3,500 lawyers, one AI research and drafting assistant
How law firms and legal departments use AI for research, review and drafting, with the verification that professional duties require.
Figures: Allen & Overy, 2023. Source and caveats below.
01 The problem
Legal work is reading and writing under time pressure. Associates spend long hours on first-pass document review, due diligence, research memos and first drafts, and clients are increasingly unwilling to pay hourly rates for that work.
General-purpose chatbots are not a safe answer. They do not know the firm's precedents, they can leak confidential information, and they invent citations. Courts have sanctioned lawyers for filing briefs that cited cases which do not exist.
02 Where AI fits
Contract review and due diligence
Models review a data room against a checklist, extract key terms and flag departures from the firm's standard positions.
Research with verified citations
Answers retrieved from licensed and internal sources, with every citation linked to the underlying text.
First drafts
Clauses, memos and correspondence drafted from the firm's own precedents and style.
Knowledge management
Semantic search across prior matters, so a lawyer can find how the firm handled a similar issue.
eDiscovery and intake
Classification, privilege screening and summarization of large document sets, and structured matter intake.
03 Reference architecture
The layers we put in place for a typical legal & professional services deployment, from your systems of record up to the people who review the output.
- L1Document management (iManage, NetDocuments)
- L2Ethical walls and matter-level permissions
- L3Private retrieval index per matter
- L4Drafting and review models
- L5Citation verification step
- L6Lawyer review and audit trail
04 Real-world implementation
In February 2023 the international law firm Allen & Overy (now A&O Shearman) announced it had deployed the generative AI assistant Harvey to more than 3,500 lawyers across 43 offices, working in multiple languages on drafting, research and due diligence. During the trial that preceded the rollout, its lawyers asked around 40,000 queries in the course of their day-to-day client work.
Source: Allen & Overy press announcement, February 2023. Allen & Overy is not a 925 Techies client.
What to keep in mind: The firm published adoption figures, not time or cost savings, and stressed that output must be reviewed by a lawyer. The 2023 sanctions in Mata v. Avianca, for a brief with fabricated citations, show why verification cannot be skipped.
05 Business impact
Leverage
First-pass review and drafting take a fraction of the time, so lawyers handle more matters or spend longer on strategy.
Fixed-fee margin
Lower cost of production makes alternative fee arrangements profitable rather than risky.
Turnaround
Faster answers and drafts for clients, which matters most in deals and disputes on tight timetables.
Knowledge retention
The firm's accumulated work becomes searchable and reusable, instead of sitting in closed files.
06 Implementation roadmap
- Weeks 1–3
Assess
Choose a practice group and a task. Review confidentiality obligations and client consent requirements.
- Weeks 4–8
Prove
Run the tool on closed matters and compare its output with what the lawyers produced.
- Weeks 9–18
Build
Integrate with the DMS, enforce ethical walls, add citation checking, train the practice group.
- Ongoing
Operate
Monitor usage and accuracy, add practice areas, update firm policy as bar guidance develops.
07 Guardrails
- Client data stays in a private tenant and is never used for training
- Every citation is checked against the source
- Ethical walls enforced at retrieval
- A lawyer reviews and takes responsibility for all output
08 Services involved
RAG & Enterprise Knowledge
Answers grounded in your documents, with a citation on every claim and access rules respected.
Small Language Models (SLMs)
Compact models fine-tuned for one job, running in your environment, at a fraction of the per-task cost.
AI Security & Governance
Guardrails, red-teaming and policy so your AI stands up to auditors, regulators and attackers.
Start here
Exploring AI in legal & professional services?
We can run a proof of concept on your own data in about four weeks, measured against the KPIs above.
