Every SaaS company eventually has the same conversation with their customers: what does this cost per month? For most software products, that is an easy question. For AI products built on LLM APIs, it is genuinely hard to answer.
Your costs vary with usage. Usage varies with user behaviour. User behaviour is unpredictable. So you either offer variable pricing and constantly justify bills to customers, add usage limits and frustrate your best users, or quote a fixed price and quietly absorb the variance yourself.
Most teams end up doing a combination of all three. None of it is satisfying. There is a cleaner way to think about this.
Why variable costs make pricing hard
The obvious response to variable costs is variable pricing. Charge per query, or per token, or per feature use. Your revenue tracks your costs and you maintain your margins.
The problem is that your customers hate it. Usage-based pricing creates anxiety. It makes customers think twice before using your product. It creates awkward conversations every time an invoice is higher than expected. And it makes you nearly impossible to include in a customer's annual budget.
Enterprise buyers in particular need to know what a tool costs before they can approve it. A quote that says "it depends on how much you use it" is not a quote. It is a reason not to buy.
The safety margin approach
The most common way teams try to offer fixed pricing is to quote a number that includes a large enough safety margin to cover the worst-case usage scenario. If your average LLM cost per seat is $8 per month but it can spike to $22 in a heavy usage month, you quote $30 and hope the average covers you.
This works until it does not. The margin that felt comfortable at 100 users stops feeling comfortable at 10,000 users with a much wider distribution of usage patterns. And quoting $30 when the actual average cost is $8 makes you uncompetitive against any product that has figured out how to tighten that spread.
Usage tiers and soft limits
Another approach is to define usage tiers: a starter plan with 100 queries per month, a growth plan with 1,000, and so on. This caps your downside exposure per customer but introduces new problems.
Customers who approach their limit start rationing usage. That is the opposite of what you want from an engaged user. Customers who exceed their limit feel penalised for finding value in your product. And your support team spends a meaningful amount of time handling limit complaints.
The products that win long-term are the ones that let customers use them without anxiety. Usage limits create anxiety by design. The goal should be to remove the constraint, not manage it more elegantly.
Fixing the cost, not the pricing
The cleanest solution is to make your underlying costs fixed before you build your pricing model. If you know your LLM costs are $X per customer per month regardless of how they use the product, you can quote confidently, budget accurately, and compete on value rather than constantly managing the gap between your costs and your pricing.
There are a few ways to approach this. You can build a significant internal optimisation layer that handles caching, prompt compression, and model routing to reduce variance. This works, but it requires engineering resources that most teams would rather spend on their actual product.
Alternatively, you can work with an infrastructure provider that absorbs the variance for you and charges a fixed fee. That fixed fee becomes a known cost of goods sold. You build your pricing model on top of it. Your customers get a clean number. Your margins are protected.
What fixed LLM costs unlock
Once your LLM costs are fixed, a few things become straightforward that were previously complicated.
You can put a real number in your financial model. Investors and finance teams can see exactly what your cost structure looks like at scale. You can tell them: at 1,000 customers, our LLM cost is $X per month, full stop.
You can set your own pricing with confidence. You know your margin on every deal before you sign it. You are not underwriting unknown usage risk on every contract.
You can offer unlimited usage tiers. If your cost per customer is fixed, there is no downside to letting customers use the product as much as they want. That changes the competitive conversation entirely.
And you can close deals faster with enterprise buyers. The moment a buyer asks about costs and you can hand them a fixed number, the deal moves forward. Variable pricing stalls deals. Fixed pricing closes them.