How AI's Soaring Costs Are Forcing LegalTech to Get Real

Jan de Vries · 2026-08-25

You know how it goes. The initial hype around any new tech is always deafening. Everyone's talking potential, painting visions of the future. But eventually, the bills start rolling in. That's exactly what's happening with agentic AI in LegalTech right now. The conversation has shifted. It's less about 'what could be' and more about 'what's the actual return?' And honestly, that's a healthy sign. It means we're moving from a playground to a proper workplace. This new focus on costs and ROI is a natural, even necessary, phase. We saw this coming. As AI models get more sophisticated, they also get more expensive to run. Companies that jumped in headfirst are now checking their bank statements. Back in April, Uber reportedly blew through its entire annual AI budget and had to cap employee spending on tools like Claude Code and Cursor. That's a real-world wake-up call. Then there was the weird 'tokenmaxxing' trend. For a minute, some teams treated high AI token usage as a badge of honor—like burning more compute meant you were working harder. Amazon reportedly had to drop an internal usage leaderboard because people started optimizing for token count, not results. Microsoft sent a similar internal memo in August, telling engineers to focus on outcomes, not token usage, and capping spending. The message is clear: smart use, not just heavy use. ### Getting Savvy About ROI Remember those flashy sales demos? The ones that felt like magic shows? They're losing their power. Buyers are getting tougher. They're walking into meetings with their own documents and their own tricky questions. They're stress-testing products against the exact scenarios they face every Monday morning. That's a good thing. Demanding proof of ROI from the start isn't being difficult; it's being smart. It sets clear expectations. It makes it easier to look back six months later and say, 'Yes, this was worth it.' Many legal teams never really tracked baseline metrics like turnaround times or task volume before. AI tools can shine a light on that, making the impact crystal clear. And it's not just the legal department using this tech anymore. Business units are getting in on the action. Sometimes, the gains are even easier to spot there because they're tied directly to revenue and costs. Think about an insurance claims team. One firm saw a person's output jump from processing about 150 claims a month manually to around 700 using agentic AI. That's a number you can take straight to the CFO. ### The Vendor's Dilemma: Scaling Without Stumbling Here's the twist: while clients are demanding ROI, vendors are sweating over their own costs. The big expense? The processing fees for the large language models (LLMs) that power their solutions. For LegalTech startups, this isn't just an operational cost—it's a question of whether their entire business model can survive at scale. The typical playbook has been per-seat pricing. Procurement teams love it because it's predictable. But there's a tension there. As customer usage (and those underlying LLM tokens) goes up, the vendor's costs skyrocket, but the per-seat price stays flat. That eats margins. So, do they switch to usage-based pricing to protect themselves? That's risky. It can spook customers who are already nervous about runaway costs and send them straight to a competitor. It's a tightrope walk. ### Making Technically-Smart Choices So, what's the way forward? Vendors have to get clever under the hood. They can't just default to the biggest, most powerful (and most expensive) LLM for every single task. That's like using a rocket launcher to swat a fly. - **Match the model to the mission.** Not every legal task needs GPT-4. Some workflows might be perfectly served by a smaller, cheaper, more specialized model. - **Build a team of agents.** Think of it like assembling a specialist team. You might have one agent for document review powered by one LLM, and another agent for contract summarization powered by a different one. Use the right tool for each job. This cost scrutiny isn't a buzzkill. It's the market maturing. It's forcing everyone—vendors and customers—to be smarter, more efficient, and more focused on what truly delivers value. And in the end, that's how you build technology that lasts, not just technology that trends.