Per-customer cost attribution links the underlying vendor costs of usage to the specific customer who caused them. It reveals real gross margin per customer, workload, or feature, not just aggregate revenue.
Per-customer cost attribution answers a question most billing systems cannot: how much does each customer actually cost to serve, and therefore what is the margin on each one? For AI products, where a single customer's usage can carry real and variable vendor cost, knowing revenue alone is not enough. Attribution links cost to the customer who caused it.
How Per-customer cost attribution works
It connects two data streams at the event level: the revenue from a customer’s usage, and the underlying vendor cost that usage incurred (model API charges, compute, GPU time). By tagging each usage event with both the customer and its cost, the system can sum cost and revenue per customer, per workload, or per feature, and compute gross margin for each.
The key is event-level granularity. Aggregate financials show total cost and total revenue, but they cannot tell you that one customer runs at negative margin while another subsidizes them. Attribution makes that visible by tracking cost at the same resolution as usage.
Per-customer cost attribution examples
An AI product sees that a flagship customer generates $5,000 of revenue but $4,200 of model cost, a thin margin hidden inside a healthy-looking total. Another customer on the same plan costs a fraction of that. Attribution surfaces both. A platform discovers one feature is unprofitable across all customers because its underlying compute cost exceeds what the plan charges.
Without attribution, these signals are invisible until aggregate margin erodes and the cause is unclear.
Per-customer cost attribution vs Aggregate reporting
| Per-customer cost attribution | Aggregate reporting | |
|---|---|---|
| Resolution | Per customer / workload / feature | Whole business |
| Reveals | Margin per customer | Total margin only |
| Catches | Unprofitable customers/features | Only the net effect |
| Needs | Event-level cost + revenue | Summary financials |
Benefits & when to use it
Per-customer cost attribution is what lets an AI business defend margin. It identifies unprofitable customers and features before they erode the whole, informs pricing (which plans actually clear their marginal cost), and turns “we are roughly profitable” into “we know exactly where we make and lose money.”
It becomes essential when cost per customer is variable and significant, the AI and infrastructure case. For flat-cost products it matters less, because all customers cost roughly the same to serve.
FAQ
What is per-customer cost attribution?
Linking the underlying vendor costs of usage to the specific customer who caused them, so you can see real gross margin per customer, workload, or feature instead of only aggregate revenue.
Why do AI products need per-customer cost attribution?
Because AI usage carries real, variable vendor cost, and one customer can run at negative margin while the total looks healthy. Attribution surfaces unprofitable customers and features that aggregate reporting hides.
How is cost attribution different from usage metering?
Metering measures how much a customer used. Cost attribution goes further: it links the vendor cost of that usage to the customer and compares it to revenue, producing margin per customer rather than just consumption.
How Credyt handles Per-customer cost attribution
Per-customer cost attribution is part of Credyt's observability. Credyt ingests vendor costs and links them to usage events at the customer, workload, and feature level, exposing real-time gross margin per customer. Because usage is already metered and attributed as it happens, margin is visible live rather than reconstructed at quarter end. See the platform. Explore Credyt →