Azimut AI
Illustrative scenario. The company on this page doesn't exist: it's a typical profile for the sector, built to explain how Azimut would be used. It doesn't describe any customer or claim measured results.
The company in this scenario
- 120 people: 80 lawyers in corporate, employment, litigation and tax, and 40 in support functions.
- Documents in SharePoint, organized by client and matter; Outlook and Teams.
- Already uses specialized legal databases for legislation and case law.
- First- and second-year associates who spend much of the day hunting for templates.
The problem
The firm has drafted thousands of contracts, claims and opinions, and every associate goes looking for the right template by asking down the hall. The legal database answers what the law and the courts say; it doesn't answer how this firm handled it last time, with which clause, or with which partner. Large matters get read end to end to prepare a summary. And some lawyers already paste client documents into public AI tools, which is exactly what the firm can't afford.
How it would work with Azimut
Azimut would connect to SharePoint and email inheriting permissions by client and matter: a lawyer would only get answers from the matters they already have access to, and ethical walls between teams would stay in place. On that foundation, the firm would publish four agents.
| Agent | What it does | Who uses it | What gets measured |
|---|---|---|---|
| Firm precedents | Finds the firm's own contracts, briefs and opinions similar to the matter, citing the source document. | Lawyers | Time to the starting template. |
| First draft | Drafts from the templates and clauses approved by the firm, not from generic text. | Associates | Hours to a reviewable first draft. |
| Matter summary | Summarizes a matter or a document set: parties, facts, deadlines and open issues. | Lawyers and partners | Reading hours per matter. |
| Recurring questions | Prepares the answer to questions clients keep asking, with the firm's internal position, for a lawyer to review. | High-volume practice areas | Response time to the client. |
Drafts and summaries of long matters call for the top-capability model; precedent search works well with a mid-tier one. Each lawyer picks the model per task from OpenAI, Anthropic, Google and Mistral, and providers don't train on the data sent to them.
Where you'd start
You'd start with the precedents agent. The whole firm uses it, it produces no text that reaches the client (it only finds the firm's own work), and it forces you to check that permissions by matter are inherited correctly before turning on anything that drafts.
What's out of scope
Azimut doesn't replace the legal databases the firm already uses for legislation and case law, and it doesn't give legal advice: everything that goes to a client is reviewed and signed by a lawyer, who keeps professional responsibility. There's no connector for specialized legal document management systems; today Azimut works on SharePoint, OneDrive or Google Drive. Integrations are confirmed during rollout and only the tested ones get activated.
How it would be measured
Before turning on any agent, you set the baseline, with the company's own numbers rather than industry averages.
Week 0. Hours to a first draft on a sample of recent matters, reading hours on the quarter's largest matters, and current AI use in the firm: who, with which tool and with which documents.
Month 3. The same numbers, next to Azimut's usage panel by user, practice area, model and tool, plus a check that no agent has answered with documents outside the asker's permissions.
Frequently asked questions
Can a lawyer see another team's matters through Azimut?
No. Agents inherit the source system's permissions: if a lawyer has no access to a matter, Azimut returns nothing from it.
Does it replace our legal database?
No. That database answers what the law and the courts say; Azimut answers what the firm has done. They're used together.
Is it compatible with professional secrecy and GDPR?
Azimut is hosted on Google Cloud in the Madrid region, under a data processing agreement, and model providers don't train on the data sent to them. The final assessment belongs to each firm's compliance lead.