Harvey AI in Big Law — What the Deployment Data Actually Shows

Harvey AI raised a $300M Series E in June 2025 at a $5B valuation, then reportedly reached $8B by December 2025, and TechCrunch reported in February 2026 that it was raising again at $11B. ARR crossed $100M in August 2025. The platform now claims more than 100,000 lawyers across 1,300 organizations, roughly half of the Am Law 100, 500-plus in-house legal teams, and 50 asset managers.

That is the funding story. It is impressive and largely beside the point.

The interesting question is the one nobody in the legal trade press has answered cleanly — what actually changed inside the firms that deployed Harvey? Hours saved on which workflows? What does the work look like now that did not look like that in 2022? And the question Big Law partners actually ask in deal rooms: what does Harvey cost, what does it replace, and is the math defensible to the management committee?

This is the answer, with the published numbers and a few caveats the marketing pages will not tell you.

The Allen & Overy case study — the one everyone cites

Allen & Overy (now A&O Shearman after the May 2024 merger) was the first magic-circle firm to deploy a generative-AI legal tool firmwide. The trial started in November 2022 inside their Markets Innovation Group. By February 2023, 3,500 lawyers had run roughly 40,000 queries against Harvey. The announcement broke the legal-IT trade press for a week — David Wakeling, the partner running the project, became the most-quoted partner in legal tech for the rest of 2023.

The follow-on numbers Harvey and A&O have publicly shared:

• 4,000+ lawyers across 43 jurisdictions on the platform within the first year of firmwide rollout
• Average 2-3 hours saved per week per lawyer who uses the tool
• 30% reduction in contract review time on the workflows where Harvey is the primary tool
• 7-hour average savings on complex document analysis (typically the long-form due diligence work)
• One in four lawyers using Harvey daily; 80% using it at least monthly

Take those numbers at the value the firms put on them — they are firm-disclosed, not independently audited. The skeptical reading is that "2-3 hours per week" assumes the lawyer would have done the equivalent work manually rather than redistributed those hours across a different mix of billables. The honest reading is that even at half-discounted those numbers represent real workflow change at a magic-circle firm where partner attention is the binding constraint, and that A&O has not unwound the deployment after three years, which is the strongest signal in the dataset.

PwC — the professional-services beachhead

PwC announced its strategic alliance with Harvey in March 2023, one month after A&O. The press coverage at the time framed it as a Big Four bet on legal AI; the actual deployment is more interesting. PwC's legal services arm covers 4,000+ legal professionals across 100+ countries — a global in-house legal team larger than most Am Law 50 firms. PwC UK has a dedicated published case study with Harvey covering regulatory and compliance analysis and a separate piece on AI-driven legal entity transformation work.

The strategic angle, not visible until late 2025, is that PwC was the bridge into Harvey's tax and accounting expansion. Harvey's post-Series E roadmap explicitly targets tax advisory as the next vertical, and PwC's tax practice is the proof point. This is the same playbook OpenAI's enterprise team has been running in adjacent professional-services markets — start in a high-stakes, document-heavy vertical, expand horizontally into adjacent ones that share the same buyers.

Paul, Weiss and the custom workflow story

Paul, Weiss — an elite Am Law 50 firm — partnered with Harvey to build out custom AI workflows using Harvey's workflow builder platform. The workflow builder is the feature Harvey has been positioning hardest in 2025-2026, and Paul, Weiss is the public reference.

What it actually does: it lets a firm encode its proprietary methodologies — the M&A diligence checklist that has been refined over 30 years of deals, the privacy-review framework, the regulatory-filing template — and run them as repeatable Harvey workflows against new matters. The output is supposed to look like work product the firm would have produced anyway, only faster and more consistently. The pattern is the same one ABA 5.12 audits address on the supervision side — the AI is checking work that the firm already has a defensible procedure for, not generating analysis the firm has no prior position on.

The honest take on workflow builders: they are how Harvey defends its margin against the rest of the legal-AI market. The base capability — query a foundation model against a fine-tuned legal corpus — has been commoditizing for two years. The defensible feature is the workflow layer the firm builds on top, because once that is built and the partners trust it, the switching cost gets steep. Harvey is not selling a model. It is selling a workflow platform with a model underneath it.

Latham & Watkins — the August 2025 firmwide

Latham & Watkins announced a firmwide deployment in August 2025. This was the largest Am Law commitment of the year and the inflection point where the trade press shifted from "is Harvey going to penetrate Big Law" to "which Am Law 100 firm is next." Reported coverage suggests the Latham deployment covers the firm's standard transactional and litigation workflows — contract analysis, due diligence, regulatory research — with custom workflows being built for the M&A and capital markets groups.

Worth noting what Latham did not deploy Harvey for: advocacy, oral argument preparation, novel-issue legal strategy, or client counsel. Those remain partner-only tasks at every firm in the disclosed customer list. The deployment is concentrated where the work is high-volume, pattern-matchable, and producible against a checklist. That is the actual shape of Big Law AI in 2026, and it is consistent across every published case study.

What Harvey actually does — workflow by workflow

Stripping out the marketing copy, Harvey's published use cases cluster into four buckets:

• Contract review and analysis. Extracting key terms across an agreement set, flagging unusual provisions against the firm's standard, surfacing risk language for partner review. This is the most-deployed workflow because it has the cleanest before/after measurement — partner reviews a contract in 25 minutes instead of 90, and the diff is the productivity gain.
• Due diligence. Reviewing commercial contracts, employment agreements, compliance policies, environmental reports in M&A diligence. Document classification, cross-document pattern recognition, red-flag surfacing, report generation. This is where the "7-hour average savings on complex document analysis" number comes from at A&O.
• Legal research. Querying across legal, regulatory, and tax sources with a fine-tuned model. The Harvey + LexisNexis partnership that took effect in 2025 brought primary law content into the platform, which is the integration that made this workflow defensible against pure-research tools like Lexis+ AI directly. Cost: roughly $400-600 per lawyer per year on top of base Harvey pricing.
• Workflow agents. Firm-built or pre-built agents that chain the above steps together for a specific matter type. M&A diligence agents, regulatory-filing agents, privacy-review agents. The Paul, Weiss custom-workflow story sits here.

What it does not do — and Harvey has been increasingly careful to say this — is replace lawyer judgment on novel issues, court advocacy, or strategy work. The firms that have deployed it have generally kept those tasks human and pushed the gains into the high-volume document-heavy work where the math is provable.

The pricing — what Harvey actually costs in 2026

Harvey does not publish pricing. Multiple secondary sources put enterprise pricing in the range of $1,000-$1,200 per lawyer per month, or roughly $12,000-$14,400 per seat per year on the base bundle. Premium bundles with Lexis primary law content and advanced workflow agents run $15,600-$16,800 per seat per year.

Minimum commitments typically start at 25-50+ seats, so the annual contract floor is $300,000 and the realistic mid-market figure is $500,000-$2M+ for a working deployment. For an Am Law 100 firm with 500-2,000 lawyers, the all-in is well into seven figures annually before professional services and workflow-builder consulting.

A&O ran the pilot with 3,500 lawyers. Latham has roughly 3,500 lawyers globally. The math on those deployments, even with negotiated enterprise discounting, is meaningful. Harvey's $100M ARR figure as of August 2025 implies somewhere on the order of 100-200 enterprise accounts at this price band. That tracks against the "1,300 organizations" claim once you assume the long tail is in-house legal teams and smaller corporate accounts paying a fraction of the magic-circle rate.

The alternatives — and where they actually win

Three serious alternatives in this market, each playing a different game.

CoCounsel (Thomson Reuters). Thomson Reuters acquired Casetext for $650M in cash in August 2023 specifically to fold CoCounsel into the Westlaw/Practical Law stack. The integration story is "you already pay Thomson Reuters for primary law content; the AI assistant rides on the same contract." CoCounsel hit 1 million professional users across 107 countries by February 2026 — a far larger user base than Harvey by raw count, because it includes solo and small-firm Westlaw subscribers, not just enterprise deployments. The Stanford 2024 hallucination study found Westlaw AI-Assisted Research hallucinated at 33% on certain query types and Ask Practical Law was accurate only 18% of the time, which is the asterisk on the user-count claim. Harvey's published numbers on hallucination are not directly comparable (no independent Stanford-style audit) but Gabriel Pereyra (Harvey co-founder) has acknowledged the platform is not at zero, and the workflow-builder design is explicitly the answer to that — constrain the model to firm-validated procedures rather than open-ended queries.

Spellbook. Self-serve, Word-integrated contract review tool priced at $99-300 per user per month. This is the small-firm answer to Harvey. A five-lawyer firm pays $18,000-36,000 per year all-in and gets a contract-drafting assistant that lives inside Word. It does not have Harvey's depth on research, due diligence, or custom workflows, but it does not need to — the buyer is a solo practitioner or a 5-30 person firm where the contract-drafting workflow is the binding constraint and the firm cannot absorb a six-figure procurement cycle. Spellbook is what most firms reading this article should actually be evaluating instead of Harvey. The math is right-sized for the matter.

Robin AI. UK-based, originally focused on contract review for in-house legal teams and PE/VC funds. Priced in the same band as Spellbook for self-serve, with enterprise tiers for larger deployments. The differentiation is contract-focused depth — Robin is built for the "I review 200 NDAs and MSAs a month" workflow, not the full firmwide-research-and-drafting platform Harvey ships.

Lexis+ AI / Westlaw AI. Bundled into existing Lexis or Westlaw enterprise subscriptions. The buyer is the firm that already pays $X per lawyer per year for primary law access and wants the AI assistant on the same line item. The Stanford hallucination numbers above are the asterisk. This is the path of least procurement resistance, not the path of best capability.

Where Harvey is going — tax, accounting, and the platform claim

The post-Series E narrative Harvey has been pushing — and which the PwC alliance was always strategically positioned for — is expansion into tax and accounting as the next professional-services vertical. The product logic is sound: tax is high-stakes, document-heavy, regulation-driven, and the buyer (Big Four tax practices) is already partially in the customer base. The execution risk is that tax advisory has incumbents (Thomson Reuters again, plus Wolters Kluwer's CCH) that are more deeply entrenched than the legal incumbents were in 2022.

The platform claim — Harvey as the operating system for professional services rather than just legal AI — is the $11B valuation story. Whether it holds depends on whether the workflow builder layer becomes the durable moat or gets commoditized by the next foundation model release. The Harvey co-founders' bet is that the firm-by-firm workflow encoding work is the moat. The skeptical bet is that the next GPT-class model with native tool use makes most of those workflows reproducible by any vendor with a Lexis API key. Neither bet has been settled in the public data.

The honest take for everyone not running an Am Law firm

If you are a 5-50 lawyer firm, Harvey is not the right tool for you. Not in 2026, not on these unit economics. The minimum commitment alone exceeds your annual technology budget, and the procurement cycle assumes a firm with a dedicated knowledge-management partner, an IT director, and a year of political capital to spend on the deployment. The honest small-firm path is a Citation Verifier for the Mata-v.-Avianca problem, an Engagement Letter Generator for consistency, an ABA 5.12 audit for the supervision posture, and Spellbook (or equivalent) inside Word for contract review. Cost: under $500/month all-in. Defensible posture: better than firms ten times your size.

If you are at an Am Law 100 firm reading this — you already know whether Harvey is on your procurement list. The interesting question is not whether to deploy it but whether the workflow-builder commitment is the right horizontal across your practice groups, or whether you should be pairing Harvey for some workflows with Robin AI for contract-volume work and CoCounsel for research, on the theory that no single vendor wins all three.

If you are a Big Four legal or tax practice — Harvey's strategic alliance posture with PwC is the relevant case study. The deployment math at 4,000+ legal professionals is more favorable than at any Am Law 50 firm because the volume is higher and the workflow standardization is built in. This is the segment where Harvey's $11B valuation story most cleanly defends itself.

The broader signal — the one everyone in legal tech is reading off this dataset — is that Big Law has crossed the chasm on generative AI faster than any other regulated industry. Three years from A&O's pilot to ~50% of the Am Law 100 deployed on a single vendor at a $300K-$2M+ annual ticket. That is unprecedented adoption velocity. The question for 2027 is whether the workflow-builder layer becomes the durable moat or whether the foundation-model commoditization catches up. Either way, the lawyers who learn the workflow are the ones who keep their billable rate intact.