By Rob Simcock

AI pricing in legal tech is starting to look like a business model I know well: metered billing that captures revenue whether or not the customer is getting value.

Thirty years ago, I sold photocopiers and printers. The pricing model was simple: charge per copy, with a contracted minimum volume built in. It didn’t matter whether the customer was getting value out of the machine that month, the revenue was captured either way.

I assumed legal tech had moved past that kind of thinking. Software pricing has mostly settled into predictable shapes: per-seat licensing, tiered subscriptions, sometimes a flat platform fee. You know roughly what you’re paying before the year starts, and so does your finance team.

At iManage ConnectLive recently, a conversation with another delegate made me question that assumption. I asked what they thought happens next with AI in legal. Their answer: “We run out of money.”

They weren’t talking about firms overspending through poor planning. They meant the pricing model itself. Much of the conversation at the event centred on per-token and per-search charging for AI tools, and the legal sector is squarely in that market. iManage themselves weren’t suggesting this as their own direction, but the pattern among other vendors there was hard to miss.

We don’t charge this way at Kutana+Brochet, so it hadn’t fully crossed my mind that firms were already being billed for every search, every query, every interaction with an AI tool. Once it was described plainly, the logic was identical to what I was selling in the photocopier trade three decades ago…new technology, same model.

Why usage-based pricing is a harder problem for law firms specifically

Per-seat software pricing has one real advantage: predictability. You know your headcount, so you know your cost, and budgeting is a matter of multiplication.

Usage-based AI pricing removes that. A firm’s AI spend now depends on how much lawyers actually use the tool, how complex their queries are, and how quickly adoption grows across the firm. None of those are things a finance director can forecast with much confidence in year one.

Here’s the part that should give firms pause: success is what breaks the model. If an AI research or drafting tool works well and lawyers start relying on it for genuinely useful, everyday tasks, usage goes up. Under a per-token model, that’s exactly when costs start climbing fastest, often just as the firm is trying to demonstrate return on the investment in the first place.

That’s the same trap I used to sell without fully realising it. A customer who copied more paid more, regardless of whether the extra copying reflected genuine business growth or just a machine that had become indispensable. Firms adopting AI tools priced this way are walking into a similar dynamic, just with higher stakes and less visibility into what’s driving the monthly bill.

What firms should be asking before they sign

None of this means usage-based pricing is automatically the wrong choice. For some tools, at some volumes, it may genuinely work out cheaper than a flat licence. The problem is signing up without understanding which side of that trade a firm is likely to land on.

Before committing to an AI tool priced by token or by search, it’s worth asking the vendor a few direct questions. What does typical usage look like for a firm of our size, and what would that cost under this model? Is there a cap, and what happens once we hit it: does the tool stop working, throttle, or does the price simply keep climbing? If adoption across the firm succeeds beyond expectations, what does the bill look like in twelve months, not just during the pilot?

That last question matters more than it might seem. Pilots are usually run by a small, engaged group of early adopters, so the token counts look modest. The real cost only becomes visible once a tool has been rolled out across a full practice group, and by then the firm has already built workflows and expectations around it.

Where I’d push back

Does usage-based AI pricing ever make sense for law firms? Sometimes, but only if the vendor can show real numbers up front, not just a range of possible outcomes.

I don’t think every AI vendor pricing by usage has bad intentions. There are genuine reasons a token-based model can make sense, particularly for tools where usage varies enormously between firms and a flat fee would overcharge light users to subsidise heavy ones.

But the burden should sit with the vendor to make that pricing transparent and predictable, not with the firm to work it out after the invoices start arriving. Ask for real numbers from comparable firms, not hypothetical ranges. Ask what proportion of similar customers have gone over budget in year one. If a vendor can’t or won’t answer, that itself is useful information.

I’ve worked in a sales role built on exactly this kind of pricing before. I know how well it works for the supplier when the customer doesn’t fully understand what they’ve signed up for. Firms evaluating AI tools now have an advantage that photocopier customers didn’t really have thirty years ago: the model is well understood this time round, and firms can ask for transparency from the outset rather than discovering the shape of it in year two.

I wonder whether other firms are already seeing this play out in their AI vendor conversations, or whether the push-back is still to come?

Here’s where you can get in touch – Contact us