Should our firm pay a subscription for AI or pay per use?

Neither model is right in the abstract. A subscription makes sense when you already know AI will be used steadily by most fee earners, because the per-hour cost falls as usage rises. Pay per use makes sense when you do not yet know that, because it lets you find out without committing the firm to twelve months of licence fees for seats that may sit idle.

Most firms of four to fifty fee earners are in the second position, whether or not they admit it. The honest question is not which model is cheaper in theory but which one fails more cheaply if the tool turns out not to suit the work you actually do.

What each model really asks of you

A subscription asks you to predict usage. You buy seats for a period, usually a year, and you pay for them whether they are used or not. The supplier gets predictable revenue and often gives you a lower headline rate in return. Your risk is that adoption is patchy: three enthusiasts use it daily, everyone else logs in twice and forgets, and you have paid for all of them.

Pay per use asks you to tolerate a variable line in the management accounts. You pay for what you consume. Costs rise in a heavy month and fall to nothing in a quiet one. Your risk is that a busy period produces a bill nobody budgeted for, which is why it matters whether the supplier lets you control spend rather than simply invoicing you afterwards.

There is a third arrangement worth naming: credit bought in advance. You top up when you choose, the balance runs down as it is used, and nothing renews automatically. That caps your exposure by design rather than by promise.

Questions to ask before you sign anything annual

• How many seats do we have to buy at minimum, and can we reduce the number mid-term?
• Does the price include everything, or are there tiers where the useful features sit above the one we are being quoted?
• What happens at renewal? Is it automatic, and what notice must we give to stop it?
• If we stop paying, do we lose access to past work and conversations, and in what form can we take them away?
• Is the quoted price per fee earner, per user, or per firm? Support staff and paralegals often need access too, and per-user pricing can double a quote quietly.
• Is training, onboarding or configuration charged separately?

The renewal question matters more than firms expect. An annual contract signed in a burst of enthusiasm has a way of rolling over twice before anybody checks whether the tool is still being opened.

Work out the cost against something real

A licence fee means nothing on its own. Compare it against the work it displaces, and be specific about which work.

Pick two or three tasks you can measure: reading into a new file, pulling the key dates out of a bundle, finding the relevant part of a statute or piece of official guidance. Estimate what those tasks cost you now in fee earner or paralegal time at the rate you actually recover, not your headline charge-out rate. Then run the same tasks with the tool for a month and see what falls out.

Two warnings on this exercise. First, time saved is only money saved if it goes somewhere useful; if it is absorbed into the day, the saving is real for the person but invisible in the accounts. Second, count the checking. Any output that a qualified person must read against the source before it is relied on carries a review cost, and that cost belongs in the comparison. A tool that saves an hour of drafting but adds forty minutes of verification has saved twenty minutes, not an hour.

What tends to go wrong with each

With subscriptions, the common failure is quiet non-use. Nobody cancels because nobody owns the decision, and the cost becomes part of the furniture. If you go this route, put a named person in charge of reviewing usage before renewal, and ask the supplier for usage figures in writing.

With pay per use, the common failure is nervousness. Fee earners avoid the tool because they are unsure what a query costs, so the firm never learns whether it works. The fix is to tell people plainly what the budget is and that they are expected to spend it during a trial period.

With both, the failure that costs most is buying before you know what problem you are solving. Firms that decide first which tasks they want help with, then test against those tasks, spend less and abandon fewer tools.

Where Alesis fits

Alesis is funded by credit rather than a subscription. The first firm a person sets up starts with £25 of free credit, with no card needed, and each colleague who joins with a verified account adds £15, for up to 20 colleagues. The firm tops up only when it chooses: nothing recurs and there are no feature tiers, so there is no annual commitment to review and no seat you are paying for while it sits unused.