Mid-sized law firms (5-50 attorneys) are in a uniquely good position with AI in 2026, large enough for the productivity to matter, small enough to actually ship rollouts without enterprise paralysis. This is the playbook for the firms doing it well.
AI changes legal margin math fundamentally. A contract review that took 4 hours now takes 1. If billed hourly, that's a revenue cut. If priced by deliverable, that's margin expansion.
Firms that have moved more of their book toward fixed-fee, subscription, or value-priced work are capturing the AI upside. Firms holding to pure billable hours are watching margins compress.
The strategic move for most mid-sized firms: progressively migrate the book toward fixed-fee work for predictable, repeatable matters (NDAs, employment agreements, standard contracts, simple M&A). Hourly remains for novel, judgment-heavy work. This shift takes 12-24 months.
Build a Project loaded with the firm's 30-50 most-used clauses and 100 past contracts. First-draft contract generation drops 60-80%. Highest single-workflow ROI we see at firms.
Counterparty drafts run through a Review Project flag deviations from firm playbook. Faster, more consistent, catches more.
First-pass document review for relevance and privilege. Reduces hours; humans still do the substantive judgment passes.
Synthesize case law and statute research into structured memos. Use traditional research tools (Westlaw, Lexis) for the underlying research; use Claude for synthesis. Do NOT use Claude as a primary legal research tool, citation accuracy matters too much.
Status updates, draft emails, meeting summaries. High-frequency lower-stakes work that nonetheless eats partner time.
Track realization rate (collected vs billed) by practice area pre and post AI rollout. Firms doing this well see realization rise 8-15% in the first year because AI-assisted drafts have fewer write-offs.
Track profit per attorney, not just revenue per attorney. AI's biggest gift to firms is profit margin expansion, not top-line growth.
Traditional model: juniors do hours of foundational drafting and learn judgment from the repetition. AI removes the hours but should not remove the learning.