How AI's Price Tag Is Changing LegalTech for the Better

Jan de Vries · 2026-08-25

Remember the early days of agentic AI? It felt like the wild west. Everyone was throwing money at the latest shiny tool, dazzled by demos and promises. But that initial exuberance has definitely waned. Now, the conversation in LegalTech—and honestly, across the board—has shifted. It's less about potential and more about practicality. Vendors and customers are sitting down and talking about real business cases and something much more concrete: return on investment, or ROI. And honestly? This shift in tone is a breath of fresh air. After a period of unfettered enthusiasm that led to some wild spending, it's about time. Let me give you a couple of real-world examples. 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. Then there's the whole 'tokenmaxxing' trend that took hold. Companies were treating high AI token usage as a badge of developer performance. Amazon reportedly had to drop its internal token-usage leaderboard because employees started optimizing for usage rather than actual results. They were gaming the system. In early August, Microsoft sent an internal email to its engineers making the company's stance crystal clear: outcomes, not tokenmaxxing, should be the objective. They also put caps on employee AI spending. This is all a natural part of the technology adoption cycle. As these advanced models get more complex, they also get more expensive to use. So, of course, discussions about cost and ROI become front and center. ### Getting Savvy About the Bottom Line Flashy sales demos worked for a minute there. But buyers are getting smarter. They're adopting more rigorous approaches to evaluating legal AI. Some are now walking into sales meetings with their own documents and specific questions, stress-testing products against realistic scenarios they face every day. Demanding evidence of ROI from the start isn't being difficult; it's being smart. It creates clearer expectations about the value a tool should deliver and makes it way easier to assess later on if the implementation is actually working. Here's the thing: many in-house legal teams haven't traditionally tracked baseline metrics like turnaround times or task volumes. AI tools can actually create greater visibility into these measures, making outcomes easier to quantify in the first place. And it's not just legal teams anymore. Business units are adopting legal technology too. In some cases, the gains are even easier to measure because these teams are directly tied to revenue and costs. - For example, one insurance firm's claims-handling team used agentic AI to boost their output. They went from processing about 150 claims per person per month manually to around 700 in the same period. That's a game-changer. ### The Vendor's Dilemma: Scaling Sustainably Companies are now demanding ROI and solid proof points after getting those first shocking AI bills. For the vendors themselves, AI costs are a massive headache. Their solutions depend on large language models (LLMs), and the processing costs for those LLMs are no joke. For LegalTech startups, these costs directly impact whether their business model is viable as they try to scale and monetize. It's a tightrope walk. Typically, vertical AI solutions charge on a per-seat basis. Procurement teams like that—it gives them cost clarity. But this creates a structural tension. The vendor's costs are often token-based (tied directly to usage), but they're charging a flat seat fee. As usage grows, their margins get squeezed. Switching to a pure usage-based pricing model to protect margins sounds logical, but it's risky. It can fuel customer fears of spiraling, unpredictable costs and send them running to competitors. ### Making Technically-Smart Choices Is Key So, what's the path forward? To face these issues head-on, AI vendors need to get technically clever about how they build their solutions. They need the right tool for the job, not just the biggest hammer. It's essential to match the right-sized model to each specific task instead of defaulting to the most powerful (and most expensive) option for every single query. You wouldn't use a industrial crane to hang a picture frame, right? The same logic applies here. Multiple specialized agents can be built to empower different workflows, and each one could be powered by a different LLM, or a different version of a model, optimized for cost and performance. This thoughtful, layered approach is how LegalTech will mature from a cost center into a genuine value driver. As one industry observer recently noted, *'The end of the hype cycle is where real business gets done.'* We're finally moving past the spectacle and into the substance. That's a win for everyone who wants technology that actually works and pays for itself.