Rising AI costs have become a talking point for LegalTech, and other industries, as the initial exuberance over agentic AI has waned. These days, vendors and customers are more likely to discuss specific business cases and return on investment (ROI) rather than get carried away with AI’s potential.
This shift in tone is welcome news, especially after a period of unfettered enthusiasm that led to some wild spending. In April, for example, Uber had reportedly blown through its AI budget for the whole year and capped employee spending on agentic tools like Claude Code and Cursor.
Costs also became a problem as the “tokenmaxxing” trend took hold, treating high AI token usage as a proxy for developer performance. Amazon reportedly dropped its internal token‑usage leaderboard after employees began optimising for usage rather than results.
And in early August, according to an internal email, Microsoft told engineers that outcomes, not tokenmaxxing, should be the objective, while also capping employees’ AI spending.
This shift is a natural development given where we are in the technology adoption cycle. As advanced AI models have become more complex, and more expensive, for companies to use, these discussions about costs and ROI have become more pertinent.
Flashy sales demos may have worked for some companies during the early days of agentic AI. That is becoming less common as buyers adopt more rigorous approaches to evaluating legal AI. Some now bring their own documents and questions to sales meetings to stress-test products against realistic scenarios.
Demands for evidence of ROI can also be a positive sign for a project. Asking these questions from the outset creates clearer expectations about value and makes it easier to assess whether an implementation is delivering the intended results.
Many in-house legal teams have not traditionally tracked baseline metrics such as turnaround times or task volumes. AI tools can create greater visibility into these measures, making outcomes easier to quantify.
Legal technology is also being used by business units, rather than only by in-house legal teams. In some cases, the gains may be easier to quantify because these teams are more directly connected to revenue, costs and profit-and-loss measures. For example, one claims-handling team at an insurance firm increased the number of claims processed per person from around 150 a month manually to around 700 in the same period using agentic AI.
After grappling with AI costs and large bills, companies are seeking ROI and proof points of legal AI. For the vendors, AI costs are also a pressing issue because of the processing costs of the underlying large language models (LLMs) that their solutions depend on.
For LegalTech startups and vendors, these costs play into the viability of their business models as they reach scale and seek to monetise their solutions.
Typically, vertical AI solutions have charged on a per-seat basis to meet expectations of procurement teams that want clarity over future costs. This creates a structural tension between rising token-based usage of the LLMs and costs passed on to customers.
On the other hand, switching to a usage based pricing model to protect margins risks further fuelling customers’ fears of spiralling costs and losing them to competitors.
To face these issues head on, AI vendors need to make smart technical choices about how they build their solutions. They need specific tools for the task at hand; not every LLM is suitable, or necessary, for the LegalTech solution they are building.
It is essential they match the right-sized model to each task instead of defaulting to the most powerful (and most expensive) option for every type of query.
Multiple agents can be built to empower different workflows, and each of these could be powered by a different LLM, or different version of the LLM. The usage of the LLM needs to be more targeted; there is no need to bring a bazooka to a query when a penknife would do.
The issue of rising AI costs has brought into sharp focus the solid foundations that legal AI needs to be built upon, and how LegalTech startups need to build in a way that generates returns for them, and their investors. The focus on AI costs and ROI is good news for LegalTech as a whole, and for the customers that use it.
Source: EU-Startups
