Prepaid credits vs subscriptions: when to sell a balance instead of a plan

Somewhere between "charge a flat subscription" and "meter every unit" sits a third model that AI products have made mainstream again: sell a prepaid balance and let usage draw it down. Credits are not new. Phone cards worked this way. But they solve a specific set of problems, and they create a few of their own.

What credits are actually for

Credits remove bill shock. A customer who bought 10,000 credits cannot wake up to a surprise invoice. They spent what they chose to spend, in advance. For products where usage is spiky or hard to predict, this is the single best reason to offer credits.

Credits move cash forward. You get paid before the usage happens, not thirty days after the invoice. For a young company that difference is not cosmetic.

Credits make small purchases possible. A customer who is not ready for a $99 monthly commitment might happily put down $20 to try the product properly. Credits are a low friction door.

Credits abstract the unit. If your product does five different things with five different costs, pricing each in dollars gets noisy. Pricing them in credits (a small request burns 2, a heavy one burns 40) gives you one currency to explain.

What credits are bad at

Predictable recurring revenue. A wallet that gets topped up when it runs out is not a subscription. Revenue arrives in lumps, and a customer with a full wallet has no renewal moment where they re decide anything.

Signaling commitment. Enterprise buyers often want a contract with a known annual number, not a balance. Procurement does not love wallets.

Simplicity, if you are careless. A credit system with confusing conversion rates or hidden burn rates feels like a casino. The trust you gained by removing bill shock disappears fast.

The hybrid most products land on

The choice is rarely either or. The pattern that keeps showing up: a subscription for the predictable core (access, seats, an included usage allowance) plus credits for burst usage beyond it. The plan gives you recurring revenue and gives the buyer a known monthly number. The wallet absorbs the spikes without anyone renegotiating a contract.

The mechanics that matter in production

If you offer credits, four details decide whether they work:

1. A clear conversion rate. State it plainly and keep it stable: 2,500 credits = $25.00. Customers should be able to do the dollar math in their head.
2. Auto top up. A production workload that stops because a wallet hit zero is an outage you caused. Let customers opt into automatic top up at a threshold they choose.
3. A low balance threshold. Before the balance runs out, the customer should hear about it, at a threshold they set.
4. Visible burn. Customers should be able to see what drew the balance down, request by request, the same way they would read a phone bill.

yRecurring wallets ship all four: a stated conversion rate shown next to dollars, auto top up, a configurable low balance threshold, and a recorded movement history.

One more policy decision: whether credits expire. Expiry improves revenue recognition and nudges usage, but it is also the thing customers resent most about credit systems. If you expire credits, say so loudly at purchase, not in a footnote.

A simple way to decide

Ask two questions. Is usage predictable month to month? Is the buyer a company with procurement, or an individual with a card? Predictable plus procurement points at subscriptions. Spiky plus self serve points at credits. Both, which is the common case, points at the hybrid: plan plus wallet.

yRecurring runs all three shapes on one subscription: flat plans, included allowances with capped overage, and prepaid credit wallets with auto top up. The billing for AI page shows the credit mechanics in more depth, the token pricing guide covers how credits combine with per token rates, and the trial will let you set up a wallet and watch it burn down.