Copilot agent for contract analysis at a 14-lawyer law firm

A Warsaw law firm serving primarily B2B and corporate clients employed 14 lawyers across 3 practice groups (commercial law, employment, tax). Each spent 60–90 minutes at the start of every day reading newly delivered contracts — often standard contracts for their segment, with recurring risks and clauses.

The issue wasn't the analysis itself — that's where lawyers shine. The issue was uneven quality of the first pass. A senior commercial lawyer with 12 years' experience would spot the risks in 25 minutes. A junior (3 years post bar) needed 2 hours and potentially missed a clause they hadn't yet seen across their first hundred contracts. There was no playbook to level the depth of the first review without burning weeks on internal QC.

On top of that, the firm wasn't getting leverage out of its own know-how. A few hundred prior analyses, opinions and memos sat in SharePoint, but no lawyer used them systematically — because no one remembered that a colleague from another group had written about an identical risk two years earlier. The firm's intellectual capital wasn't an operational asset.

Weeks 1–3: workshops with the firm's partners (sessions per practice group) — writing down typical risks per domain, good-clause playbooks, anti-patterns. Each of the 3 groups produced its own risk and pattern dictionary. Those dictionaries + 380 historical opinions from the last 4 years were indexed in Azure AI Search with permissions respected (a commercial lawyer doesn't see tax opinions without SharePoint access).

Build took 4 weeks. Microsoft 365 Copilot as the interface in Word and Outlook: the lawyer opens a new contract in Word, hits „analyze contract”, and gets a side panel with identified risks mapped onto their group's dictionary. Every risk links to a paragraph in the contract and to the most relevant historical firm opinion. Copilot Studio as a Teams agent: for questions like „did we have a similar clause in our opinion for client X two years ago?” the agent searches the archive and returns concrete documents with citations.

Security was central given attorney–client privilege. Microsoft Purview with DLP policies: client data never leaves the firm's M365 tenant, every Copilot interaction is auditable, the agent doesn't export content externally. The last 2 weeks were a pilot on 3 lawyers from 3 groups, dictionary tuning and refinement of the client-memo templates the Copilot auto-generates.

The most visible effect: risk-spotting accuracy for lawyers with 1–3 years' experience went up 45% in the first 8 weeks after rollout. The agent didn't replace their judgment — it leveled the starting point. A junior lawyer now begins the analysis from the agent's risk list grounded in the group's playbook, then layers their own interpretation on top. A senior gets the same baseline but modifies it based on the unique details of the case.

The firm's partners noticed an important shift: the time they used to spend correcting first drafts from juniors dropped 60%. First drafts are now meaningfully closer to the final version. That freed partners up for strategic client work — which translates into higher client satisfaction and better account retention.

An unexpected effect: activation of the firm's own know-how. The 380 historical opinions previously sitting unused in SharePoint are now a real asset — when a new contract comes in, the agent automatically references the most relevant historical opinions. That changed how the firm sees the value of the work it produces: today's opinion isn't just a deliverable for today's client, it's material that will help the firm in three years.