Section 1
The billing model decides the roadmap
Fixed-fee and subscription work is where automation is unambiguously good news. Every hour removed is margin retained, and the client experience improves at the same time. Volume work, standard agreements, compliance filings, and routine incorporation sit here naturally. Hourly matters are the opposite. Efficiency reduces the invoice unless the firm changes how it prices, and it will not change how it prices because a pilot suggested it. This is why so many legal AI deployments stall after a successful trial. The technology worked. The economics did not. Decide the pricing question before the pilot, or accept that adoption will be voluntary and therefore partial. The sequencing discipline for this is in [Building an AI Automation Roadmap for Your Startup](/blog/building-an-ai-automation-roadmap-for-your-startup).
Section 2
Retrieval beats generation in legal work
The uses that hold up in a firm are mostly about finding and comparing, not writing. Locating the relevant clause across a thousand contracts. Comparing an incoming draft against your standard position and flagging every deviation. Summarizing a matter file for a partner who is taking it over. Extracting obligations and dates from an executed agreement into something that can be tracked. Each of these has a verifiable answer sitting in a document you own. That is the crucial property. A summary can be checked against the source in seconds. Generated reasoning about the law cannot be checked as quickly, which is why the failures that reach the press involve invented authority rather than a mis-summarized clause.
Section 3
Nothing goes out without a lawyer reading it
This rule is not negotiable and it is also not a limitation. It is the product. The firm's value is a qualified person taking responsibility for advice. Automation that shortens the path to that person's review is valuable. Automation that removes them is not a cheaper law firm, it is a different and uninsurable business.
Section 4
Rolling it out inside a firm
Start with one matter type and one clearly bounded task. Contract review against a playbook is the usual best first choice, because the firm already has the playbook written down and disagreements are easy to adjudicate. Baseline it in hours and in quality: how long a first-pass review takes today, how many issues a senior lawyer catches that a junior missed, and how often a matter comes back with a problem. Then run the system on matters already completed, where you know the answer. That backtest is the cheapest possible evaluation and it costs no client anything. Only after it holds should the tool touch live work, and the first live use should be a second pair of eyes rather than the first.
Section 5
Privilege, confidentiality, and the citation problem
In a law firm the governance question is narrower and sharper than the NIST language of design, evaluation, and use suggests: does sending this material to a third-party system affect privilege or breach a confidentiality undertaking. Get that answered by someone qualified before a pilot, not after. The operational controls follow from it. Client matter data stays inside tools with contractual guarantees on retention and training. Conflicts and confidentiality walls must be respected by any system that searches across matters, which is a real technical requirement and frequently overlooked. Every generated citation is verified against the source before it leaves the firm, without exception. And log what was automated on each matter, because a professional negligence claim years later will ask.
Section 6
Measuring it without fooling yourself
Hours saved is the wrong headline unless you know what happened to those hours. Track realisation rate, matter cycle time, write-offs, first-pass review time, issues caught in review, and the share of automated output a lawyer had to correct. Write-offs are the honest number in a firm. Time written off is work that was done and could not be billed, and it is where inefficiency actually shows up in the accounts. Review quarterly with the partners who own the matters. The comparison with other regulated professions is covered in [AI Automation in Healthcare Startups](/blog/ai-automation-in-healthcare-startups), and [Storytelling in the Age of AI and Automation](/blog/storytelling-in-the-age-of-ai-and-automation) is useful on how to describe the change to clients.