An internal AI copilot that reads the studio's full project memory — SharePoint, Teams, project plans, timesheet exports — and answers leadership questions in plain language. With citations. Without a byte leaving the tenant.
The studio runs over one hundred active projects across India and the United States. Leadership relies on weekly status reports, project plans, timesheet exports, and the institutional memory of three senior managers to know where things stand. Most weeks, getting a clear picture of even ten projects took half a day.
The data existed. It lived in SharePoint, Teams, a handful of Excel files, and the heads of three people who had been at the firm for fifteen years. Leadership wanted one place to ask "what is the actual state of project X?" and have the answer come back in plain language, with citations, and never leave their own Microsoft tenant.
We separated the work into four phases. The first two were unglamorous — auditing the data estate, designing the architecture. The visible work came later. Most AI projects fail in the first two phases. We refused to skip them.
We mapped the firm's data estate. Eleven document libraries, six Teams channels, two timesheet exports, three SharePoint sites. We graded each source on quality, recency, and access permissions. The first deliverable was a one-page memo on what was worth indexing and what should be excluded. Half of the available data made the cut.
We designed an indexing pipeline that respects role-based access. A senior manager's queries pull from documents she has rights to read. A project manager's queries do not. Tenant-isolated. No data, queries, or embeddings leave the firm's Microsoft tenant. We chose every component for auditability, not novelty.
We built the agent, the retrieval layer, and the evaluation harness in parallel. Every release ran against a hand-curated set of eighty internal questions before going live. We tracked accuracy, citation correctness, and refusal quality. Releases that did not improve scores were rolled back automatically.
Released first to four members of the leadership team. They asked the questions they actually ask in Monday review meetings. The agent answered correctly on the first try in eighty-nine percent of cases, partially on a further seven, and refused on the rest. Which is what we wanted. PM teams are being rolled on now.
Leadership now starts Monday reviews with the agent open. Status reporting time is down. The conversations that follow are richer because everyone is reading from the same baseline. The institutional knowledge that lived in three senior managers' heads is being made searchable by everyone else who needs it.
The first week I used it, I asked it the questions I had asked my PMs the week before. The answers matched. The difference was speed.