Law firms across Europe are calling their IT departments in a panic. The reason: Legora, a widely used legal AI platform, quietly switched from a flat monthly subscription to pay-per-use pricing. Some firms saw their bill double. Others saw it triple. It sounds like a legal-industry problem, but the lesson applies to any company currently paying a flat fee for AI software.
What actually happened at Legora?
Legora charged a fixed price per user per month for years, typically between $110 and $325 (roughly €100 to €300), depending on the plan. Last month, the platform switched to usage-based billing. You no longer pay for the seat. You pay for what you actually do with it.
The impact was immediate. Elgar Weijtmans, a former IT lawyer and now head of technology at HVG Law, put it this way:
“I'm getting calls from firms in a panic, asking how they're supposed to budget for this.”
Elgar Weijtmans, head of technology, HVG Law
Legaltech analyst Douwe Groenevelt of Datacation has run the numbers: firms that want to keep using the best available models should expect their AI costs to run two to three times higher than before, according to reporting in the Dutch financial daily Het Financieele Dagblad. The pattern is not unique to one platform or one country. Uber blew through its entire 2026 AI budget by April and now caps individual employees at $1,500 a month in token spend, according to reporting by industry analyst Josh Bersin. When Anthropic moved automation platform Workato to token-based pricing in May, the company's bill jumped seven-fold in a single day.
Why is the flat fee disappearing?
The reason is simple: AI agents don't cost what a human costs. An employee opens a tool, asks a question, reads the answer. That might burn a hundred tokens. An AI agent that independently reviews a contract, cross-checks case law, and drafts a summary can burn tens of thousands of tokens on a single task. Those costs scale with the number of tasks, not the number of people using the tool.
For a vendor like Legora, the flat fee was quietly losing money on its heaviest users. Here's the thing: the same math applies to any AI tool built around agents, whether that's a legal platform, a marketing assistant, or a code reviewer.
Think of it like a gym membership. A flat monthly fee works fine as long as most members show up twice a week. But if half of them suddenly start training three hours a day, the gym has to either raise prices or go under. That's exactly what's happening with AI agents: usage per person keeps climbing, and the flat-fee model can't absorb it.
Which tools have already made the switch?
Legora isn't the first, and it won't be the last. The pattern has been building for months.
- Anthropic bills API access to Claude by the token. The $20-a-month Pro plan has a built-in usage cap; go past it and you either pay overage or move to the $90 or $180 Max plan.
- GitHub Copilot introduced a pay-per-use option alongside its flat-rate license in 2025. Once you exceed the standard allowance of code completions, extra requests are billed individually.
- Cursor runs a hybrid model: a flat subscription with a cap on fast requests, then throttling or top-up fees after that.
Groenevelt expects other major legal-AI vendors to follow the same path. But this isn't confined to law firms. Amazon, Walmart, Cisco, and Meta are all capping internal AI budgets and pushing staff toward cheaper models, warning against what one executive called "AI for the sake of AI." Hybrid pricing, a base subscription plus usage overage, is now the industry default: 41 percent of AI vendors use it, according to Bessemer Venture Partners' 2026 AI Pricing Playbook, up from 27 percent the year before. Any AI platform built on agents, whether it powers customer service bots, marketing assistants, or internal reporting tools such as those covered in TheAIDaily's breakdown of what AI actually costs a team, runs into the same wall: a flat per-seat price stops covering the cost the moment someone uses the tool hard.
How much more could this cost you?
Here's a worked example. Say you currently pay $50 a month for an AI writing assistant, for a team of five people. That's $250 a month, total. Everyone uses it a few times a day for short tasks.
Now the vendor switches to usage-based billing. Two of your five team members turn out to be heavy users: they have the AI generate full reports, draft entire email sequences, and analyze documents end to end. Their usage runs five to ten times higher than the other three. Suddenly the team isn't paying $250 a month. It's paying $500 to $700, without a single new hire.
For context, McKinsey's State of AI 2025 survey found that 88 percent of organizations now use AI in some form, up ten points from 2024, and that 13 percent of employees already use generative AI for at least 30 percent of their daily work, three times what most leaders assume. That gap between assumed and actual usage is exactly where usage-based bills spike hardest. If you want a deeper international breakdown of how employees actually use these tools day to day, TheAIDaily's own AI workforce statistics track that shift month over month. The question isn't whether your vendors switch. It's when.
How to check if your AI tools are exposed
Five concrete steps you can take this week:
- Read your contract terms. Look for phrases like "fair use policy," "usage limits," "consumption-based," or "overage fees." If your vendor reserves the right to change the pricing model unilaterally, you're exposed.
- Ask your vendor for a usage report. How many tokens, requests, or credits does your team burn per month? Without that number, you can't size the risk.
- Identify your heavy users. Every team has one or two people who use an AI tool ten times harder than everyone else. They're the ones who will drive the bill up when the pricing model changes.
- Set a budget cap. More and more tools now let you set a hard monthly ceiling. Use it. A fixed limit prevents surprises.
- Calculate cost per outcome, not per seat. What does it actually cost to generate a report, review a contract, or produce a campaign? That's the number you need for a fair comparison.
When usage-based pricing actually works in your favor
It isn't all bad news. Companies with many licensed seats but low average use often come out ahead under usage-based pricing. If you have ten licenses but only three people touch the tool daily, you're currently paying for seven empty seats. Usage-based billing means you pay only for what you consume.
Legora now offers dashboards that break down usage by organization, by user, and by project. That's more visibility than most firms ever had under a flat subscription. The invoice becomes feedback: you can see exactly which tasks burn the most tokens and which don't, which helps answer the harder question of what to hand off to AI and what to keep doing by hand.
Worth noting: model choice still matters just as much as pricing structure. Per-token prices for the underlying models themselves vary enormously, from a few cents to more than $200 per million tokens depending on the model and provider. Which model sits behind your tool has as much influence on your bill as the pricing plan itself.
Make the invoice your steering wheel
The shift to usage-based pricing is not going away. AI agent costs scale with task complexity, not with headcount. Vendors that hold onto a flat price are effectively subsidizing their heaviest users with money from their lightest ones, and that model breaks down the moment everyone starts using the tool hard.
For any company running lean, the lesson is straightforward: treat your AI budget as a variable that moves with usage, not a fixed line item. Ask your vendor for usage transparency now, before that transparency gets forced on you by an invoice that's suddenly double what it was last month.
Teams that start measuring early are the ones least likely to be blindsided when the pricing changes. And knowing which questions to ask about AI usage before the bill lands puts you in a stronger position for the negotiation that's coming either way.
