Build, Buy or Host: How Law Firms Now Evaluate Legal AI Vendors

Legal AI vendor evaluation now follows three paths. Some large firms build, running their own models on owned hardware, as Latham & Watkins has started to do. Many buy from well-funded platforms such as Harvey. Others back providers that host data in-country, as White & Case did with Clauze.AI in Saudi Arabia. Vendors win by proving data handling, residency, accuracy and integration.

Below: the week's signals, what each path means for buyers, what vendors should prove and a six-step pipeline plan. For how AI assistants assemble legal tech shortlists, see AEO for legal tech and CLM software.

The week's signals

Five announcements between September 9 and September 11, 2026 show one shift from different angles: buyers want control over where client data goes, what AI costs and which models run.

Harvey's round, covered widely the week before, is included for context.

The three evaluation paths: build, buy or host

The paths are not exclusive: one firm can build for its most sensitive work, buy a platform for daily drafting and require in-country hosting for a regional office.

Build: firms that want to own the stack

Latham, which Legal IT Insider describes as the second-largest US law firm with $8.3 billion in revenue last year, has bought several Nvidia GPU servers and is customizing open-weight models itself. Chief information officer Rene Mendoza told the Financial Times that some client information is sensitive enough that the firm does not want to put it "to any cloud vendor," and he pointed to coming consumption costs as a reason to keep flexibility. Legal Cheek called Latham the first major law firm publicly known to have bought its own AI hardware and fine-tuned models, and reported that the firm still uses general tools including ChatGPT, Claude and Gemini.

Building is not a rejection of vendors. Legal IT Insider lays out the trade-off: substantial capital investment, specialist technical expertise and full responsibility for securing the infrastructure. A build-path firm judges vendors on what it cannot or will not build itself.

What build-path buyers need to see:

• Deployment inside their environment, or a clear statement of which components can run there.
• Model choice, including support for open-weight models the firm has customized.
• The value above the model: workflow, review screens, matter permissions and integrations the firm's engineers would otherwise have to build and maintain.
• A pricing structure the CIO can forecast as usage grows.

Evaluators here are technical: the CIO, AI leads, information security and knowledge management.

Buy: firms choosing well-funded platforms

Harvey says it serves more than 3,000 customers, including 80% of the top 100 law firms, according to LawNext's report on its $550 million round. The same report notes that Harvey introduced Tenet, its first post-trained open-weight model, and launched Harvey LAB, a Legal Agent Benchmark. For a specialist vendor, that means many large firm prospects already own a general platform, and the evaluation turns on fit alongside it.

Xapien shows how a specialist competes in that market. It sells one workflow, AI-native due diligence on people and companies, to corporations, law firms and private banks, and names Greenberg Traurig and KPMG among representative clients. Its new capital goes into the US, where it already generates 50% of its revenue.

What buy-path buyers need to see:

• A workflow the incumbent platform does not cover well, described in the practice group's own terms.
• Named references from comparable firms, or a pilot structure that can produce one.
• How the product works with tools the firm already pays for, without duplicate spend.
• Durability: funding, ownership and roadmap, which matter to a firm justifying yet another tool.

Host: buyers that need data to stay in-country

White & Case's strategic investment backs a tool built around residency: Clauze.AI offers bilingual capabilities, full data residency in Saudi Arabia and on-premises deployment. Its founder, Waad Alkurini, spent a decade at White & Case, most recently as executive partner of the firm's Riyadh office. Chair Heather McDevitt linked the investment to the Kingdom's focus on digital transformation and innovation under Saudi Vision 2030. The firm has around 40 lawyers in Saudi Arabia, including nine partners.

For vendors selling into the GCC, a local competitor now leads with residency, bilingual capability and on-premises deployment, backed by a global law firm.

What host-path buyers need to see:

• Where data is stored, where model inference runs and from where support staff can access it.
• In-country hosting or on-premises deployment, with named hosting partners.
• Bilingual performance on the buyer's own documents, not only English demos.
• Local references and a team that can meet in person.

Which residency rules apply is a question for counsel; make the facts about your product easy to find.

What legal AI vendors should prove

None of this calls for invented benchmarks. Prove five things with evidence a reviewer can check.

1. Security and confidentiality

State plainly whether client data trains any model, which subprocessors touch it, how long it is retained and how matter-level permissions and ethical walls carry into the product. List only the certifications and audit reports you actually hold, and share trust documentation before the first evaluation call.

2. Data residency and deployment options

Publish a deployment matrix: cloud regions, single-tenant or private cloud, in-country hosting and on-premises, with what each option supports. Name the model providers behind each option and say whether a firm can bring its own model. After Latham, expect build-path firms to ask.

3. Accuracy evidence

Offer a pilot on the firm's own documents with success criteria agreed in writing before it starts: which tasks, which reviewers and what counts as an error. Publish your evaluation method, including where the product is weak. If you cite a public benchmark, link it and say who built it. Harvey LAB, for example, was launched by a vendor.

4. Integration with DMS and CLM

Lawyers work in the document management system, and many in-house teams work in contract lifecycle management platforms. Show the connection to the DMS, such as iManage or NetDocuments, and to CLM, including how permissions are inherited and whether outputs write back to the matter or contract record. Our Aavenir case study shows what integration-led positioning looks like for AI contract lifecycle management built natively on ServiceNow and sold to procurement, legal, finance and IT.

5. Cost predictability and procurement readiness

Legal IT Insider notes vendors beginning to move away from subsidized token use toward consumption pricing. Explain what drives cost, how usage can be capped or forecast and how a customer moves between tiers, without publishing prices. For corporate legal deals, prepare for procurement early. In ProcureAbility's 2026 CPO Benchmark Study, 50% of respondents named poor data quality as the leading barrier to scaling AI, and 43% cited late procurement engagement as the biggest barrier to achieving value. A vendor pack that answers data, security and commercial questions up front gives procurement what it needs before it asks.

How to build pipeline with law firms and in-house legal

Legal AI buying committees are narrow at the top and wide underneath: a sponsoring partner or general counsel, then innovation, IT, security, knowledge management and procurement. Pipeline comes from reaching the right people at named accounts with proof they can forward.

1. Build named account listsTier firms and legal teams by build, buy or host.
2. Map partners and innovation leadersSponsors, evaluators and blockers per account.
3. Publish proof contentDeployment, security, integration and pilot pages.
4. Host peer roundtablesClosed-door sessions on one hard question.
5. Pre-book meetingsBook legal buyers before events, not at the booth.
6. Win AI search visibilityGet named when buyers ask for a shortlist.

1. Build named account lists by path

Start from public signals: firms announcing AI infrastructure, platform rollouts, innovation hires or regional office openings. Tag each account by its likely path, because a build-path firm needs a different first message than a firm that has just rolled out a platform. Run the list through account-based marketing so sales and marketing work the same accounts.

2. Map partners and innovation leaders

For each account, name the sponsor (managing partner, practice group leader or general counsel), the evaluators (chief innovation officer, CIO, knowledge management, information security) and, for in-house deals, procurement and legal operations. Record what each firm has already announced so outreach starts from its own words.

3. Publish proof content

Turn the five proof areas into pages and short PDFs: a deployment and residency matrix, a data handling explainer, a pilot method and integration guides for named DMS and CLM platforms. Founders and product leaders can carry the same material on LinkedIn through CXO branding, for buyers who want to know who stands behind a product.

4. Host peer roundtables

Invite innovation and IT leaders from a small group of non-competing firms or legal departments to a closed-door session on one question, such as when building beats buying or what residency should require. Keep the pitch out of the room and follow up with a written summary attendees can share internally.

5. Pre-book meetings around events

Legal technology conferences put many buyers in one place for a few days. Pre-booked meetings tied to a roundtable or briefing give an innovation leader a reason to meet beyond a booth demo. For Saudi buyers weighing residency, see our page on revenue pipeline generation in Riyadh.

6. Win AI search visibility

General counsel and innovation leaders can now ask AI assistants for legal AI shortlists and comparisons. If your deployment, residency and integration pages are thin, the answer will rely on someone else's description. The mechanics are covered in AEO for legal tech and CLM.

Where Lemniscate fits

Lemniscate Growth builds pipeline for CLM, legal AI and procurement software vendors selling to general counsel, legal ops, procurement and finance. See our legal and procurement tech practice and the Aavenir case study, where qualified meetings grew from single digits to tens a month, with 90 to 95% of results coming from inbound. We work with 35+ active clients and generate up to $10M in pipeline per client. Book a free growth audit and we will map which firms sit on each path, who evaluates and what proof to put in front of them.