AI consumption pricing is the change your next renewal quote is likely to carry, and few small firms have been warned it is coming. The flat subscription that let a fee earner use a tool as often as they liked is giving way to billing by how much work the tool does.
Legal IT Insider set the position out on 21 August 2026 in an interview with Rudy DeFelice, global head of AI strategy at the legal technology consultancy Harbor. His summary is blunt, that the token subsidy era has ended. The companies behind the models started as research bodies under no commercial pressure and grew into some of the largest businesses in the world, and they now run hard against the computing capacity they own. Prices set while they were buying market share do not survive that shift.
His answer is not a single tool
DeFelice argues for running several models behind one workflow, with routing that decides which model sees which task. An inexpensive model summarises a document or pulls the clauses out of a contract perfectly well, and the reasoning work goes to something dearer. He compares it to the way a firm already sends work to a partner, a paralegal or a legal secretary depending on who is needed for it.
Harbor puts the saving at 70 to 80 per cent, which is a consultancy describing its own results rather than an independent measurement. Take it as a direction and not as a number to budget against. That several models sit behind one product is something we wrote about in July.
Two of his reasons for spreading the work have nothing to do with cost. A firm tied to one model is exposed the moment a government makes a decision about that model. Clients set conditions of their own as well, ruling out particular models or requiring the work to run inside a private data centre.
What to ask before the quote lands
Put it to your supplier in writing, ahead of the renewal, whether your pricing is moving to a consumption basis and what the unit of measurement is. Ask what happens when the allowance runs out, because there are two answers and they carry different problems. Work that stops mid matter is a service failure you have to plan around. Work that carries on at a rate nobody has approved is a bill you will be explaining to a client.
Then look at what your people send the tool. Where most of it is summarising, extracting dates and reading correspondence, you are paying a premium rate for work at the bottom of the range, and the question worth asking is whether the product lets you choose at all. Ask as well what the firm does if the price doubles. A tool folded into how a team works is not one you drop in a week, and the moment to price that dependency is before you deepen it.
DeFelice sets out the argument, the risks and the reasoning in an interview with Legal IT Insider, which is open to read and asks nothing of you.
If you want your AI spend and your supplier terms read before the renewal rather than after it, that is work we do with firms: talk it through with us.