AI for Legal Document Review: What Actually Works in 2025
Big Law and AmLaw 100 pilots are showing 40-70% time savings. Here's the reality check.
Legal is one of the few professional services categories where AI is genuinely changing the unit economics. AmLaw 100 firms are reporting 40–70% time savings on specific document review tasks. The number is real — but the path to it has more nuance than the press releases suggest.
Where AI is delivering for legal teams
- Contract review: 40–70% time savings with proper review workflows. First-pass clause extraction, risk flagging, comparison to firm playbook. Always paired with attorney sign-off.
- Litigation discovery: 50%+ time savings on first-pass relevance review. Embedding-based search + LLM screening replaces the bulk of doc review associate hours.
- Patent prior art search: Dramatic improvements with embedding-based retrieval over keyword search. Still requires patent attorney verification, but the haystack shrinks faster.
- Deposition prep: Auto-summarized exhibit binders, key admission extraction across thousands of documents.
Where it doesn't work yet
- Novel legal arguments — the model produces plausible-sounding text that doesn't survive opposing counsel.
- Jurisdiction-specific nuance — state-by-state variation defeats general-purpose models.
- Credibility judgments — "is this witness lying" is not a vision-language model task.
- Anything requiring confidentiality of draft strategy (use of cloud LLMs for case strategy is a privilege risk).
The compliance layer
Firms shipping AI in production typically have: a vendor-managed BAA-equivalent for legal work, on-prem or VPC-isolated deployment for the most sensitive matters, explicit client notice and consent for AI use, and a documented partner-level review for any AI-touched work product.
The unit economics shift
The billable hour model breaks when junior tasks compress from 8 hours to 30 minutes. Forward-thinking firms are moving to fixed-fee for specific work types where AI predictably collapses the time. Holdouts are quietly losing the work to firms that price for outcomes.
Bottom line
If your firm hasn't piloted AI on contract review or e-discovery this year, you're behind. The technology works; the operational discipline to use it well is the differentiator.