Strategy before technology
As I happen to speak with law firm partners often, the AI topic is always on the table. And the discussion often centres on which AI tool is a better fit for their firm. It is a reasonable question, and the market has trained everyone to ask it. The harder question that sits underneath has had less attention. Before any tool is bought, what does the operating model need to look like?
AI productivity gains are real, and they are accelerating. The capabilities available now were not available twelve months ago, and the curve is steeper than anything the legal industry has managed to absorb before. Any honest reading of where this is heading has to start from there.
What that productivity does inside a practice is an interesting question. The technology lands inside whatever structure is already there. A vertical hierarchy routing every decision through a partner check. A tech stack assembled over twenty years from offline and cloud tools that do not speak to each other. Billing built around the hour. Intake designed for fax and email. None of this disappears because someone bought a Copilot licence. The structure decides what happens next.
The evidence for this is in the public record. The hallucination cases database maintained by Damien Charlotin at HEC Paris lists over 1,455 court decisions where AI-generated content turned out to be fabricated, most of them from situations where someone applied the technology without operational controls. The Stanford RegLab study in the Journal of Empirical Legal Studies found that purpose-built legal AI tools, including those marketed as hallucination-free, produced incorrect citations on 17 to 33 percent of queries. Verification time has to live somewhere in the operating model. From 2 August 2026, the EU AI Act makes the deployer responsible for human oversight, accuracy thresholds, and documentation — operating model requirements, not procurement ones.
Most procurement conversations skip past all of this. The pitch is about output speed, draft quality, time saved per matter. The redesign of how work actually moves through the practice — who reviews what, when, with what audit trail, against which cost — is not on the vendor slide. It cannot be. That decision sits with the buyer.
Most competitive benchmarking compares the firm to its peers. That is fair enough. Peers carry the same legacy systems, the same partner-track economics, the same intake habits, the same procurement timelines. They will move at roughly the same speed and arrive at roughly the same place.
The competitors that change the picture sit outside the peer set. AI-native entrants are building entire operations around the workflows the technology enables, rather than retrofitting tools to workflows designed for an earlier era. They carry no twenty-year-old document management system. Their cost base is different. Their pricing assumptions are different. They are smaller now and easy to dismiss, but they are also winning mid-market clients that established practices have historically considered safe.
Strategy makes procurement straightforward. A clear view of where the operations need to be in five years narrows the tool field on its own. The reverse rarely works — buying tools to discover the strategy tends to produce neither.
Some of these entrants will take work that established practices considered safe. The shift is happening outside the frame most firms are watching.