Legal

AI for law firms: what the evidence actually supports

79% of legal professionals now touch AI, but only 8% of solo firms use it seriously. Between those two numbers sits the whole opportunity. Here is what the studies, the court sanctions, and the billing data really say, for firms in BC and anywhere in North America.

By Adi Huric, founder of Most AI Labs·July 2026·9 min read

Legal is the profession where AI hype meets the hardest evidence, in both directions. It is the only industry with a randomized controlled trial showing AI made practitioners 34% to 140% more productive. It is also the industry where AI has put fabricated citations into more than 1,400 court cases, cost one pair of lawyers $110,000 in sanctions, and made a BC family lawyer personally pay opposing counsel’s costs. Both things are true. A small firm that understands both wins.

This article is written for solo and small firms (roughly one to twenty lawyers) in Vancouver, anywhere in Canada, or the US. Every claim is cited. Where a number comes from a vendor, we say so.

Where adoption actually stands

The headline surveys measure different things, so read them precisely. Clio’s 2025 Legal Trends Report (the industry’s largest dataset, from a company headquartered in Burnaby, BC) found 79% of legal professionals now use AI in some capacity. But dig into firm size and the picture flips: only 8% of solo practitioners and 4% of small firms use AI widely or universally. The American Bar Association’s more conservative survey puts overall adoption at 30%, with solos at 18%. Thomson Reuters puts firm-level generative AI deployment at 41% in 2026, nearly tripling in two years.

The direction is identical in every dataset: steep growth, but broad-and-shallow. Most lawyers are dabbling with ChatGPT. Very few firms have actually wired AI into how the practice runs. That gap is the competitive opening, because the evidence below says the payoff is real for the firms that go from dabbling to deploying.

The strongest evidence: a real randomized trial

Most industries have vendor case studies. Legal has an actual RCT. Researchers from the University of Minnesota and University of Michigan ran a randomized controlled trial (Schwarcz et al., published 2026) giving law students realistic legal tasks with and without AI tools. The AI-assisted group was 34% to 140% more productive depending on the task, and the quality of the work improved rather than degraded. An earlier 2023 trial found the biggest gains went to the weakest performers, which is exactly what a small firm without associates wants to hear.

Separately, Clio’s task-level analysis of billing data estimates up to 74% of hourly billable tasks are exposed to AI automation, with admin and document work most exposed. That is not a prediction that lawyers disappear. It is a measurement of how much of a firm’s week is mechanical.

The money math for a small firm

·The average US lawyer bills at $349 per hour but captures only 3.0 billable hours in an 8-hour day, and after realization, roughly 2.6 hours actually get invoiced (Clio 2025 benchmarks). The other five hours are admin, intake, drafting overhead, and chasing.
·At those rates, winning back one extra billable hour per day is worth roughly $87,000 per lawyer per year. That is the honest ceiling for what automation of non-billable work can be worth. Even capturing a quarter of it pays for a serious project.
·Legal is the most expensive customer-acquisition market on the internet: an average of $131.63 per lead from Google Ads, the highest of any industry (WordStream 2026 benchmarks). Every mishandled inquiry is a three-figure bill with no case attached.

The leak nobody talks about: your phone and inbox

Clio ran a secret-shopper study, emailing 1,000 real law firms and calling 500 as a prospective client. In 2019, 60% of firms never answered the email and 27% never picked up or returned the call. When they repeated it in 2024, it had gotten worse: email responses fell from 40% to 33%, phone answers from 56% to 40%.

Put the two numbers together: firms pay $131 per lead, then fail to answer more than half of the resulting contact attempts. This is why the highest-ROI first AI project for most small firms is not a research tool. It is intake: 24/7 answering, qualification, and scheduling. Clio’s platform data (correlation from their own users, not a trial, but the largest dataset available) shows solos using client intake and portal tech report 53% higher revenue and 48% more leads.

The firms winning with AI are not the ones with the fanciest research tool. They are the ones where every inquiry gets answered in minutes, every document starts from a draft, and every citation still gets checked by a human.

Where AI pays off first, ranked by evidence

·1. Intake and responsiveness. The secret-shopper data above is the proof of the problem; the intake-tech revenue correlation is the proof of the fix. 27% of solos already run virtual receptionists (Clio 2025). Start here.
·2. Document drafting and automation. Backed by the RCT productivity gains. Vendor claims go further (Clio markets “up to 80% of drafting time saved”, and that is a vendor number), but even the conservative trial evidence justifies the project for any firm producing repetitive documents.
·3. Research and summarization. The most-used application (80% of AI-using firms per Thomson Reuters) and genuinely useful, but read the risk section before trusting it. This is assistance, not delegation.
·4. Timekeeping and billing capture. Plausible and widely marketed, but we found no credible independent measurement of the gains. Vendor claims only. Treat it as a bonus, not a business case.

The risks and the rules. Do not skip this section.

AI-fabricated citations have now been caught in more than 1,400 court decisions worldwide, growing at two to three new cases per day. In April 2026 a US federal judge issued a $110,000 sanction, the largest yet. And this is not a “just buy the legal-grade tool” problem: a Stanford study found even dedicated legal research AI hallucinated on 17% to 33% of test queries depending on the product.

Canada has its own casebook. In Zhang v. Chen (2024 BCSC 285), a BC family lawyer cited two ChatGPT-invented cases and was ordered to personally pay opposing counsel’s wasted costs. The Federal Court of Canada now requires a declaration, in the first paragraph of a filing, when content is AI-generated. The Law Society of BC published guidance covering competence, confidentiality, and responsibility for AI output. In the US, ABA Formal Opinion 512 says the same things with teeth: get informed consent before client information goes into self-learning AI tools, verify everything, and do not bill clients for the hours AI saved you.

Watch out for this

Two non-negotiables for any firm, any size: never put confidential client information into a consumer AI tool without informed consent and a tool configured not to train on your data, and never file anything containing a citation a human did not verify. Every sanction in the database traces back to breaking one of these two rules.

The BC and Canada notes

·Law Society of BC has published generative AI guidance; Alberta’s law society ships a full playbook; the Federal Court requires AI disclosure in filings. Check your court’s current practice direction before filing anything AI-assisted.
·Canadian small-firm adoption data is thin (most Canadian surveys cover firms of 20+ lawyers, where 80% are exploring AI but only 7% have fully implemented). We flag this rather than pretend precision exists.
·The world’s biggest legal-practice dataset comes from a BC company: Clio, headquartered in Burnaby. The benchmarks above are as relevant on West Broadway as in Chicago.

The honest bottom line

·The productivity evidence is real and unusually strong: an RCT showing 34% to 140% gains with quality improving.
·The first project should almost always be intake: answered calls and same-hour responses, where firms are measurably bleeding $131-per-lead marketing spend.
·Drafting automation is second, with trial-grade evidence behind it.
·Research AI is useful but must be verified, because even the legal-grade tools hallucinate.
·The rules are settled enough to act: LSBC guidance, ABA Opinion 512, and court disclosure requirements are checklists, not blockers.

If you run a firm and want to know which of these applies to your practice specifically, that is exactly what our AI consulting work does, and it starts with the free 7-day audit: we map where your hours and leads actually leak, and show you the math before you spend anything. Fixed price, no retainers, and you own whatever gets built.

Wondering what an ad budget buys here? Our free lead calculator runs the verified benchmark math for this industry: a legal lead runs $131.63 on Google, the most expensive of any industry, and firms convert about 17.6% of intake leads.

Sources

If any of this is your week

Start with the 7-day audit.

7 business days. A real document. Yours to keep — whether you hire us or not.

Start with the 7-day audit Or talk to the founder first

Not all AI is the same. Which model fits which job?

The tier system behind every tool your firm will evaluate.

Read→

Buyer Guide

What does a custom AI agent actually cost?

The budget conversation, with real numbers.

Read→