Unit Economics of AI Features
Calculate cost per request, per user, and per outcome; model gross margin under different usage patterns; and spot features that lose money at scale.
In traditional SaaS, serving one more customer costs almost nothing, so pricing is mostly about what customers will pay. AI features break that assumption. Every summary, draft, and chat message has a real, variable cost, and the customers who love your feature most are the ones who cost the most to serve. It's entirely possible to build a feature people adore that loses money on every heavy user, and to discover it only after it's popular.
Unit economics is the practice of working out what one unit of your product costs and earns. In this lesson you'll calculate cost per request, per user, and per outcome for Deskmate's AI features, model margin under realistic usage patterns, see why success can make margins worse, and find the features and customers that lose money before they become a crisis.
Three units, three questions
Each unit answers a different question, and you need all three:
| Unit | Question it answers | Where the number comes from |
|---|---|---|
| Cost per request | What does one call to this feature cost? | Your cost ledger: total cost ÷ calls, per feature |
| Cost per user (or per seat, per account) | What does one customer cost to serve each month? | Calls per user × cost per call, summed across features |
| Cost per outcome | What does one useful result cost? | Total feature cost ÷ number of successful outcomes |
The first is an engineering number, the second becomes a margin once you put it next to what the customer pays, and the third tells you what you're actually buying with your AI spend.
All three start from the cost ledger built in Understanding Your AI Bill, with every call tagged by feature and customer. If you don't have that yet, estimate from token counts and list prices, then replace the estimates with real data as soon as you have it.