Skip to main content

Your blended margin numberis hiding the real story

Break down costs at the customer, segment, feature, and usage-pattern level. See which parts of your business subsidize others.

Blended margins obscure what matters

Aggregate cost data hides the customer-level economics that drive pricing decisions

The Blended Margin Problem

Three disconnected systems create a gap that hides cross-subsidies.

Provider invoices show total spend

OpenAI, Anthropic, and AWS bills show aggregate costs. They don't show cost per customer or per feature.

Revenue systems show total revenue

Stripe shows payments and subscriptions. It doesn't connect revenue to the costs incurred serving each customer.

The gap reveals cross-subsidies

Some customers generate margin. Others consume it. Without request-level cost attribution, you can't tell which is which.
How revenue connects
Connect Stripe via OAuth — subscriptions, invoices, and payments sync automatically. Revenue is attributed to customers and matched with cost-to-serve data to calculate margins.

How granular margins work

SDK captures request-level costs. Dashboard breaks down margins by any dimension you need.

Frequency
Continuous
Granularity
Per Customer
Dimensions
Multi-Dimension

Customer Margin Table

Revenue, cost-to-serve, and margin for each customer. Sort and filter by any dimension.

Customer
Revenue
Cost
Margin
Acme Corp$4,200$1,890+$2,310
StartupCo$200$12,000-$11,800
DataFlow Inc$500$413+$87
Customers are automatically classified as profitable, break-even, or unprofitable. StartupCo uses 60x the inference of Acme Corp on a plan that costs 95% less.

Margin Over Time

Track margin trends by customer, feature, or model. Granularity adapts automatically — daily for recent data, weekly and monthly for longer ranges.

October
58%
November
42%
Change
-16 pts
-16 pts

Cross-Subsidy Analysis

See margin contribution by feature, model, customer tier, or custom metadata. Identify which dimensions generate margin and which consume it.

By Model— not all models are profitable
Model
Provider
Margin
Claude HaikuAnthropic
+84%
GPT-4o MiniOpenAI
+61%
GPT-4oOpenAI
+22%
GPT-4 TurboOpenAI
-12%
Claude OpusAnthropic
-31%
Claude Opus and GPT-4 Turbo are loss-making at current pricing. Haiku generates 84% margin on the same workloads.

Break down margins by any dimension

The same cost data, sliced however you need it

By Customer

Revenue minus cost-to-serve for each customer. See who generates margin and who consumes it.

By Feature

Which product features drive cost? Which are margin-positive? Attribute costs at the endpoint level.

By Model

Compare margin across AI providers and model versions. GPT-4o vs Sonnet vs your fine-tuned model.

By Segment

Group customers by tier, region, or plan. See margin patterns across your customer base.

See your real margins

Request-level cost attribution. Margin by customer, feature, and model. Updated continuously.