I hear a version of the same conversation with law firm leaders almost every week.

They’ve bought the AI tools. They’ve run the pilots. And yet, when I ask how it’s going, the answer is usually some version of “fine, but not what we expected.” Lawyers are still re-prompting, still verifying everything by hand, still not fully trusting what comes back. 

For a while, I assumed this was a model problem – that the next release, the next context window, the next benchmark score would close the gap. It won’t. The gap isn’t in the model. It’s in what the model can see. 

Legal work has never been about finding information. It’s about understanding how that information connects – to a client, a matter, a prior negotiation, a risk the firm has already priced in once before. Lawyers carry that connective tissue in their heads, in hallway conversations, in the institutional memory of whoever happened to work the file. AI tools don’t have access to any of that. They see whatever gets uploaded into a session window, and nothing else. So they answer confidently, and sometimes wrong, because they’re filling gaps we never gave them the context to fill correctly. 

The numbers back this up more than I’d like. McKinsey finds that fewer than 10% of enterprises have scaled agentic AI to real value, and eight in ten point to data limitations as the reason. A 2025 Stanford RegLab study found that even purpose-built legal AI research tools hallucinate between 17% and 33% of the time. That’s not a model quality issue. That’s a context issue, and it’s exactly why we built the industry’s first Legal Context Graph. 

Colleen Baehrend, a lawyer and NetDocuments legal solutions director, and I put together a practical guide for firms trying to figure out where they actually stand – not in theory, but in the day-to-day reality of how their AI tools perform. 

The Legal Context Graph guide on AI Spend to AI Impact walks through: 

• What a Legal Context Graph is and how it’s different from the taxonomies and knowledge graphs firms have tried before 

• What changes for associates, partners, and IT once matters are truly connected, and how that plays out differently in litigation, transactional work, estate planning, and firms of every size. 

• A checklist with 10 questions organized around cost, accuracy, usability, governance, and institutional knowledge, that you can sit down with your team and answer candidly in 20 minutes. There’s no passing score – the point isn’t to make anyone feel behind. It’s to surface, specifically, where the gap between what you’re spending on AI and what you’re getting back from it is actually coming from, so the next conversation about AI investment is grounded in something real instead of another vendor demo. 

AI models will keep changing. Every one of them will eventually be replaced by something better. What won’t change is that your firm’s context – your matters, your relationships, your institutional judgment – has to live somewhere durable enough to outlast all of it. That’s the foundation this guide is about. 

If any of this sounds familiar, start with the checklist and see what it turns up. 

Get the Complete Guide > From AI Spend to AI Impact – A Law Firm’s Guide to Getting the Foundation Right 

The same context problem shows up just as often inside corporate legal, compliance, and risk teams at regulated companies. We’re putting together a version of this guide built specifically for that audience. Coming soon.