AI Agents for Law Firms: What Copilot Actually Looks Like Inside a Practice Management System

Every legal tech vendor is talking about AI. Most of them mean a chatbot bolted onto their help page. Some mean a summarisation tool. A few mean a prompt that drafts a letter.

None of that is what we built.

MatterX ships with 38 API endpoints that give AI agents — Copilot, Claude, ChatGPT, or your firm's own — full operational access to your practice management system. Not read-only. Not summarisation. Creating matters, recording time, running conflict checks, preparing invoices, querying trust balances, tracing transactions through the general ledger.

Working infrastructure. Not a roadmap.

Why APIs, not chatbots

Most practice management systems that "support AI" embed a chat window inside the application. You click a Copilot icon and ask it a question within that walled garden. The AI can only do what the vendor coded. You're locked into whichever model they chose.

We went the opposite direction.

The endpoints sit behind Microsoft Entra ID authentication and module licensing. Once authenticated, an AI agent can do anything a human user can do through the UI. The practice management system doesn't care which AI is calling — it authenticates the request and processes it.

38 endpoints across seven systems

We designed the Copilot API surface to cover every system in the client lifecycle. Click any system below to see the AI capabilities it unlocks, the specific endpoints, and real-world examples.

The endpoint count isn't the point — coverage is. An AI that can read matters but can't create time entries is a reporting tool. One that creates time entries but can't check trust balances is a liability. The 38 endpoints exist because that's what it takes to cover the full operational surface.

What it actually looks like

Theory is cheap. Here are three real scenarios showing what happens when an AI agent has full API access. The API calls shown are actual endpoint patterns and response structures.

Under the hood

Time entry scenario — the agent parsed a natural-language week description, looked up matter numbers and billable rates, and batch-created six entries in one API call. A human would open the time entry page six times. The API call takes milliseconds.

Conflict check — two names in, structured JSON out. The system searches customers, contacts, matter contacts, and related entities. Match confidence levels (exact, partial, fuzzy) give the agent data it can reason about — not a PDF to interpret.

Billing review — multiple endpoints combined: WIP entries for unbilled work, budget alerts for matters approaching caps, and enough context for a nuanced recommendation. Invoice two now, hold the third for a client conversation.

Trust accounting and AI

Trust accounting is where firms get nervous about AI. Rightly so — in Australian jurisdictions, trust errors trigger regulatory investigation.

An agent that monitors trust accounts and alerts you to anomalies is a compliance tool. The API makes it possible. The posting engine makes it safe.

{
  "performed": true,
  "matchCount": 2,
  "matches": [
    {
      "entityType": "Customer",
      "entityNo": "C-1847",
      "name": "Michael Chen",
      "matchField": "Name",
      "confidence": "Exact"
    }
  ],
  "matchesByEntity": {
    "Customer": 1,
    "MatterContact": 1
  }
}

Rate management: a practical example

MatterX's rate hierarchy — project override → client override → resource default → system fallback — is exposed through ratePreviews. An agent looks up the effective rate before creating a time entry.

Without this, associates create entries without knowing the rate. It gets applied at invoicing time, weeks later. If wrong — a missed client discount, a mid-engagement rate change — it's caught during billing review, when it's expensive to fix.

With the API, the rate lookup happens at creation time. If $375/hour appears on a matter that should be fixed-fee, the agent flags it immediately.

Impact by firm size

5–15 fee earners — time entry recovery. A conversational agent reads calendar and email, suggests entries for billable time that would otherwise go unrecorded.

15–30 fee earners — billing workflow compression. Surface unbilled WIP, flag budget overruns, prepare draft invoices. Plus conflict checking at the scale where manual searches become unreliable.

30–50 fee earners — conversational analytics. Utilisation tracking, realisation analysis, matter profitability — without a BI implementation or a data analyst.

The Microsoft ecosystem advantage

MatterX's API is one layer in a stack that includes Copilot across Outlook, Word, Excel, and Teams. The compounding effect is significant.

Copilot in Outlook reads and summarises emails. MatterX's API creates matters and links contacts. An agent combining both reads an engagement letter, extracts client details, runs a conflict check, and creates the matter — from a single prompt.

We've been running this workflow in production using the Business Central MCP server. Creating matters from email. Generating time entries from calendar activity. Filing documents based on content analysis. Production workflows, not demos.

What's next

The 38 endpoints cover core operations. Three capabilities in development will make the integration significantly more powerful.

Document intelligence — combining SharePoint document metadata with Business Central matter context. An agent that files, classifies, and tags documents based on content analysis. SharePoint's AI capabilities plus MatterX's matter context.

Workflow triggers — agents subscribing to events (new matter, low trust balance, budget exceeded) for proactive action instead of waiting to be asked.

Multi-matter analysis — aggregate views across the entire practice: utilisation trends, revenue forecasting, client lifetime value. Conversational access to analytics that traditionally requires a BI implementation.

The API surface grows with the platform. Every feature gets an endpoint. Every endpoint makes the AI more capable. The compounding effect is the strategy.