A hedge fund kept its Excel models and got a data platform underneath them

The fund's research lived in hundreds of analyst-built Excel models, one per company, built over nearly two decades. Nobody was going to give those up. So the platform went underneath them: a nightly job opens every model, runs its scenarios and writes the results to a warehouse with the date attached. An Excel add-in returns the firm's own numbers to the cell the analyst is already in.

The problem

A workbook has no API, no memory of earlier versions and no way to say a number is stale. Comparing estimates across companies meant opening workbooks and copying cells. Market data was pulled straight into each file, so a failed refresh silently reused last week's value and nobody noticed until someone looked.

What we built

A data platform under the models. An analyst saves a model to the shared drive with a short config file describing how to run it. Every night the platform opens each model, cycles its scenarios and writes the outputs with the date. Every model gets a history: what the analyst believed in June against what they believe now.

Running every model the same way also validated them. The first full run exposed models linked to moved files that had been quietly working from stale data. A report groups those breakages by the file that needs fixing and ranks fixes by how many models each one unblocks.

An Excel add-in. Analysts type a function and get a value from the platform: the firm's own estimates, market data, figures as of a given date. Each number shows its source and capture time. When a lookup fails, the cell says so, which never happened before.

Where it goes next

A conversational layer over the full dataset is in progress: questions across every model, answered with scenario, definition, capture time and a stale flag. It is planned for the warehouse, for Slack morning briefs and over MCP for agents.