–100 against a completeness rubric (timeframe, metric definition, business purpose, output format, audience), classifies request type and data domain, estimates an effort band, and generates the specific clarifying questions the request is missing. Below the quality threshold, those questions go back to the requester as an Adaptive Card in Teams with an inline resubmit option, and no analyst is assigned yet. Above threshold, a second AI call checks the request against previously closed requests in the same domain for semantic duplicates; if a high-confidence match is found, the requester is offered the prior answer instead. Clean requests route to the least-loaded analyst in that domain with an SLA date attached. The AI only ever sees the request description — never any banking data.
Claude surface
Chat, Cowork