What Legal AI Really Costs: Seat Licenses, Usage Caps and a 60-Lawyer Firm's Bill
The seat price is only part of the bill. We cost a hypothetical 60-lawyer firm over three years under three buying models, and the cheapest quote doesn't produce the cheapest total.
Why the Per-Seat Quote Is the Wrong Number to Compare
Ask a firm what its legal AI tool costs and you'll hear a per-seat monthly price. In the worked example below, licenses make up only about 40–65% of the three-year spend. The rest goes on seat minimums, AI uplifts on research subscriptions, training hours nobody bills, security review and the internal time it takes to sustain adoption.
The aim here is to replace the seat price with a three-year total cost of ownership you can put in front of a management committee. We build it for a hypothetical 60-lawyer firm under three buying models. The firm is invented. The cost categories are not. Every figure is a typical 2026 market range that varies by contract, region and negotiating position, so confirm it with vendors before you budget.
Legal AI pricing is hard to compare because the market uses three different commercial structures:
- Legal-specific platforms sold per seat, usually with a minimum seat count or annual minimum spend.
- Research-embedded assistants, sold as an uplift on an existing research subscription.
- General enterprise assistants priced per user, by consumption, or as a platform fee plus usage.
Each one pushes cost into different budget lines. That's why a quote that looks cheap often isn't once everything is added up.
Typical 2026 Price Ranges by Product Type
Most major vendors don't publish enterprise prices. The ranges below are indicative USD figures based on how these products are usually packaged. Local pricing, bundles and currency terms differ by market. Ask for a written quote and don't rely on a list.
| Product type | Examples | How it's typically priced | Indicative 2026 range |
|---|---|---|---|
| Legal-specific AI platform | Harvey, Legora | Per seat on an annual or multi-year term, often with a seat floor or minimum spend | Roughly $150–$500 per seat per month; minimums of 20–50 seats or $50k–$250k+ a year are common |
| Research-embedded assistant | CoCounsel (Thomson Reuters), Lexis+ AI | Uplift on the Westlaw, Practical Law or Lexis subscription, or a higher bundled tier | Often 15–50% on top of existing research spend, or $75–$300 per user per month where sold per user |
| General enterprise assistant | Microsoft 365 Copilot, enterprise chat platforms | Per user per month, or a committed spend plus consumption | $25–$60 per user per month for seats; usage commitments often $25k–$150k a year |
Harvey's pricing is negotiated case by case. Expect seat minimums and multi-year terms, with price depending heavily on firm size and the number of practice areas in scope. CoCounsel and Lexis+ AI costs are best read as part of your research contract, so your leverage sits at the renewal of the underlying subscription, not in the AI line item.
Jurisdictional coverage varies too. The depth of primary law behind each assistant differs across England and Wales, India, EU member states, Singapore and Brazil. Test against your own practice jurisdictions before you compare prices.
The Hypothetical Firm and the Assumptions Behind the Model
Hypothetical example. Say a 60-lawyer full-service firm has 15 partners and 45 associates and counsel. It already pays for a major research subscription and runs a standard document management system. It wants a three-year view before signing. These assumptions apply to all three models:
- Escalator: licenses rise 7% a year at each anniversary. Multi-year contracts commonly carry increases in the mid-single digits unless you negotiate a cap.
- Training time: non-billable hours valued at $200 per hour. That's an internal opportunity cost, not the billing rate. Year one is 6 hours per new user (8 for pilot users, who also give structured feedback). Later years are 2 hours per user for refreshers and new features.
- Security and procurement review: $25,000 one-time for internal IT, risk and outside review time (vendor questionnaire, data processing terms, pen-test summary), plus $5,000 a year for re-review. The usage-based model starts at $35,000 because model choice and data residency need extra scrutiny.
- Research content add-on: an AI-enabled research tier adds $25,000–$30,000 a year to the existing subscription.
- Internal ownership: a knowledge or innovation lead spends part of their time running adoption, valued at $100,000 fully loaded per FTE.
Swap in your own numbers. The structure carries over to other firms even where the inputs don't.
Three Buying Models Costed Over Three Years
Model A: firm-wide seats
All 60 lawyers get a legal-specific platform at $250 per seat per month. Licenses come to $180,000 in year one, $192,600 in year two and $206,082 in year three. Add the $30,000 research uplift, training, security review and 0.2 FTE of ownership, and year one costs $327,000.
Model B: a pilot cohort, then expansion
The firm wants 15 pilot users, but the vendor's minimum is 20 seats. Because the commitment is smaller, the per-seat price is higher at $300 a month. The firm pays for 20 seats in year one ($72,000) and expands to 30 seats in years two and three. Ownership is 0.15 FTE. The research uplift is priced on firm-wide entitlement, so it only falls to $25,000 a year. That's the trap in partial rollouts: the content cost barely shrinks when the rollout does.
Model C: usage-based enterprise assistant
A general enterprise assistant goes firm-wide with a $60,000 annual commitment. Overage is $20,000 in year two and $35,000 in year three as usage grows. The assistant isn't grounded in authorities, so the firm still pays the $30,000 research uplift. It also spends $40,000 one-time on configuration: a document management connector, precedent prompt libraries and access controls. Training runs 7 hours per lawyer in year one because users have to learn to work around a general-purpose tool, and ownership is 0.3 FTE.
| Three-year cost line | A: Firm-wide seats | B: Pilot cohort | C: Usage-based |
|---|---|---|---|
| Licenses / platform and usage | $578,682 | $311,209 | $235,000 |
| Research content add-on | $90,000 | $75,000 | $90,000 |
| Integration and configuration | Included | Included | $40,000 |
| Training time | $120,000 | $60,000 | $132,000 |
| Security and procurement review | $35,000 | $35,000 | $45,000 |
| Internal ownership | $60,000 | $45,000 | $90,000 |
| Three-year total | $883,682 | $526,209 | $632,000 |
| Year-one cash | $327,000 | $161,000 | $279,000 |
| Cost per active lawyer-year | $4,909 (180 lawyer-years) | $7,016 (75 lawyer-years) | $3,511 (180 lawyer-years) |
Model B has the smallest total and the highest cost per active user, and for most of the term half the firm has no tool. Model C is the cheapest per lawyer, but its usage line is a forecast, not a price. If overage runs at twice the expected level, the three-year total rises to about $687,000. Neither the extra tool nor the lower sticker price answers the question of whether lawyers will use it.
Where the Bill Climbs: Caps, Minimums and Escalators
The worked example assumes clean contracts. Real ones rarely are. Five clauses do most of the damage to a generative AI budget for a law firm.
- Seat floors that only move up. Many multi-year deals let you add seats mid-term but not remove them until renewal. If adoption stalls at 40% in Model A, the firm still pays for all 60 seats, and cost per active lawyer-year rises from $4,909 to about $12,300.
- Undefined "fair use" or premium-feature caps. Some tiers meter the expensive features: long-document review, multi-step research agents, bulk extraction across data rooms. Ask what happens at the cap. Is the feature throttled, billed per unit, or does it trigger an upgrade conversation?
- Uncapped escalators. A 7% annual uplift adds roughly 7% to Model A's license line over three years compared with flat pricing, about $38,700. Push for a cap tied to an index, or a flat price across the term.
- Co-terminous bundling. When the AI tier is tied to your research renewal, leaving one can mean losing pricing on the other. Ask for the AI component to be priced separately in the order form, even if it's sold as a bundle.
- Consumption without alerts. For usage-based contracts, require monthly usage reporting, spend alerts at 50%, 75% and 90% of the commitment, and an agreed overage rate in writing.
One check that isn't strictly about price: confirm the contract says firm and client data won't be used to train shared models, and that retention periods are stated. A clause that fails your clients' outside counsel guidelines will cost far more than any discount saves.
How Region Changes the Number: Tax, Residency and Currency
Most vendors quote in USD. The figure that lands in your management accounts can look quite different.
Indirect tax. Software subscriptions usually attract VAT or GST, depending on how the supply is treated locally. Standard rates run from about 5% in parts of the Gulf to 20% or more in the UK and many EU states, with India's GST on software typically at 18%. Firms that can recover input tax may pass it through. Firms that can't, for example where some of their supplies are exempt, should budget it as a cost.
Withholding tax. In several jurisdictions across South Asia, Latin America, Africa and the Middle East, payments to foreign software vendors can carry withholding tax. If the contract requires you to gross up, the effective price rises. Have a tax adviser read the payment clause before you sign.
Data residency. Firms subject to the GDPR or UK GDPR, India's DPDP Act, Saudi Arabia's PDPL or client-imposed residency rules may need regional hosting. Some vendors charge a premium for in-region processing or offer it only on higher tiers. Others don't offer it in every region. That limits which models are viable in practice, whatever the price.
Currency. On a three-year USD contract, FX risk sits with you. A 10% currency move over the term can matter as much as the escalator. Ask whether local-currency billing is available, or budget at a conservative rate.
None of this is tax or legal advice. Treatment depends on your entity structure and jurisdiction, so confirm it with a qualified tax adviser and your data protection officer.
A Decision Rule for Firms of 30 to 100 Lawyers
For most mid-sized firms, the best buy is a time-boxed pilot with a contractual path to firm-wide pricing. Think of it as Model B with Model A's unit price locked in. Before you sign, negotiate three things in writing: the per-seat price at full rollout, the right to apply pilot spend as a credit against it, and a seat floor that matches the pilot rather than the vendor's standard minimum. If a vendor won't price the expansion up front, you lose your leverage the day the pilot succeeds.
Go firm-wide from day one only if three things are true: at least two practice groups have named partner sponsors, you've identified a measurable use case such as first-draft review or diligence summaries, and you can commit 6 or more training hours per lawyer in the first quarter. A usage-based assistant fits better when your work is mostly drafting and summarizing rather than research, and when someone will actually read the consumption report every month.
Your next step: rebuild the table above with your own headcount, research spend and internal hourly value. Then send the same three-year scenario to each shortlisted vendor and ask them to quote against it line by line.
Quick answers
How much does Harvey AI cost per lawyer?
Harvey doesn't publish prices. Contracts are negotiated, usually per seat, with minimum seat counts and multi-year terms. As a typical 2026 range, legal-specific platforms in this category run roughly $150–$500 per seat per month, depending on firm size, scope and region. Get a written quote that includes the seat floor and the escalator.
Is CoCounsel or Lexis+ AI cheaper than a standalone legal AI platform?
Sometimes it looks cheaper, because the AI is priced as an uplift on a research subscription you already pay for. That uplift is often 15–50% of existing research spend, and it may be priced on firm-wide entitlement, not on the number of users. Compare the incremental three-year cost against a standalone platform plus your existing research contract, not just the headline figures.
What hidden costs should a law firm budget for beyond licenses?
The big ones are non-billable training time, security and procurement review, research content uplifts, integration work and someone owning adoption internally. In our hypothetical 60-lawyer example, these ran from roughly half of license cost in the firm-wide seat model to well over license cost in the usage-based model. Outside the US, also budget for VAT or GST, possible withholding tax and currency movements.
Is usage-based pricing safer than per-seat pricing for generative AI?
It avoids paying for idle seats, but the cost becomes a forecast rather than a fixed price. Usage-based contracts suit firms that can monitor consumption monthly and have negotiated spend alerts and a fixed overage rate. If you can't commit someone to watching usage, a capped per-seat contract is usually easier to budget.
Written by the Counsel Core Tech editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.