Law firms report 80% time savings on due diligence using AI. But can you trust a model with legal liability?
Law firms spent decades resisting technology that threatened the billable hour, and yet legal has quietly become one of the most enthusiastic adopters of AI in any professional field, with major firms now reporting 60 to 80 percent time savings on document review, contract analysis, and due diligence using tools like Harvey, Thomson Reuters' CoCounsel, and Luminance.
Contract review, from six hours to ten minutes
Contract review is the most mature use case in the category, and the mechanics explain why. A 200-page commercial agreement traditionally required a junior associate to read every clause manually, extracting payment schedules, liability caps, termination triggers, and change-of-control provisions, then flagging anywhere the language deviated from the firm's standard templates, a process that typically ate four to six hours per document. Current AI systems perform the same extraction and flagging in under ten minutes, producing a summary memo that captures the same key terms and deviations a human reviewer would have caught, at a small fraction of the time cost.
That speed doesn't just save money, it changes what's economically feasible to review at all. Deals that previously only justified spot-checking a sample of contracts because full manual review was too expensive can now have every single document reviewed in full, which surfaces risks that sampling-based review would have missed entirely.
Due diligence at a scale no human team could match
The due diligence use case for M&A transactions shows the same dynamic at a much larger scale. In one recent five billion dollar acquisition, an AI system processed 3.2 million documents inside a virtual data room in 48 hours, identifying 47 material risks that would have taken a team of 50 associates several weeks to surface working manually. Among those findings was a previously overlooked environmental liability worth 120 million dollars, exactly the kind of buried risk that a time-pressured manual review, constrained by how many documents a team of humans can physically read before a deal deadline, is statistically likely to miss.
That example captures both the promise and the stakes of the technology in one story: the AI didn't just save time, it caught something material that a slower, human-only process plausibly would have let through, with a nine-figure price tag attached to the miss.
The accuracy ceiling that still requires a human in the loop
None of this has eliminated the need for a lawyer to check the AI's work, and the accuracy numbers explain why. Current AI legal tools achieve 94 to 97 percent accuracy on routine document review tasks, which is genuinely comparable to a senior associate's performance, but still sits below the 99-plus percent threshold that partners expect for anything that becomes final work product attached to a firm's name and malpractice exposure. The standard practice that has emerged across the industry is to use AI for the first pass, capturing the bulk of the extraction and flagging work, and have a human lawyer review and sign off on the AI's findings before anything goes to a client, which is a meaningfully different workflow from either pure automation or pure manual review.
This is also where the legal liability question becomes concrete in a way it isn't for most other industries adopting AI. A missed clause in a contract review isn't just an inconvenience, it can be the basis of a malpractice claim, which is why the human sign-off step is unlikely to disappear even as the underlying accuracy improves further.
What this does to the billing model
The billable-hour structure that has defined law firm economics for generations is under direct pressure from this shift, because clients are increasingly unwilling to pay hourly rates for work that both sides know is now largely automated. Firms that treat AI purely as a cost-cutting measure while preserving the old hourly billing model are going to face client pushback as soon as clients understand how much of the work is machine-generated. The firms actually positioned to thrive are the ones restructuring around delivering more value faster rather than defending an hourly rate for work AI now does in minutes.
For legal tech teams and firms evaluating which model to build on, the differences between LLMs on legal reasoning, clause extraction, and long-document summarization are substantial enough to matter, and Vincony's Model Playground lets teams compare those models side by side on exactly these tasks before committing to one as the backbone of a review pipeline that will eventually carry real liability.