Every managing partner I talk to in 2026 wants the same thing: a clear answer to a simple question. How much does this actually cost, all in, over three years? And every legal-AI vendor in the market has organized itself to make that question hard to answer. Seat pricing that tiers upward. Feature gates. Implementation fees. Minimum commitments. Auto-escalation clauses buried at the back of the order form. The list price is a starting line, not a finish line.
What follows is the three-column spreadsheet I build for every partner group before we talk about scope. It is calibrated to a real-world anchor — a 20-attorney U.S. firm with roughly 10 paralegals, an operations lead, and an active interest in using AI for one to two meaningful workflows. The numbers are 2026 market ranges, pulled from published pricing, firms that have shared terms with me, and the back-of-envelope math any partner can redo themselves in an afternoon.
Read it as an argument, not a quote. Every firm is different. But the shape of the comparison — and where the real costs hide — is remarkably consistent.
Why legal-AI pricing is opaque on purpose
Legal-AI vendors in 2026 are in a scramble for logos. Every signed Am Law firm shows up in the next pitch deck. That dynamic has two predictable effects on pricing. First, list prices are inflated so that each negotiated discount feels like a win. Second, the most expensive line items — implementation, integration, premium support, enterprise security tiers — are quoted separately, often after the partner committee has already said yes to the headline number.
The result is that the real cost of a legal-AI deployment is usually 1.8 to 2.4 times the advertised per-seat rate by the end of year one. At year three, after seat expansion and the standard 8 to 12 percent annual price escalator, the real cost is 3 to 4 times the year-one headline.
That is not a criticism. It is how the category prices itself. But a partner who prices the build alternative without modeling the SaaS alternative on the same all-in basis is comparing apples to a carefully curated subset of oranges.
The three columns I build every time
Column A — Multi-tenant legal-AI platform. A market-leading legal-AI product, deployed as a SaaS subscription, scoped to cover research, drafting, and document review for the whole firm.
Column B — Single-tenant hosted AI. The same or similar product, deployed as a dedicated instance in the vendor’s cloud, with isolation between firms and a stronger security posture. Enterprise-tier pricing.
Column C — Custom Brightline module(s). A focused custom build targeting the firm’s highest-value workflow, delivered on a fixed-price Sprint, hosted on the firm’s own cloud account, with source code handed off through the firm’s GitHub. Optional retainer for ongoing support.
The same firm sits underneath all three columns: 20 attorneys, 10 paralegals, one litigation-heavy practice, and a real appetite to change how at least one workflow runs.
Column A: Multi-tenant legal-AI platform, fully loaded
Market pricing for a multi-tenant legal-AI platform in 2026 ranges from roughly $110 to $280 per attorney per month, depending on tier, contract length, and negotiated discount. A 20-attorney firm on the mid-tier, committing to a 24-month deal, typically lands around $160 to $220 per attorney per month. Most vendors will throw in paralegals at a reduced rate if pushed.
Year one, line by line:
Seat subscriptions, 20 attorneys at $190/month blended, plus 10 paralegals at $70/month: roughly $54,000 for the year. Implementation and onboarding fee, typically quoted as a one-time charge on enterprise deals: $12,000 to $25,000. Integration services, if the firm wants the AI to read from its document-management system or practice-management software: another $15,000 to $40,000, billed separately or bundled into professional-services hours. Training and change management, either from the vendor or a third party: $8,000 to $15,000 if done well.
Year-one all-in, for a firm that does the rollout right: $90,000 to $135,000.
Year two adds the price escalator (usually 8 to 12 percent), the cost of the inevitable seat expansion as non-attorney staff want access, and a renewal negotiation that rarely reduces the per-seat rate from year one. Call it $72,000 to $95,000.
Year three looks like year two plus another escalator, plus — in most cases — an upsell to a higher tier as the vendor releases features that were previously included and are now behind a new paywall. Call it $80,000 to $110,000.
Three-year all-in, Column A: roughly $240,000 to $340,000.
None of that money builds a firm-specific asset. At the end of year three, the firm has access to the same product its competitors have, minus whatever the firm has loaded into the system as prompt libraries and reference documents. A good product, no question. But a rental.
Column B: Single-tenant hosted, the enterprise tier
Single-tenant pricing for the same class of product runs about 2.5 to 4 times the multi-tenant tier. The rationale the vendor gives is reasonable: dedicated infrastructure, stronger isolation, a tighter security posture. Everything on the due-diligence checklist that mattered in the previous post looks better here.
A 20-attorney firm on a single-tenant enterprise contract typically pays between $350 and $600 per attorney per month, with minimums that sometimes force the firm to buy capacity it will not use in year one.
Year-one all-in, including the higher implementation fee, integration, and training: $180,000 to $260,000. Year two and three scale similarly to Column A, just from a higher base. Three-year all-in, Column B: roughly $500,000 to $750,000.
This tier exists for a reason. For a certain kind of firm — high-stakes regulatory work, defense-side litigation with classified or near-classified material, anything where the bar’s ethics opinions make multi-tenant a non-starter — Column B is the only responsible rental option.
But it is still a rental. At year three, the firm has the same generic product every other Column-B customer has, hosted on dedicated infra. The dollars have bought isolation, not differentiation.
Column C: A custom build the firm owns
Here is where the comparison gets interesting. A focused custom AI module, scoped to one genuinely painful workflow at the firm, costs roughly 15 to 50 percent of a one-year Column-A spend in upfront build cost — and carries no ongoing seat subscription.
A Brightline Micro-App — a narrowly scoped agent that automates one workflow, priced as a two-week fixed-price Sprint — lands in the $15,000 to $50,000 range, depending on the data sources it has to integrate with and the degree of calibration against firm templates required. A Brightline AI Sprint, for a more substantial two-week build covering a larger or more technically involved workflow, is priced at $30,000, fixed.
Most 20-attorney firms end up in one of two patterns. Pattern A: two Micro-Apps, one in month one and one in month four, covering the firm’s two highest-value workflows. Pattern B: one AI Sprint up front, scoped ambitiously, covering a more integrated workflow end to end. Either way, the first-year build spend lands between $40,000 and $90,000.
On top of the build, the firm pays for the underlying foundation-model API directly (a small fraction of what the SaaS equivalents cost — typically $200 to $800 per month at this workload), hosting costs on its own cloud account ($150 to $400 per month), and an optional retainer for post-handoff support ($3,000 to $6,000 per month, if the firm wants it).
Year-one all-in, typical scenario: $55,000 to $115,000.
Year two is where the math really changes. The build is done. There is no seat subscription. The firm pays API and hosting costs (~$5,000 to $15,000 annually) and, if retained, a modest monthly for support and minor enhancements. Adding a second or third workflow to the custom module, or building a sibling module, is a new fixed-price Sprint — not an escalating seat expansion.
Year two all-in, maintaining and modestly extending: $30,000 to $65,000. Year three, similar: $30,000 to $65,000.
Three-year all-in, Column C: roughly $115,000 to $245,000. Less than half of Column A and a third or less of Column B. And at the end of year three, the firm owns three things that no Column-A or Column-B customer has: the source code, the data, and the institutional knowledge required to extend the system on its own or through any competent contractor.
The line items that never appear on vendor slides
The spreadsheet is not complete until a few line items the vendor never shows up front are priced in. Every partner group I have walked through this exercise has missed at least two of these.
Data-portability cost at exit. On the day a Column-A or Column-B contract ends, the firm’s loaded prompt library, reference documents, and tuning work may or may not travel. Read your contract. If the answer is “may not,” the firm is effectively paying to build and repeatedly abandon institutional knowledge. In Column C, the data and the code are already the firm’s.
Vendor switching cost. Firms that switch legal-AI vendors — and by year three, a third of firms do — pay a second implementation fee, a second integration bill, and a second round of training. That is a real cost that should be modeled at a 30 to 40 percent probability in any Column-A three-year scenario. Column C has no vendor to switch away from.
Partner and associate time spent in vendor meetings. Enterprise-tier legal-AI contracts come with quarterly business reviews, annual renewal negotiations, security questionnaires from general counsel, and the regular cadence of feature-request meetings. Multiply the blended hourly rate of the partners in each meeting by the meeting count. It is a larger number than most firms realize, typically $8,000 to $20,000 per year in partner time.
Feature regression. Multi-tenant products ship updates on the vendor’s timeline, not the firm’s. Sometimes that is a net positive. Sometimes a feature the firm relied on gets deprecated, paywalled, or changed enough to require rework. Budget one or two of these per year. They almost always happen.
The ROI math — and why it is quicker than partners expect
Pick one workflow. Estimate the hours the firm currently spends on it in a year. Estimate how much of that an AI module can realistically displace (usually 60 to 85 percent for a well-scoped workflow, sometimes higher). Multiply by the blended rate of the staff doing the work today. That is the upper bound of the annual savings.
A concrete example. A 20-attorney litigation firm spends roughly 2,500 hours a year turning deposition transcripts into issue-tagged outlines, done by paralegals at a $95/hour blended cost (fully loaded). A custom module that replaces 75 percent of that work recovers 1,875 hours, worth $178,000 a year in capacity. Against a Column-C year-one cost of, say, $85,000, the workflow pays for itself in under six months and returns 2.1x in year one alone. Column A would need a comparable productivity lift across every attorney in the firm to justify its larger annual spend, which — honestly — it sometimes does.
The point is not that Column C always wins. The point is that the math is legible. A partner can sit down with the spreadsheet, plug in the firm’s own numbers, and come to a defensible answer in an hour. The vendors would prefer that not happen.
When each column is the right answer
Column A is the right answer when: the firm wants broad, shallow coverage across many workflows; the partner group has no appetite for managing any technology; and the firm is comfortable renting a product that every competitor in its market can also rent.
Column B is the right answer when: the firm has specific regulatory or client constraints that rule out multi-tenant deployment, and the partnership is willing to pay 3x to 4x for the stronger isolation posture.
Column C is the right answer when: the firm has at least one high-value workflow it actually knows well enough to specify, wants the savings to compound as assets rather than as subscriptions, and cares about owning the institutional knowledge that emerges from the build.
For most 20-attorney firms — especially ones with one or two practice-defining workflows — Column C is a better match than the partnership originally assumes. The hesitation is almost never about the math. It is about the unfamiliarity of owning software at all. That hesitation is worth naming, but it is not a pricing argument.
Run the spreadsheet yourself
Build the three columns. Put your firm’s own seat count and workflow in the inputs. Add the line items the vendor did not show you. Apply the escalators for years two and three. Compare against the ROI of one well-chosen custom build.
Most partner groups that do this exercise for the first time end up surprised by the same two things. Column A is more expensive, over three years, than they thought. And Column C is cheaper, and simpler, than the sales materials of the past decade conditioned them to believe.
If you would like a second set of eyes on your own three-column spreadsheet, that is exactly what a thirty-minute bottleneck audit is for. Bring your contract and your assumptions. I will tell you which column I would bet on for your firm, and, if the numbers say Column A is the right answer for you, I will say that too.
