Firm Economics
Pricing, procurement, and what AI actually costs a firm.
Opening Argument
One argument, drawn from the week to date.
Legora told clients this month it is moving its most advanced products to consumption-based pricing, and Gartner expects the rest of the market to follow: by 2028, more than 35 percent of new legal-tech spending will be tied to how much AI a team actually uses, not how many seats it bought.
The bill that logic produces is already landing. Harvey's co-founder told Legal IT Insider that customers should expect "consumption bills of $10 million," and that the firm's own token usage grew fourteenfold in six months. Latham & Watkins says it is deliberately not hitching its AI stack to one vendor.
Most firms have not caught up to either half of that story. A survey of more than 800 legal finance professionals found 63 percent still price a matter the way they priced it before AI, and only 25 percent discount when a client authorizes AI use. Whatever a consumption model ends up costing, most firms have not yet decided who keeps the savings it was supposed to create.
The Implication
Model what a consumption contract would bill your firm at peak usage before you sign it, not after the first invoice. And decide now, before a client asks, how much of any AI efficiency gain your pricing intends to keep versus pass through.
The Findings
What moved in practice this week, and what each one tells you.
Legora is moving its most advanced AI products to consumption-based pricing, and Gartner says the rest of the legal-tech market is right behind it.
Legora's Alex Fortescue-Webb said flat per-seat pricing "was never the right long-term model for AI, because usage varies too much between people for a flat fee to hold up." Gartner projects that by 2028 more than 35 percent of new legal-tech spending will tie to AI consumption rather than seats, and estimates the shift could raise IT costs roughly 25 percent for information-heavy work like litigation and M&A due diligence.
Why it matters That 25 percent is Gartner's estimate, not a measured result, so treat it as a planning number rather than a forecast. Baker McKenzie's Ben Allgrove framed the shift as a procurement posture, telling the Post: "the days of throw the latest and newest AI at every problem are over."
Harvey's co-founder is telling customers to expect "consumption bills of $10 million" as usage-based pricing replaces flat licenses.
Gabe Pereyra said Harvey's own token usage grew fourteenfold in six months. A complex drafting query can run about £15 in model costs; reviewing 100,000 contracts can run to roughly £15,000. Latham & Watkins CIO Rene Mendoza said the firm is deliberately "not hitching our wagon to one particular company."
Why it matters A token bill scales with the work a firm actually does, so a document-heavy quarter will no longer cost the same as a quiet one for the same software. Ask every AI vendor for a worst-case bill at your firm's peak usage, not just its list price, before you sign anything.
A survey of more than 800 legal finance professionals found most firms are still pricing AI-assisted matters the way they priced them before AI, and few are passing savings to clients.
BigHand's 2026 Legal Pricing and Budgeting Trends survey, of senior legal finance professionals across the UK and North America, found 63 percent default to conventional pricing when quoting a matter and only 25 percent discount when a client authorizes AI use. Roughly 30 percent hold prices flat while letting AI efficiency widen margin; another 30 percent are still deciding.
Why it matters Those two groups put something close to 60 percent of firms nowhere near ready to answer a client who asks where their AI discount went. "We're evaluating" will not satisfy a general counsel doing the same math.
Chamelio raised a $26 million Series A to sell in-house legal teams autonomous agents for contract review, not another assistant a lawyer has to prompt.
Entrée Capital led the round, joined by existing investors Work-Bench and Emerge Ventures plus new backer Bright Pixel Capital. CEO Alex Zilberman says the platform acts on "the corporate intelligence of each individual business" rather than a generic playbook; the money is earmarked for a proprietary legal action model and onboarding tooling.
Why it matters This is a bet on the in-house buyer, where the customer is the client rather than the firm billing them, and the product is judged on headcount avoided, not hours saved. Watch whether Chamelio prices per seat or per contract reviewed; that choice answers this issue's central question.
A UK legal commentator argues that an AI vendor's regulatory risk to a law firm is set at procurement, not when a lawyer first opens the tool.
Eleonora Dimitrova lists five baseline demands for the contract: documented data flows from ingestion to deletion, enforceable limits on subcontractors and cross-border transfers with audit rights, named processing jurisdictions, plain-language explanations of how outputs and errors are generated, and a verified, clean way to extract and delete data on exit.
Why it matters A clean exit is the one firms skip most often, because it only matters on the way out, exactly when a firm has the least leverage left to negotiate it. Put exit terms in the first draft, not the termination letter.
The Practice
This week's deep read on AI in practice.
Three things happened to Firm Economics this week that only make sense read together. Legora told the market it is moving to consumption-based pricing. Harvey's co-founder said the resulting bills could hit $10 million for some customers. And a survey of 800-plus legal finance professionals found most firms have not changed how they price a matter since before any of this started. The gap between the second and third fact is where the risk sits.
Legal IT Insider's reporting put real numbers on the abstraction: a single complex drafting query can run about £15 in model costs alone, and reviewing 100,000 contracts can run to roughly £15,000, before any lawyer's time is added. Harvey says its own token usage grew fourteenfold in six months. None of these are audited third-party figures; they are the vendors' own reporting, which is still the best data available this early in the shift.
A consumption bill does not land the same way on every firm. An hourly firm is structurally ambivalent about efficiency, because faster work is less billable work unless the rate card changes with it. A contingency firm treats efficiency as pure margin, measured in cycle time, so the same invoice is a cost it can model against settlement value. A volume practice prices per matter, so a token bill either fits inside that price or erodes it, with no billable-hour cushion to absorb the difference. The BigHand numbers below describe firms in aggregate; which bucket a firm falls into changes what the 63 percent figure means for it.
63%
Share of surveyed firms that still price a matter the same way they priced it before AI, per BigHand's 2026 survey of legal finance professionals.SOURCE · BIGHAND 2026 LEGAL PRICING & BUDGETING TRENDS, VIA EDISCOVERY TODAY
On the Methodology
BigHand surveyed more than 800 senior legal finance professionals across the UK and North America; respondents opted into a pricing-focused survey, which likely skews the sample toward firms that already track pricing formally. The report does not break results out by firm size or practice area, so it does not show whether a 20-lawyer plaintiff shop and a 2,000-lawyer defense firm are making the same choice for the same reasons.
None of this is fixed once a vendor sends a contract. Exit terms, price caps, and audit rights are still negotiable at signing, before the firm has generated the usage data that gives the vendor leverage to hold price. A commentator writing in Legal Futures this week made the same point about regulatory risk: the moment that sets a firm's exposure is procurement, not the day a lawyer opens the tool. Financial exposure works the same way. A firm that waits to negotiate a price cap until after its usage has scaled fourteenfold is negotiating from the position the market already expects it to be in.
The firms best positioned for what comes next are not the ones with the most AI licenses. They are the ones that can say, today, what a bad month of AI usage would cost them, and what a vendor is already contractually barred from doing about it if they want to leave.
Exhibit AI
When AI goes wrong in the legal profession. Not to assign blame, but to learn the lesson before it is ours.
In October 2025, DK Law, a California personal injury firm, told its AI and telecommunications vendor ConnexAI that it would not renew their contract, citing what DK Law describes as a year of persistent system failures. ConnexAI kept invoicing anyway.
What followed is now two lawsuits, according to The Register. ConnexAI sued DK Law in New York in March 2026 for non-payment; DK Law countersued in California in April, alleging breach and misrepresentation. In August, the California court stayed its case pending the New York matter; on 1 September a federal magistrate denied a further stay. ConnexAI's general manager called the allegations "one part of a wider dispute" and said they are "rejected." Neither court has ruled on the merits, so this is a live dispute, not a verdict against either party.
Whatever the two sides eventually prove, the dispute turns on exactly the clause this issue's procurement checklist flags as the one firms skip: what happens when the relationship ends, not when it begins. A contract that is unclear on notice periods and automatic renewal is a contract two parties can each read in good faith and still land in different courtrooms.
The Lesson
Negotiate your exit before you negotiate your onboarding call. A renewal clause neither side remembers agreeing to does not care how good the AI is, and it will not resolve itself before the invoice does.