How EM3 Law brought every Claude surface under one governed, auditable view
One platform now tracks, attributes and audits the firm's Claude use across web, desktop, Microsoft 365 and its own legal assistant. Attorneys learn good practice in the same place.
4 to 1
Separate Claude surfaces now reporting into one view
28%
Of monthly Claude spend that was invisible, now attributed
200K+
Activity records under governance since June 2026
Minutes
To produce each attorney's monthly AI usage report
Client
EM3 Law (Maxson Mago & Macaulay)
Industry
Legal services, US law firm
Built on
Claude Enterprise, Cowork, Claude in Microsoft 365 and the Claude API
Status
Live in production
Bring us the workflow you dread
One session. We map where the hours actually go and show you what we would build.
Book a working sessionThe Challenge
EM3 Law is a US firm working across insurance policy review, corporate and M&A, healthcare and commercial transactions. It adopted Claude broadly: on the web, on the desktop through Cowork, inside Microsoft 365, and alongside an internal legal assistant of its own. Leadership, legal operations and IT had no way to govern that use as a whole.
No single view
Each surface reported separately, or not at all.
No cost attribution
One monthly invoice, with no breakdown by attorney, matter or surface, in a firm that bills onward.
Adoption was guesswork
IT's list of installed machines didn't match real usage, and web-only users were invisible.
No governance signal
No alerts on unusual spend and no record an auditor could be walked through.
No easy answers
Any ad-hoc question about usage meant someone writing SQL.
No structured enablement
Attorneys had access to Claude but no firm-specific guidance on good and ethical use in legal practice.
What Attri Built
Attri delivered three connected pieces covering control, capability and enablement. That combination is why adoption stuck instead of stalling after rollout.
OpenWatch, AI governance and observability
One view over every surface where the firm uses AI: cost over time by model, user and matter, budget tracking, and sessions drillable to the individual message. Claude agents run on demand and on a schedule, each with a hard cost cap: an assistant that answers usage and cost questions in plain English, read-only, and can file a ticket; an anomaly agent that flags cost spikes, unusual usage and policy violations, with citations back to the source records; and a report agent that produces each attorney's monthly AI usage statement and improvement report as a PDF.
Mike, an AI legal document assistant
Attorneys upload DOCX and PDF files and work over them by matter, with versioning, tracked-change accept and reject, tabular extraction and reusable workflows. Usage is recorded per model provider, so Mike is governed inside OpenWatch too.
The Attorney AI Learning Library
A firm-only site with curated Claude adoption and ethics guidance for legal practice, plus recordings of the recurring Office Hours sessions.
Built on Claude
Claude Enterprise
Attorneys on the web, including Projects.
Cowork and Claude Desktop
Local work with file and tool access.
Claude in Microsoft 365
Work inside Office documents.
Claude API
Powers OpenWatch's own governance agents. Models in production: Claude Opus 4.6, Opus 4.7, Opus 5 and Sonnet 4.6.
Claude surfaces
Every surface reports into one warehouse in the firm's own Azure environment.
- Claude Enterprise (web)
- Cowork and Claude Desktop
- Claude in Microsoft 365
- Mike, the firm's own assistant
- read
OpenWatch agents
on the Claude API, each run on demand or on a schedule with a hard cost cap
- Usage and cost assistant
- Anomaly agent
- Report agent
- draft
Findings that need a person become tickets
Including data-subject requests, erasure and retention sweeps, each with a lifecycle and history.
Oversight becomes an auditable process
Not a dashboard someone has to remember to check.
People act on what the agents surface; the agents never change the underlying data.
The Outcome
| Metric | Before | Now |
|---|---|---|
| Visibility | Four surfaces, each reporting separately or not at all | Desktop, web, Microsoft 365 and the firm's own assistant in one view with a common cost model |
| Cost attribution | One invoice, no breakdown | Cost per attorney, per matter and per conversation |
| Web-only usage | Absent from desktop tooling | Fully attributed, 28% of monthly spend |
| Governance | Spend issues found when the invoice arrived | Scheduled scans surface spikes and policy violations as cited findings |
| Questions about usage | Needed someone to write SQL | Asked in plain English, in the context of the page |
| Attorney reporting | Manual reconciliation | Generated from the dashboard in minutes per attorney |
| Adoption | IT's install list didn't match reality | Directory reconciled against observed usage |
The firm now sees, attributes and audits all of its Claude use in one place.
The web portion alone was 28% of monthly spend and had been effectively invisible.
Figures are from the platform’s own data as of 29 Sep 2026. The platform handles about 2,700 activity events a day.
Why It Worked
Control, capability and enablement together
Governance shows what is happening; the Learning Library shows attorneys what good use looks like.
Numbers the firm can defend
Every figure traces back to source records, which is what makes it usable for client billing.
The governance agents follow their own rules
Each run is cost-capped and records its own cost, steps and tool calls.
Work stays in lawyers' formats
Mike keeps AI edits as reviewable tracked changes in the documents attorneys already use.
Security & Compliance
Ready to govern Claude across your firm?
See how Attri gives law firms one auditable view of every Claude surface, with cost attributed to the matter it was worked on.