Harvey

What Harvey does

Harvey is an enterprise AI work platform for law firms, in-house legal departments, tax teams, and professional-services organizations. Its product family covers purpose-built Agents, Vault for bulk document analysis, Knowledge for legal and regulatory research, secure Spaces, Contract Intelligence, and Command Center adoption analytics. Harvey II adds agents that inherit project and matter context and memory of how a team works.

The useful question is whether those capabilities improve a defined workflow. Do not buy from a polished demonstration alone. Use representative work, realistic permissions, and the same acceptance standard applied to the current process.

Features that matter

• Purpose-built legal Agents for multi-step work
• Vault for bulk analysis of document collections
• Knowledge research across legal, regulatory, and tax domains
• Secure Spaces for firms, clients, and co-counsel
• Contract Intelligence and Command Center analytics

These are testable functions, not a promise that every output will be correct. Confirm integrations, exports, regional availability, administrator controls, and limits in a trial account or statement of work. If the system can write to another application, give it the smallest useful permission scope and require approval before consequential actions.

Pricing and total cost

Harvey does not publish numeric self-serve pricing. The requested /pricing route returned HTTP 404, while the homepage directs buyers to request a demo. Request seat minimums, model and document allowances, implementation fees, support terms, renewal increases, data-region costs, and separate Contract Intelligence charges in writing.

Calculate cost per accepted outcome rather than cost per seat. Include implementation, integration work, usage charges, training, security review, support, storage, and human checking. Model expected volume plus a 25% buffer, then ask whether reaching a limit causes a hard stop, slower service, automatic overage, or forced upgrade.

Pros

• Research, document analysis, contracts, and collaboration share one governed platform
• SAML SSO, audit logs, IP allow-listing, and lifecycle controls support procurement
• The vendor reports more than 2,400 legal organizations and 75+ AmLaw 100 firms
• Command Center helps leaders measure adoption

Cons and risks

• No public numeric price makes early cost comparison difficult
• Legal output can omit authority or misstate law and needs qualified review
• Broad rollout requires access design, training, and knowledge governance
• Vendor-reported 25+ monthly hours saved is not a buyer-run benchmark

These limitations should shape the pilot. Keep a named human responsible for every output that can affect learning, health records, legal rights, an external campaign, production data, money, privacy, or security. Verify authentication, roles, audit logs, encryption, retention, deletion, subprocessors, model-training use, incident notification, and data residency before sensitive deployment.

Practical use cases

• Research a legal question and verify every authority
• Extract clauses and obligations from a Vault collection
• Route repeatable diligence through a legal Agent
• Share matter work with outside counsel in a governed Space

Run a two- to four-week comparison with 20 to 50 representative tasks. Record the current baseline and measure minutes per reviewed document, citation validity, material issues missed, lawyer correction time, accepted work-product rate, and annual cost per professional. Include hard cases: missing data, ambiguous instructions, provider timeouts, expired credentials, duplicate retries, unsupported files, and attempts to cross permission boundaries. Review failures by severity and remediation time instead of averaging them away.

Alternatives and recommendation

Relevant comparisons include CoCounsel, DocuSign IAM, GC AI, ChatGPT. Compare each candidate on the same tasks with the same people, inputs, and rubric. Feature counts matter less than accepted-output rate, correction cost, integration depth, governance, and annual cost at real volume.

Choose Harvey when its specialized workflow removes measurable work and its controls fit the risk level. Delay broad rollout if ownership is unclear, paid terms cannot be documented, users cannot explain review, or the pilot fails to beat the current process after correction time. The sound buying decision is a written pilot result with costs, failure examples, security findings, and a named owner.