How to Price AI Features: Credits, Add-on or a Price Increase?

There are four ways to price an AI feature: include it in your existing price (with or without a price increase), sell it as a paid add-on, meter it through a credit allowance, or charge per outcome achieved. Which one fits comes down to two questions: how high are your variable costs per use, and is the feature a selling point for everyone or extra value for a few?

AI features differ from classic SaaS features in one fundamental way: they have real variable costs. Every call costs compute or API fees. A pricing model that ignores those costs can eat up the margin on an entire account once usage gets heavy.

Do the maths before you price: the margin check

Before you choose a path, you need two numbers:

1. Cost per unit of usage: What does an average AI operation cost you (inference, API fees, infrastructure), calculated realistically, including failed attempts and retries?
2. Usage per customer: How often does a typical customer trigger that operation per month, and what does the heaviest tenth of your customers look like?

Multiply the two and compare the result with your monthly price. If the AI costs of a heavy user take up a significant share of their subscription price, "just include it" is not an option, it is a margin risk. Step 5 of the 9-step process walks you through the calculation, and your break-even price sets the floor.

Path 1: Included, possibly with a price increase

The AI feature becomes part of the existing plan. It is the right choice when usage costs are low and the feature strengthens the core of your value proposition: it makes the product better, so it justifies the price rather than needing one of its own. A moderate price increase for everyone is often more honest than an artificial add-on, but it has to be communicated properly. Be careful when usage varies widely: included means your heaviest users run up costs on your tab.

Path 2: Paid add-on

Customers buy the feature separately, per account or per user. It is the right choice when only some of your customers need the feature and that group has a clearly higher willingness to pay. The add-on keeps the base price competitive and captures that willingness to pay exactly where it sits. The weakness: every add-on is a second buying decision with its own conversion hurdle, and too many add-ons fragment your offer.

Path 3: Credits

The customer receives or buys an allowance and uses it up depending on the action. Credits have become the standard when usage costs are significant and fluctuate a lot: they cap your risk, make consumption visible and let actions with different costs run through a single counter. Two rules decide whether customers accept them: the customer has to know what an action costs before taking it, and the ratio of credit price to your costs has to support your margin, even for the heaviest tenth of users. Confusing credit logic is the most common reason these models fail.

Path 4: Outcome pricing

You charge per result achieved, for example per support case resolved automatically. This is the purest form of value-based pricing and a strong selling point: the customer only pays when the AI delivers. But it requires an outcome you can measure clearly and beyond dispute, and the capacity to carry the risk of failures. For agent products with clearly countable success it is a serious option; for fuzzy outcomes it is a dispute waiting to happen.

The decision logic in brief

Where this fits

Choosing a model for AI features is a special case of step 8 in the 9-step process; the SaaS pricing models guide gives an overview of all the basic models. For a deeper look at agent pricing, including the margin calculation, see AI agent pricing. And for why you should not let a generic AI name the price itself, see asking ChatGPT for your price. As with every pricing decision: the path you choose has to work within your pricing corridor, otherwise the problem is not the model but the numbers behind it. In PricingOS, step 8 gives you AI-assisted model suggestions that take your cost structure from step 5 into account.