Why the AI Cost Reckoning is Actually Good for LegalTech

Jan de Vries · 2026-08-25

Remember when every conversation about LegalTech felt like a breathless hype session about AI's potential? Well, that's changing. And honestly? It's a good thing. Rising AI costs have become a major talking point across industries, including legal tech. The initial wave of pure excitement over 'agentic AI' has settled into a more mature, practical discussion. These days, vendors and customers are far more likely to talk about specific business cases and hard return on investment (ROI) than get swept away by possibilities. This shift isn't a sign of failure—it's a sign the industry is growing up. This new tone is actually welcome news. It comes after a period of what you might call unfettered enthusiasm that led to some pretty wild spending. Back in April, for example, Uber had reportedly blown through its entire annual AI budget and had to cap employee spending on popular agentic tools. They weren't alone in hitting a cost wall. ### When Measuring Usage Backfires Costs became a real problem as a strange trend called 'tokenmaxxing' took hold in some tech circles. The idea was simple but flawed: treat high AI token usage as a proxy for developer productivity. It led to some odd incentives. Amazon reportedly dropped an internal token-usage leaderboard after employees started optimizing for high usage rather than actual results. They were gaming the system, not building better products. Microsoft sent a similar message in early August. An internal email told engineers clearly that outcomes, not tokenmaxxing, should be the only objective. They also put caps on employee AI spending. This is all a natural part of the technology adoption cycle. As the AI models themselves have become more complex and expensive for companies to use, these tough conversations about cost and value were bound to happen. ### Getting Seriously Savvy About ROI Flashy sales demos might have sealed the deal in the early days. That's becoming much less common. Buyers are getting rigorous. Some now bring their own legal documents and specific questions to sales meetings to truly stress-test a product against their real-world scenarios. They're not just watching a scripted show anymore. This demand for evidence is a positive sign for any project. Asking the hard ROI questions from the start creates clearer expectations about the value you're buying. It also makes it way easier later to check if the implementation is actually delivering the intended results. Here's the thing: many in-house legal teams haven't traditionally tracked baseline metrics like turnaround times or task volumes. It just wasn't part of the culture. Ironically, AI tools can create greater visibility into these exact measures, making their own outcomes easier to quantify. It's a self-fulfilling cycle of improvement. And it's not just legal teams anymore. Business units are adopting legal technology directly. In these cases, the gains can be even clearer to measure because these teams are directly tied to revenue, costs, and profit-and-loss statements. > For example, one insurance claims team saw their output skyrocket from about 150 claims processed per person per month manually, to around 700 in the same period using agentic AI. That's a tangible impact you can take to the bank. ### The Vendor's Tightrope Walk After getting those shocking bills, companies are demanding ROI and clear proof points. For the LegalTech vendors selling these solutions, AI costs are a massive internal issue too. Their own solutions depend on underlying large language models (LLMs), and the processing costs for those models are real and significant. For startups and vendors, these costs strike at the very heart of their business model's viability as they try to scale. There's a real tension here: - **The Per-Seat Model:** Most vertical AI solutions charge per user. Procurement teams love this—it offers cost clarity and predictability. - **The Hidden Cost:** This creates a structural squeeze. The vendor's own costs are based on token usage, which can rise unpredictably, but they're locked into a fixed per-seat price for the customer. Switching to a pure usage-based model to protect margins is risky. It can fuel a customer's worst fear: spiraling, unpredictable costs that send them running to a competitor. ### Making Smarter Technical Choices So, what's the path forward? To face these issues, AI vendors need to get technically clever. They need the right tool for each specific job, not just the biggest hammer for every nail. Not every LLM is suitable, or necessary, for every function within a LegalTech platform. It's essential to match the right-sized, right-priced model to each specific task instead of defaulting to the most powerful (and most expensive) option for every single query. You can build multiple specialized agents for different workflows—drafting, research, review—and each can be powered by a different, optimally chosen LLM. This architectural efficiency isn't just good engineering; it's becoming a critical business survival skill in the new cost-conscious landscape. The era of easy AI money is over, and that's forcing everyone to build smarter, more sustainable solutions that deliver real value.