Scaling through embedded engineers
Harvey deploys through embedded engineers who spend six to nine months inside a firm to map workflows and rebuild knowledge bases. These engineers spend one third of their time interviewing partners, associates, and IT staff, one third on data engineering against systems like iManage or NetDocuments, and one third on retrieval and fine-tuning work. The deployment follows a three-phase loop that begins with firm-discovery during weeks one to six. During this phase, engineers conduct 40 to 80 hours of qualitative interviews to map practice groups, document management architecture, and partner-level workflow variants. The second phase, workflow-mapping, runs from weeks four to fourteen and selects concrete tasks like M&A diligence or regulatory memo drafting; this phase produces a workflow specification that names the inputs and the required human escalation points. Finally, the knowledge-base-build phase lasts from weeks eight to twenty-eight as the engineer ingests the firm’s precedent vault into retrieval-augmented generation indexes. Because BigLaw firms operate as federations of partner-owned practices, the engineer must win adoption one partner at a time to ensure the tool succeeds. The work requires firm-by-firm execution and is not a productized solution. A full rollout touches three distinct populations: partners, associates, and knowledge management lawyers. FDEs also manage the technical plumbing required for integration with systems like Aderant and Intapp, ensuring the AI respects the strict ethical walls and confidentiality regimes that govern client matters.
Document intelligence and automated workflows
The platform centralizes document analysis through Vault, a collaborative environment that holds up to 100,000 documents per project. Firms use Vault to load entire deal rooms or contract portfolios. This allows the AI to extract fields and identify clauses across a massive corpus of documents with 97% field extraction accuracy while managing tens of thousands of files per project. This capability supports the 15,000 custom legal workflows users have created through Workflow Builder. The system also uses multi-model routing to select between Claude Opus 4.6, Claude Sonnet 4.5, GPT-5.1, and GPT-5.2 depending on the specific task. Users also benefit from native MCP support, which allows the tool to function as both a client and a server within existing firm toolchains. The 18,000-workflow library provides patterns for M&A due diligence, litigation motion drafting, regulatory compliance, and real estate lease review. Workflow Builder allows firms to codify their specific positions on contract clauses with deterministic three-outcome evaluation. The LexisNexis strategic alliance allows lawyers to verify citations and check case validity without leaving the Harvey environment. The platform also provides horizon scanning, regulatory tracking, and usage analytics to help legal teams manage their workload.
The economic and risk reality
The cost of Harvey makes it a tool for AmLaw 100 firms and large corporate legal departments. The pricing starts at roughly $1,200 per lawyer per month with a twenty-seat minimum and creates an annual floor of $288,000. You should know that a small practice cannot afford this enterprise-grade pricing. The tool can produce incorrect citations and is not a replacement for legal judgment. Every lawyer must verify the output against primary sources to prevent the risk of fabricated authorities. Repsol reported saving 4 to 6 hours per lawyer per week after reaching 96% adoption. However, the steep learning curve requires significant change management, including training lawyers to use Vault and building custom workflows. Implementation requires a dedicated team to manage the transition from manual tasks to AI-driven processes.
The platform meets industry standards through annual SOC 2 Type II and ISO 27001 audits, and it allows customers to set data retention policies. Large organizations like HSBC and DLA Piper use the platform to manage high-stakes work and automate repetitive tasks. DLA Piper expanded its relationship to 5,000 licenses after a pilot phase proved use cases across various practices. The platform offers data residency options in the EU, Switzerland, the US, and Australia. How will firms manage the risk of fabricated citations in high-stakes litigation?
