Spellbook's review agents run about 530,000 contract reviews a month for lawyers
Spellbook runs about 15 Claude agent configurations inside Microsoft Word to spot issues, apply each legal team's playbook and suggest edits. It reports cutting an agreement from roughly 10 hours of lawyer work to about one.
- Organization
- Spellbook
- Industry
- Legal software
- When
- 2026
- Evidence
- Vendor case study
- Agents and tools
- ClaudeMicrosoft Word
Reported outcome
About 530,000 contract reviews a month. Spellbook reports a 10x cut in time per agreement, from roughly 10 hours of lawyer work to about one.
- about 530,000Contract reviews per month
- roughly 10 hours → about 1 hourTime per agreement
- more than 700,000Chat messages from lawyers per month
- 5,000 across 80 countriesCustomers
Caveats
Published by the company that sells the agent. Treat numbers as marketing until confirmed elsewhere.
- Published by Anthropic, which sells the model, using Spellbook's own numbers. Both companies benefit from the story.
- The page gives no method for the 10x figure, and no accuracy or error rate.
- 530,000 counts reviews run, not reviews that were correct or accepted.
- Spellbook measures how many suggestions lawyers accept, but the page does not share that rate.
What the agent does
Spellbook works inside Microsoft Word. Its agents spot issues, suggest negotiation language and revise agreements. Legal teams write their review standards into playbooks, and the agents apply those standards to each incoming contract.
About 15 Claude agent configurations run on the review surface alone, with stronger models on the hardest subtasks and faster models elsewhere. Edits stay within the paragraphs the task specifies.
What happened
Spellbook reports about 530,000 contract reviews a month and more than 700,000 chat messages from lawyers. It reports that the time to complete an agreement fell from roughly 10 hours of lawyer work to about one.
Its CEO says the team measures how many suggestions users accept, because that shows whether lawyers find the output useful.
What to copy
- Write down your review standards first. The agent is only as good as the playbook.
- Limit edits to named paragraphs so a change cannot spread through a document unseen.
- Measure acceptance of each suggestion, not only how many ran.
Starter prompt
Written by us, not by Spellbook or Anthropic. A lawyer must review the output.
Our playbook: {playbook}
Contract clause: {clause}
Compare the clause with the playbook.
Reply with: MATCHES, or DEVIATES plus the playbook rule it breaks.
If it deviates, suggest replacement text for this clause only.
Do not change any other clause. Do not give legal advice beyond the playbook.
Sources
- Spellbook runs 530,000 contract reviews a month with Claude — Anthropic, 2026
Added Nov 9, 2026 · Sources last checked Nov 16, 2026.
Get new use cases weekly
One short email. Sources included.
The newsletter is paused. Past issues stay in the archive.
Related use cases
EvenUp reports personal injury drafts in about 30 minutes, down from 8 to 15 hours of skilled work
Drafting that took 8 to 15 hours of skilled work per case now returns in about 30 minutes, per EvenUp. Discovery responses went from 5 hours to 30 minutes.
Freshfields gave Claude to 5,700 staff and reports a 500% usage rise in six weeks
Freshfields reports Claude is available to 5,700 employees and that adoption and usage rose about 500% within the first six weeks.
TRM Labs' security agent handled about 10,000 interactions in 30 days and closed 17% of alerts alone
In the past 30 days: about 10,000 interactions, 45% of helpdesk requests resolved with no ticket, and about 238 security alerts (17%) closed without paging anyone.