What do your different customers cost?

(We've written before about where a single request's cost goes. This time the question is different.)

For some businesses, it's relatively easy to determine customer specific costs. Imagine a chocolate company. They can calculate the exact cost of producing each bar. They can allocate things like marketing and back office. They know the exact promotional deals offered to each customer. They may have to determine transportation and storage based on distance from the production facility, but that's doable. There, you know how profitable each customer is, or is not.

For SaaS companies, it's another story entirely. There can be as many as 20 (!!!!!!) different costs. Just scan the table below.

Seventeen of these twenty can be traced to a specific customer. Some of them the moment a request happens - which model answered it, whether it hit a cache, what it cost to send the answer back. Others sit there whether the customer is active or not - how much they store, how it's stored, how many background jobs quietly run on their behalf while nobody is watching. A couple only show up after the fact - a support ticket, a compliance review, and take a little more work to trace back to the account that caused them, but they still belong to somebody.

Only three of the twenty don't belong to any one customer. Shared development environment, general platform overhead, capacity the company committed to and hasn't used yet. Those three are shared, and no amount of clever accounting should pretend otherwise.

Here's the uncomfortable part. The chocolate company has always known its version of this list, and it's a short one. A SaaS company's list is five times longer, harder to see, and changes shape depending on choices nobody in finance ever gets asked about. Some of those are which AI model answered a request, whether a cache happened to be warm, how much of last year's uploads are sitting untouched in expensive storage? Most SaaS companies can tell you the total bill. Very few can tell you, customer by customer, which seventeen of these twenty numbers belong to whom.

That gap used to be a technology problem. It mostly isn't anymore. The data behind most of these seventeen items already exists somewhere in your systems, the moment the event happens. What's usually missing is connecting it, routing that data to a customer's name and adding it up.

Once you do that work, the table stops being a taxonomy and starts being a tool. Take item seven - storage tier. On its own it's just a line on a bill. Paired with how often that data gets read, it becomes a specific, defensible sentence: this customer is costing us an extra $200 a month because their data sits in expensive storage and nobody has opened it in ninety days. That's not a reporting exercise. That's a decision waiting to be made.

The chocolate company would never let that sentence go unsaid. There's no reason a SaaS company should either.