AI Is Now Standard in Law Firms. The ROI Still Depends on Who Runs It.

AI is no longer the future of legal work. It is the present. Clio’s 2025 Legal Trends Report found that 79% of legal professionals now use AI in some form, and that adoption in mid-sized firms jumped from 19% to 93% in a single year. Clio also found something that should get every managing partner’s attention: firms with wide AI adoption are nearly three times more likely to report revenue growth.

So the tools work, and they can pay off. But that same report found that 53% of legal professionals say their firm has no AI policy, or they do not know if one exists. And the risk side of the ledger is growing fast. Stanford researchers testing purpose-built legal research tools in 2025 found they still fabricated information in a meaningful share of queries, from around 17% for Lexis+ AI to roughly a third for Westlaw’s AI-assisted research, with general chatbots like GPT-4 far worse. Public trackers of the problem have now logged more than 1,300 court filings containing AI-invented citations, close to 500 of them from licensed attorneys, and judges are losing patience as sanctions climb.

Read those numbers together and the lesson is clear. The firms getting revenue growth from AI and the firms getting sanctioned are often using the same tools. The difference is not the software. It is the person running it.

The legal AI stack, and why every layer needs an operator

Whatever your firm has adopted, it almost certainly falls into one of these categories:

• AI legal research: Lexis+ AI, Westlaw Precision with CoCounsel, Vincent by Clio (from Clio’s vLex acquisition)
• Drafting and legal assistants: Harvey, Legora, CoCounsel, Clio Work
• Practice management with built-in AI: Clio (Clio Work and Manage AI), Smokeball, MyCase
• eDiscovery and document review: Everlaw, Relativity aiR, DISCO
• Contract review and analysis: Spellbook, Luminance, Kira
• Intake, client screening, and transcript or deposition summarization

Every one of these does the same thing: it does the heavy lifting on volume and hands a draft to a person. None of them decides whether the output is right. That judgment call, on every layer of the stack, is human. It is also exactly where the 17% to 33% hallucination rate gets caught, or does not.

A practical framework for getting legal AI ROI

1. Point AI at the work where a tool plus a person compounds

Do not try to bolt AI onto everything. Start with the high-volume, judgment-heavy work:

• Document review and discovery, where the tool surfaces and summarizes and a person confirms every result that matters
• Legal research, drafted by the tool and verified against primary sources before it reaches an attorney
• First drafts of correspondence, memos, and routine filings, prepared for attorney review
• Intake and client screening, where AI handles volume and a person handles the judgment
• Summarizing depositions, medical records, and long case files

These are the workflows where an hour of the tool plus a skilled operator returns three or four hours you used to bill by hand.

2. Put a skilled operator behind every tool

This is the step most firms skip. A capable operator asks the sharp question, catches what Westlaw or Harvey got wrong, and turns a rough output into finished work an attorney can rely on. That person does not need to be a licensed attorney. They need to be genuinely proficient with the specific tools your firm uses and carry the judgment to know when an answer is off.

Here is where the math trips firms up. AI is not cheap. Between per-seat licenses, the platform fees baked into your practice management stack, implementation, and training, your firm is already carrying real cost before a single matter moves faster. The instinct is to hire a full-time local specialist to run it all, and that stacks a second expensive line item, salary, benefits, overhead, on top of the first. Now you are paying premium rates twice, once for the tool and once for the person, and the ROI you were chasing gets thin fast. The goal is to add the operator without compounding the cost.

3. Build verification into the workflow, not around it

This is the step the sanctions are punishing. With even the best legal tools fabricating citations in a real share of queries, “the software checked it” is not a defense. Decide, in writing, who verifies what: AI-assisted research checked against primary sources, AI-drafted filings signed off by an attorney, every citation confirmed before anything is filed. A firm that skips this is one bad brief away from being the next name in the sanctions tracker. A firm that gets it right turns AI from a liability into leverage.

4. Give the firm an AI policy and someone to own it

More than half of firms have no policy. That is not a compliance footnote, it is the reason adoption stalls. Write down which tools are approved, what data can go into them, and who reviews the output. Then give one person ownership of it. This alone separates the firms that compound gains from the ones that quietly stop using the tool after the first scare.

5. Measure what moves, and keep operators current

Track turnaround on the work you pointed AI at, cost per matter, and the capacity your team picked up. If the number has not moved in a quarter, adjust the workflow or the operator behind it. And because the tools change every few months, keep the people running them current, or hire people whose proficiency is kept current for you.

The shortcut most firms miss

Every step above depends on one thing: the right person in the operator seat, fluent in your tools and carrying the judgment to check them. That is the hard part, and it is the part firms try to solve last.

This is where we help, and where the cost problem solves itself. ShiftSixOS places Philippines-based professionals who are already proficient with the AI tools your firm uses, vetted for the judgment to verify output rather than just generate it, and matched to the way your firm works, at a fraction of the cost of a local specialist hire. You get the operator your tools need without stacking a second premium salary on top of your AI spend. We are not another platform to buy. We are the skilled people who make the platforms you already bought actually pay off. Based at One World Trade Center in New York with our team in the Philippines, the model is simple: you manage the day-to-day work, and we handle recruitment, HR, payroll, and compliance, with dedicated US-based support behind you.

So you can put your budget where it moves the work forward, into better tools and skilled people to run them, and keep your firm on the revenue-growth side of the AI divide instead of the sanctions side.

If you want to see what that looks like for your firm, take a look at our AI-Ready Talent page, or book a consultation and we will show you the talent your firm can hire.

Sources: Clio 2025 Legal Trends Report; Stanford legal AI hallucination study (2025); Stanford CIS.