AI Unit Economics: Cost to Serve, Margin, and Profit per Customer
AI unit economics measures the profit of your AI product one unit at a time, usually per customer or per request. You take the revenue a customer generates, subtract the variable cost to serve them (model spend, compute, and per-customer overhead), and the margin that remains tells you whether growth makes you money or loses it.
The three numbers that define AI unit economics
Unit economics for an AI product comes down to three figures per customer: the revenue they pay you, the variable cost to serve them, and the margin between the two. Revenue is what you already know. Cost to serve is the part most AI companies cannot see, because it is buried inside an aggregate provider bill.
Once you have both, margin is simple subtraction, and it is the number that decides whether adding customers is growth or a slow leak.
Customer margin = Revenue − Cost to serveWhy AI changes the math
In classic SaaS the cost to serve one more customer is close to zero, so margins are high and roughly uniform. AI breaks that. Model calls are a real variable cost that scales with how heavily each customer uses the product, so two customers on the same plan can have very different margins.
That variance is where the risk hides. A power user on a flat plan can consume more in model cost than they pay, and you would never see it in an average. Provider prices also move, so a margin that was healthy last quarter can erode without a single change on your side.
A customer pays $200 a month. Their model and compute cost to serve is $150. Margin is $50, or 25 percent. A second customer on the same $200 plan runs heavy agent workflows costing $240 to serve. That account loses you $40 every month, and the average hides it.
How to calculate and monitor it
Start by attributing model cost to each customer, so cost to serve is measured rather than guessed. Map that against the revenue each customer generates, then track the margin over time and flag accounts whose margin falls below a healthy threshold. Bear Lumen does this continuously, so unit economics is a live dashboard instead of a quarterly spreadsheet.