Harvey
Each of these is a real decision Harvey is working through, written so you can make the call yourself — incomplete information, someone who pushes back, no answer key.
Harvey is an enterprise legal AI platform for research, drafting, document review, and workflow automation, used by law firms, corporate legal departments, US federal agencies, and professional services teams, including in-house counsel at private equity firms and Fortune 500 companies. It is an integrated SaaS product on web, desktop, and mobile, deeply tied into existing legal work environments, embedded in Google Docs, MS Word, Outlook, and Slack, plus an API. Harvey Deals is built off its Benchmark acquisition. It processes trillions of tokens and millions of daily requests, analyzes 178M+ documents monthly, works with 150+ asset management firms, and has employees in a dozen countries. Its AI platform has an agent harness, custom agent and workflow building, tool calling, retrieval over petabyte-scale document corpora (ingestion growing 10x every six months, data collection across 50+ jurisdictions), memory and context management, personalization and tone, a thread experience, and evals that turn model quality into product quality. It runs multi-cloud on Azure and GCP and integrates with AI model providers OpenAI, Anthropic, Azure OpenAI, Fireworks, and Baseten.
Your first one is free. No card.
21 live right now.
The rest
Contain unsupported redlines before the renewal review locks in
Complex agreements are receiving approved-position redlines that counsel cannot support from the cited record. You must weigh useful precedent-based judgment against a review path that lawyers can verify before relying on it.
Speed long reviews while lawyers stay in the contract
Reviews for contracts over 80 pages are arriving after counsel has started manual markup. You have to weigh smaller context and caching against citation-backed playbook findings.
Decide if the vendor cutover is ready to sign
Forty-one of 293 supplier obligations rest only on generated summaries, and seven critical dependencies are unvalidated 30 days out. You have to weigh the committed cost date against evidence you cannot test in production.
Decide whether the deletion promise can go live
Six weeks before launch, 34 of 500 deletion requests lack conclusive proof for at least one recovery or workflow path. You have to judge whether the evidence is strong enough to turn the commitment on as promised.
Decide whether the ethical wall is ready
The wall held in 99.4% of tests, but six traces leave the lineage in doubt across add-in exports and automated outputs. You have to decide whether that is enough to sign off on a control you cannot take back.
Set launch criteria for matter workspaces
Mappings agree 97.8% of the time, but 34 of 1,540 need a person to settle the matter boundary, and the activation date is a month out. You have to say what readiness means and which metrics show that adoption is real.
Split event data by region before cutover
The residency commitment you signed forces a one-way split of event data, and 9% of derived datasets still lean on labels a machine guessed. With six weeks left, you weigh completeness by region against keeping analytics intact.
Show reviewers when a clause has changed
In pre-release testing, only 61% could tell a generated clause from one they had materially edited under time pressure. You settle the review states now, because the release is hard to unwind once reviewers learn it.
Lock the look before the rollout ships
Only 43% of lawyers see the workflow materials as belonging together, though trust scores stay strong. You have to weigh a conservative legal look against a more distinctive one, on materials that are costly to reissue.
Decide how to tie costs to each matter
Only 62% of billed activity can be traced to a matter, against the 95% threshold just signed. You have to pick an allocation method you can defend, knowing the machine-made classifications need paid human checking.
Ship the clause workflow before the policy window
Reviewers disagree on 14% of citations pulled from legacy policy documents, and the customer cannot move its refresh cycle. You have to sequence the rollout and set review gates so nothing unverified reaches legal work.
Finish the secrets move before support ends
Rehearsals hit 99.97% retrieval success, but 0.6% of credentials still have lineage gaps that only application owners can clear. You have to decide what evidence is enough before cutover, when rollback gets expensive.
Settle the launch rules for Outlook rewrites
In testing, 11 of the 60 rewrites dropped or reworded legal-advice disclaimers and negotiation reservations. The launch date is fixed, so you decide now what gets preserved, flagged, or sent for review.
Decide if the new route is safe to cut over
The new route matches the old one on aggregate scores, but dropped instructions in 8.2% of the long drafts reviewed. Your capacity reservation expires soon, so you weigh that timing against the lawyer review this needs.
Settle on one way to review redlines
The same redlines get judged differently in Word and on the web, a gap of 18 points across 480 sets. You have to pick the review model and the acceptance bar now, and you cannot undo it later.
Decide where the assessment sends each prospect
Reviewers disagree on 22% of the routes the assessment recommends, and leadership locks the campaign this month. You have to weigh the launch date against routing you can defend in public.
Move the audit trail before storage runs out
A rehearsal matched only 92.4% of the event chains that matter, and storage runs out in nine weeks. You have to decide what evidence you can guarantee before the cutover makes the design hard to unwind.
Move matters to regional indexes without losing citations
Rehearsals show the routing works, but recall of sources varies by 0.9% and provenance records are incomplete. You have to weigh the contractual launch against how much you verify before each matter moves.
Win the three-year deal without overpromising
The buyer ranks you first on quality but two redlines are open, and the incumbent is offering a bundled renewal and implementation guarantee. You have to protect the commercial model without promising what you cannot verify yet.
Close the regional gap in draft acceptance
Verified first-draft acceptance varies by 31 points across regions, and the legacy intake channels retire at fiscal-year start. You have to weigh the speed legal wants against the review evidence compliance needs.