Legal AI Cost Calculator
Three-year cost of Harvey, Legora, CoCounsel, Spellbook, Lexis+ AI, Clio Duo or a custom build, including the costs that never appear on the proposal.
A legal AI proposal shows a seat price. It does not show the 10-seat minimum on your 6-lawyer team, the onboarding fee that arrives in the order form, the professional-services line for connecting your DMS, or the contractor you hire because half of your precedents are scanned PDFs the platform cannot read. Those items routinely add 20 to 50 percent to year one. If you want the reported deal ranges behind the defaults first, start with what Harvey actually costs and our Harvey vs Legora vs CoCounsel comparison.
Set your seat count and say where your documents live. The calculator prices every option, including a custom build on Claude or GPT APIs, on the same basis over three years, with every extra cost as its own column. It counts only new money leaving the firm, the lines that show up in a budget, not your own people's time. Overwrite any default with the number on your actual quote.
Compare Legal AI Costs for Your Firm
Everyone who needs a login, not just partners. Most firms land between 10 and 25.
Someone has to pull it together, throw out the duplicates and file it properly first. Budget about $400 per seat.
Vendors raise the price at renewal. Buyers report 5–10% a year; we use the low end on years two and three.
Edit per-seat rates, onboarding fees and the custom-build figures
Three-year total by option
Each bar stacks the licence with every one-off cost. Hover a segment for the amount.
Where the money goes
The annual subscription, the four one-time invoices, and what that adds up to in year one and over three years. Scroll sideways for every column.
| Option | Privilege safe?Can you put privileged client material into it without waiving privilege? Yes when the provider is bound to confidentiality, does not train on your data and does not retain it (ABA Formal Opinion 512). All options here meet that under their enterprise terms; only the custom build can also keep the data inside your own cloud account or building. | SubscriptionWhat you are invoiced each year to keep using it: seat fees with the contract minimum applied (rises 5% a year). For the custom build it is the monthly retainer, hosting and model usage, billed together. | OnboardingOne-time implementation fee the vendor charges to set you up. For the custom build, the one-time engineering cost. | IntegrationOne-time cost to connect iManage, NetDocuments or SharePoint: vendor professional services or an outside IT contractor. Included in the custom build; nothing for a Word add-in. | SecurityOutside spend before signing: counsel on the DPA, a security consultant for the vendor questionnaire, and a penetration test for the custom build. | DocumentsContractor cost to make your documents readable: consolidating scattered files, OCR on scans, labelling. Priced per seat from where your documents live today. Same for every option. | Year 1Everything invoiced in the first twelve months: the subscription plus all four one-time items. | 3-yr totalThree years of subscription (rising 5% a year) plus the one-time items paid once. | $/seat/moThe three-year total divided by 36 months and by your seat count. The number to put next to the seat price on the proposal. |
|---|---|---|---|---|---|---|---|---|---|
| Harvey Research, drafting, due diligence and multi-document analysis for large firms · minimum applied | Yes data leaves the firm | $50K | $15K | $5.0K | $8.0K | $10K | $88K | $196K | $217 |
| CoCounsel Research on Westlaw, document review, deposition prep and drafting | Yes data leaves the firm | $68K | — | $3.0K | $6.0K | $10K | $87K | $232K | $258 |
| Legora Research, drafting, review and tabular document analysis; strong in EU and multi-jurisdiction work | Yes data leaves the firm | $75K | $10K | $5.0K | $8.0K | $10K | $108K | $269K | $299 |
| Clio Duo Practice-management assistant: matter summaries, time capture, client emails; not research or drafting · base plan not included | Yes data leaves the firm | $17K | — | — | $2.0K | $10K | $29K | $64K | $71 |
| Spellbook Contract drafting and redlining inside Word; not a research tool | Yes data leaves the firm | $45K | — | — | $5.0K | $10K | $60K | $156K | $173 |
| Lexis+ AI Legal research with cited answers, plus drafting from Lexis content · base plan not included | Yes data leaves the firm | $60K | — | — | $4.0K | $10K | $74K | $203K | $226 |
| Custom build (Claude / GPT API) The workflows you scope: precedent search, clause library, first-draft generation, intake, review checklists | Yes can stay in-house | $24K | $35K | — | $6.0K | $10K | $75K | $125K | $138 |
Which one fits: what it does, what it connects to, what you sign
On confidentiality the short answer is: every option here is fine under ABA Formal Opinion 512, because each sends your data to its model provider under enterprise no-training, zero-retention terms. Only the custom build can keep client data entirely inside your own cloud account or building.
| Option | What it is for | Connects to | What you sign | Who runs it | Live in | Certifications | Data stays in-house? |
|---|---|---|---|---|---|---|---|
| Harvey Model: OpenAI (Azure), with Anthropic and Google models added | Research, drafting, due diligence and multi-document analysis for large firms | iManage, NetDocuments, SharePoint; Word and Outlook add-ins | Annual, firm-level commitment; renews with uplift | Vendor support plus a firm KM / innovation lead | 4–8 weeks | SOC 2 Type II, ISO 27001 | No: Harvey-managed cloud, single-tenant options for large firms, zero retention, no training |
| CoCounsel Model: OpenAI on a private Azure instance, plus Anthropic and Google | Research on Westlaw, document review, deposition prep and drafting | Westlaw, Microsoft 365, iManage, NetDocuments | Annual; research bundle priced separately | Thomson Reuters support; light firm admin | 2–4 weeks | SOC 2 Type II | No: Thomson Reuters cloud, zero retention, no training |
| Legora Model: OpenAI, Anthropic and Google via enterprise agreements | Research, drafting, review and tabular document analysis; strong in EU and multi-jurisdiction work | iManage, NetDocuments; Word add-in | Annual, 10-seat minimum; onboarding quoted separately | Vendor support plus a firm KM / innovation lead | 4–8 weeks | SOC 2, ISO 27001 | No: Legora-managed cloud, EU or US region, zero retention, no training |
| Clio Duo Model: OpenAI (Azure) inside Clio's environment | Practice-management assistant: matter summaries, time capture, client emails; not research or drafting | Clio Manage, Outlook, Gmail | Per seat with the Clio plan; monthly or annual | Clio support; nothing for IT to run | Same day | SOC 2 Type II | No: Clio cloud, zero retention, no training |
| Spellbook Model: OpenAI, with other providers for some features | Contract drafting and redlining inside Word; not a research tool | Word add-in; works on whatever document is open | Per seat, monthly or annual; no minimum | Vendor support; nothing for IT to run | Same day | SOC 2 Type II | No: Spellbook-managed cloud, zero retention, no training |
| Lexis+ AI Model: OpenAI (Azure) and Anthropic (AWS Bedrock) inside the Lexis private cloud | Legal research with cited answers, plus drafting from Lexis content | Lexis+, Word (Lexis Create), Microsoft 365 | Annual add-on tied to the Lexis+ subscription | LexisNexis support; nothing for IT to run | 1–2 weeks | ISO 27001 | No: LexisNexis private cloud, zero retention, no training |
| Custom build (Claude / GPT API) Model: Your choice: Claude / GPT API, your own cloud tenancy, or an open-weight model on your hardware | The workflows you scope: precedent search, clause library, first-draft generation, intake, review checklists | Anything with an API: iManage, NetDocuments, SharePoint, Clio, Outlook | None. You own the code and the data path; retainer is month to month | The build team on retainer, or your IT once handed over | 8–14 weeks to first workflow | Your cloud provider's (AWS / Azure / Google SOC 2, ISO 27001) plus your own penetration test | Yes: your cloud account, or on-prem with an open-weight model (you run the servers) |
Figures last checked 2026-09-14. Every column is money that leaves the firm: licences, vendor fees, contractors and outside reviewers. Your own lawyers' and IT staff's time is not counted. Harvey, Legora, CoCounsel, Spellbook and Lexis+ AI do not publish list prices: every per-seat default is a reported buyer figure matching the vendor page linked in its row, not a quote. Onboarding, integration and review figures are working assumptions; edit them above. Add-on rows exclude the base subscription they sit on. Data-handling statements reflect each vendor's published security page on the same date; confirm them in the DPA before signing. Directional estimate, not a proposal.
Document Readiness Is Your Cost, Not the Vendor's
Every tool on this page answers from documents. If your precedents sit in one DMS with searchable text and usable metadata, the connection is a configuration task and the clean-up column is zero. If they are spread across a DMS, shared drives, email archives and a legacy system, someone has to consolidate, de-duplicate and re-folder them before ingestion. If a large share are scans, they need OCR, quality checks and metadata tagging before any model can cite them.
Nobody knows their document count, so the calculator prices this per seat: a working lawyer's precedents, templates and closed-matter files run to a few thousand documents each, and contractor clean-up runs a few cents per document. The cost is charged to every option because it belongs to your documents rather than to the vendor. The practical difference is timing: a platform lets you defer it and upload matter by matter, a custom retrieval pipeline needs the corpus in order up front. See our law firm DMS guide if the answer to “where do the documents live” is “several places”.
Attorney–Client Privilege: Which Options Keep Client Data Confidential
Short answer for the partner meeting: all seven options are usable without waiving privilege, because each processes your data under enterprise no-training, zero-retention terms. The difference is where the data physically goes. Six of them send it to the vendor's cloud and on to OpenAI, Anthropic or Google. Only a custom build can keep it inside your own cloud account or your own building.
The rule is simple. You cannot paste client matter into the consumer ChatGPT or Claude apps, because those products can retain and train on what you type and you have no confidentiality agreement with the provider. Under ABA Formal Opinion 512 and the state ethics opinions that follow it, using a vendor does not waive privilege as long as the vendor is bound to confidentiality, does not use the data for its own purposes, and you have done reasonable diligence on how it is handled.
Every platform in the chart clears that bar the same way: it sends your prompts and documents to OpenAI, Anthropic or Google under an enterprise agreement with no training and zero retention, inside the vendor's own cloud. That is what you are buying with the licence, and the fit table under the calculator shows each vendor's model provider and hosting next to what the tool is for, what it connects to and what you sign. What none of them offers is a stack that runs on hardware you control.
A custom build is the one option where that choice is yours. Through the API tiers of Anthropic and OpenAI the same no-training, zero-retention terms apply. Through AWS Bedrock, Azure OpenAI or Google Vertex the model runs inside your own cloud tenancy and the data never leaves your account. With an open-weight model on your own servers, nothing leaves the building at all, which is the only configuration that satisfies a client who has contractually prohibited third-party processing. See the AI vendor DPA builder for the clauses to demand from any of them.
When a Custom Build Wins
A custom stack on Claude or GPT APIs replaces the per-seat licence with a one-time build, a maintenance retainer and metered model usage. Usage is billed on tokens, not seats, and is small at drafting and review volumes, and API list prices have fallen repeatedly rather than risen, so the three-year cost flattens as seats grow while platform licences scale linearly and escalate at renewal.
The build wins when the firm has a defined set of workflows, enough seats to amortise the build, and someone to own the system. Smaller firms that mainly need contract drafting are often better served by a Word add-in such as Spellbook; see the full best AI tools for law firms roundup for the field. It loses when the firm needs the breadth of a platform on day one, cannot fund the build up front, or has no owner. The calculator's custom row includes a penetration test and a monthly retainer precisely so the comparison is not flattering.
How We Calculate Your Numbers
- Subscription, year one: seat rate × 12 × seats, or the reported contract minimum if higher. Custom build: retainer + hosting + model usage × 12.
- One-time costs: onboarding fee + DMS integration + security review + document clean-up (seats × the per-seat rate for where your documents live).
- Three-year total: subscription in years one, two and three with a 5% uplift on years two and three, plus one-time costs once. Custom-build model usage is held flat; only its retainer and hosting escalate.
- All-in per seat per month: three-year total ÷ 36 ÷ seats. This is the number to put next to the seat price on the proposal.
Harvey, Legora, CoCounsel, Spellbook and Lexis+ AI do not publish list prices. Every per-seat default is a reported figure that matches the vendor page linked from its row, and the check date is shown under the table. Onboarding fees, integration and review costs are our working assumptions, printed in the edit panel so you can replace them.
Frequently Asked Questions
- Harvey does not publish pricing. Buyer reports put it at roughly $100 to $200 per seat per month at AmLaw scale, with small pilots reported at $800 to $1,200 per seat, firm-level access from about $50,000 a year, and a median contract near $175,000 a year. The calculator uses $150 per seat with a $50,000 annual floor by default. Replace it with the figure on your proposal.
- Legora is quote-only. Third-party 2026 coverage reports roughly $200 to $500 or more per seat per month billed annually, with a 10-seat minimum that puts the practical floor near $30,000 a year. Implementation, onboarding and DMS integration are quoted separately. Reported discounts after negotiation run 40 to 60 percent off list.
- Often, once the seat count clears roughly 15 to 25 and the firm has a defined set of workflows. A custom stack swaps the per-seat licence for a one-time build plus a maintenance retainer and model API usage, which has been falling in price rather than rising. It loses when the firm needs the breadth of a platform on day one, has no one to own the system, or cannot fund the build up front.
- Because every option can only search what it can read. A platform that ingests scanned PDFs with no metadata returns weak answers just as a custom retrieval pipeline would. The clean-up cost is a property of your documents, not of the vendor, so it appears in every row, priced per seat from where your documents live today. The only thing that changes is whether the vendor sells the clean-up as a service or you hire it yourself.
- Contract minimums, for small teams: a 10-seat floor on a 4-lawyer team more than doubles the effective seat price. For larger firms it is document clean-up. If precedents are spread across systems or sit in scanned PDFs, a contractor has to consolidate, OCR and label them before any tool can use them, and that invoice can rival the first year of licences.
- No. It counts only new money leaving the firm: licences, vendor fees, contractors and outside reviewers, the lines that appear in a budget. Lawyers learning the tool and IT staff administering it are real costs, but they are already on the payroll and are left out so the total matches what finance will see.
- Yes, when the vendor is bound to confidentiality and does not train on or retain your data, and you have done reasonable diligence, which is the position of ABA Formal Opinion 512. Every platform in the calculator states no-training and zero-retention terms with its model provider. What you cannot do is put client matter into the consumer ChatGPT or Claude apps. A custom build on the API tiers, in your own cloud tenancy, or on an open-weight model you host is the only route where the data can stay entirely under your control.
- They are reported figures, not list prices, because the enterprise vendors do not publish list prices. Each default matches what our own vendor pricing pages report and is dated in the calculator. Use the calculator to structure the conversation, then overwrite every default with the numbers on your actual quote.
The complete AI playbook for law firms
The Complete Law Firm AI Implementation Guide (2026): a step-by-step rollout plan for choosing, buying and running legal AI at a solo-to-mid-size firm, built to sit next to the numbers above when you make the case internally.
Get the guide — $59 (reg. $89)