← Back to HAQQ Blog

How to Become an AI-Native Law Firm: A Seven-Part Operating Model

By Nicolas Dawalibi · · Updated · 10 min read · Guides

US law firm tech spending rose 39.3% from 2021 to 2025, yet only 18% of professionals say their firm tracks AI ROI. A seven-part operating model and a 90-day start.

Key facts

This is the firm-level companion to the AI-native lawyer, which covers the individual practitioner.

Why are law firms spending more on AI without seeing a return?

Put the first two facts side by side. Four years of compounding technology spend, and fewer than one professional in five can point to a measured return. Firms are not underinvesting. They are investing without the structure that would let them say what they bought.

That structure is what we mean by an operating model, and it is the subject of this guide. Thomson Reuters' 2025 data puts a number on it: organisations with a visible AI strategy were 3.5 times more likely to report critical benefits. The tool matters less than the plan around it.

What does AI-native mean for a law firm?

The phrase gets used for three different states, and mixing them up is expensive.

Usage is rising fast: organisation-wide AI use went from 22% to 40% in a year, and on the in-house side Axiom counts 21% of departments as mature. The structure is what lags, and structure is what the rest of this guide covers.

What are the seven parts of an AI operating model?

Order matters more than completeness. Each part depends on the ones before it, which is why firms that start at part five get the least from it.

1. Literacy as a floor

Every lawyer in the firm knows what the technology does, what it cannot do and how to write an instruction that holds up on a real matter. Think of it as a floor nobody is allowed below, kept up with short, repeated sessions tied to the work.

There is a regulatory reason too. Article 4 of the EU AI Act applies to deployers as well as builders. Regulation (EU) 2026/1744, in force since 27 July 2026, softened the duty to taking measures that support AI literacy, without removing it; our EU AI Act amendments guide has the detail. The useful content is specific: how to prompt for a jurisdiction, how to spot a fabricated citation, when to stop and open the primary source. Our legal prompting guide is the version we teach.

2. Buy narrow, build where your edge is

The default failure is a blanket licence for every lawyer, bought to look decisive and measured by seat count. Seats issued tell you nothing about use or advantage.

Buy off the shelf for general work: research, summarising, first-pass drafting, translation. These are commodities. Build only where a workflow is yours alone: the review your firm does differently, the precedent bank nobody else has, the filing pattern in a court you know better than the market. If a build does not touch something proprietary, buy instead.

A disclosure: we sell legal AI, and we are telling you to buy less of it than most vendors would like. A firm that buys 800 licences and wires them into no workflow cancels at renewal, so our incentive points the same way as yours here. For the sceptical version of buy versus build, read the small-firm guide and our pricing teardown.

3. A ranked use-case portfolio, with owners and dates

Score each candidate use case on three axes: impact if it works, feasibility with today's models, and readiness of the data and process underneath. Give each line a named owner and a date. A use case without an owner is a wish.

Build the portfolio practice group by practice group, from the work those lawyers do every week. A use case invented centrally and handed down is the most reliable way we know to produce a tool nobody opens. Plan three to five years out and expect the back half to change; the horizon is for sequencing.

One reason to look past next quarter: 15% of organisations already use some agentic AI and another 53% are planning or considering it (Thomson Reuters, Feb 2026). The second half of any roadmap will meet systems that take several steps without supervision, and rules that are still being written.

4. Codify the know-how

The most valuable asset in most firms is undocumented: how a senior partner reviews a contract, which clauses they read first, their preferred position, their fallback, where they concede. That knowledge passes on slowly by example, and it leaves when the partner retires.

Written down as a playbook (criteria, preferred positions, fallbacks, red lines), it becomes something every lawyer can run and something a model can execute against. HAQQ's contract review runs custom playbooks of this kind. This part has the worst ratio of importance to effort spent, because it is slow work no tool does for you. It is also the one part competitors cannot buy.

Treat the corpus as an asset. Models get cheaper, interfaces converge and features get copied within a quarter. Firm-specific precedent, redlines and know-how stay yours. The legal operating system makes the longer argument.

5. Integration past the chat window

A chat window is where AI gets evaluated and where adoption stalls, because it asks the lawyer to leave the work, restate the context and carry the output back by hand.

An integrated setup triggers a skill from where the work already is, pulls the playbook from the document store and routes the output into the matter file. Nobody retypes anything. This is the plain reason firms with high licence counts report low value: the tool was never connected to the place the work happens. Our legal engineering guide covers the plumbing.

6. Human review by design

A lawyer reviews every output, permanently, as part of the design. In legal work the review is what the client pays for, and the interface either makes it fast or turns it into a formality.

Fast review needs three things on screen: the source next to the claim, a trace of how the system reached its position, and a clear signal when it is unsure. With those, lawyers override the output often and still want the tool. Without them, they rubber-stamp, which is worse than not using it. We go further in human in the loop and governance by construction.

7. Measure each use case

Firm-wide AI ROI cannot be measured in any useful way, and chasing it is how firms end up among the 82% of respondents who could not say their organisation tracks a return. A single use case can be measured with four numbers:

Four numbers per use case let a management committee fund the next tranche on evidence. They are also the answer to the risk the Thomson Reuters Institute flagged in January 2026: firms that invested heavily in AI without a clear return plan face higher costs, lower utilisation and rate pressure if conditions turn.

Where does AI efficiency show up in law firm finances?

On hourly work, cutting a task from four hours to one turns four billable hours into one. The efficiency is real and the revenue goes with it. With about 90% of legal spend still billed hourly, that explains why enthusiasm for AI inside firms tracks billing structure so poorly.

On fixed-fee work, the same cut becomes margin: conveyancing, probate, incorporations, standard NDA review, high-volume contract triage. Start there. It is the fastest proof you can produce, and it survives a partner meeting.

The Thomson Reuters Institute reports profit per lawyer 8.4% above pre-2022 levels by the end of 2025 and fees worked per lawyer up 16.8%, mostly from rate increases. Whether productivity gains become durable margin or get competed away depends on fee structure; the death of the billable hour follows that thread.

How do you get lawyers to adopt AI?

Every firm splits three ways: a few front-runners already ahead of the firm's policy, a large curious middle who will use a tool that fits their day, and a sceptical tail who have seen technology promises before and were often right.

Build the training for the middle. Front-runners do not need it and the tail will not come. Then let the front-runners win over the tail with worked examples from their own matters; partners listen to other partners. About two-thirds of professionals told Thomson Reuters that GenAI should be applied to their work in some way, so the middle is more willing than most committees assume. What it needs is a version of the tool that fits a normal Tuesday afternoon.

Why do law firm AI programmes fail?

What should a law firm do in its first 90 days?

The full model is a two-year programme. Starting it in order beats planning it in full. A defensible first quarter:

A firm that has done this once has a repeatable unit and a real basis for its next decision. A firm that bought 800 licences has a renewal date.

What the evidence does not tell us

Nobody has finished building this, us included, and a vendor showing you a completed maturity model is selling the model. The published research measures adoption and spend far better than outcomes, and that limit applies to the numbers on this page: most are survey answers, several from a vendor that also sells legal AI.

What the evidence does support: spend has compounded for four years, fewer than one in five professionals can point to a tracked return, and a visible strategy multiplies the odds of benefit. The method follows from that. Pick one matter type, write the playbook, measure four numbers, and decide the next step from the result.

FAQ

What is an AI-native law firm?

A firm whose service design treats machine execution and human judgment as separate inputs, and which prices, staffs and supervises around that split. An AI-enabled firm has licences and occasional use; an AI-assisted firm has rebuilt specific tasks around AI with owners and measured outcomes. The AI-native firm adds structure: literacy as a baseline, a ranked use-case portfolio, codified know-how, integration into existing systems, designed human review and per-use-case measurement.

How should a law firm measure the return on its AI investment?

Per use case, with four numbers: time per matter before and after on the same matter type, override frequency (how often a lawyer changes the output), weekly active use inside the workflow, and margin impact on the matter type where time was saved. Thomson Reuters' 2026 AI in Professional Services Report, published 9 February 2026, found only 18% of professionals say their organisation tracks AI ROI.

Should a law firm buy legal AI or build it?

Buy for general work such as research, summarising, first-pass drafting and translation, where the capability is a commodity. Build only where a workflow is proprietary: a review your firm does differently, a precedent bank nobody else has, or a filing pattern in a court you know unusually well. Blanket licences bought for the whole headcount with no workflow behind them are the most common and most expensive failure.

Where does AI efficiency show up in law firm finances?

On fixed-fee work first. Where the price is agreed in advance, such as conveyancing, probate, incorporations, standard NDA review or contract triage, time saved becomes margin. On hourly work, cutting a four-hour task to one hour cuts the bill too. About 90% of legal spend still flows through hourly arrangements, according to the Thomson Reuters Institute's State of the US Legal Market 2026.

How long does it take a law firm to become AI-native?

The full operating model is a multi-year programme, but a first quarter is achievable. Weeks 1 to 2: pick one fixed-fee matter type and time it from real timesheets. Weeks 3 to 5: write its playbook with the partner who does it best. Weeks 6 to 9: connect it to where that work already happens. Weeks 10 to 13: report time per matter, override frequency, weekly active use and margin, then extend or stop.

Do small and boutique law firms need an AI operating model?

Yes, and the order matters more for them because there is no budget to absorb a wrong decision. The most valuable parts for a small firm are the cheapest: writing down how its best lawyer handles a recurring matter type, choosing one fixed-fee workflow to prove the case, and measuring four numbers. Dedicated build teams and multi-year portfolios can wait until a first use case has paid for itself.