Stripe shows $48,000 in monthly recurring revenue across 120 customers. Datadog shows 2.3 billion tokens processed with 99.7% uptime. CloudZero shows $14,200 in monthly AI infrastructure spend.
The CFO asks: which customers are profitable?
Three dashboards open. None of them can answer.
Last Updated: April 2026
The Escalating Stack
AI companies build their tool stack incrementally. Each addition feels like it closes the visibility gap.
Day one: Stripe. Revenue tracked by customer. The billing side looks handled.
Month three: Langfuse. Traces logged. Token counts visible. Engineering can see what the system does.
Month six: CloudZero. Infrastructure spend tracked by service, by team, by environment. Finance can see where cloud dollars go.
Month nine: Portkey. Per-call provider costs logged. The team can see what each OpenAI or Anthropic request costs.
Five tools. Five dashboards. Each answers a real question: what did we charge, how much did they use, what did the system do, what did infrastructure cost, what did each API call cost.
The CFO still cannot answer: is Customer A's $2,400/month plan covering the cost to serve them? The answer lives across all five tools. No single tool produces it. The same visibility gap that cost Microsoft $20/user/month on Copilot for two years.
The Quiet Part
The problem is not missing data. The problem is that five precise dashboards create the feeling that economics are understood.
Revenue is exact. Token counts are exact. Infrastructure spend is exact. Provider costs are exact. Each number is accurate. Together they feel comprehensive. They are not.
Precise revenue plus precise infrastructure cost does not equal precise customer margin. The connection between what a customer paid and what that customer cost across all providers, all infrastructure, all features does not exist in any individual tool. The aggregate margin looks healthy at 52%. The distribution ranges from 63% to negative 55%.
The join is missing. And the confidence that it is not missing is the most expensive part.
What the CFO Actually Needs
The CFO does not need traces. The CFO does not need token counts. The CFO does not need infrastructure spend broken down by Kubernetes namespace.
The CFO needs to know: are we profitable by customer, and what should we charge next quarter?
That requires a sixth layer. Not a replacement for the existing stack, but a layer that sits on top. Pricing intelligence connects revenue data to provider costs to infrastructure spend to usage events. It produces what no individual tool can: margin by customer, by feature, by pricing tier.
Then it goes further. Given margin data, it models pricing changes before you make them. The CFO provides business context: customer segments, competitive positioning, growth targets. The tool provides margin data and pricing scenarios. The result is pricing built on economics, not intuition.
This is the difference between cost attribution and pricing intelligence. Cost attribution is a report: which customer is expensive. Pricing intelligence is a decision tool: what should we charge, and will it sustain the business.
The Six Layers
| Layer | Category | Examples | Question Answered |
|---|---|---|---|
| 1 | Billing and payments | Stripe, Chargebee, Zuora | What did we charge? |
| 2 | Usage metering | Orb, Metronome, Amberflo | How much did they use? |
| 3 | Observability | Datadog, Langfuse, Helicone | What did our system do? |
| 4 | Infrastructure cost | CloudZero, Kubecost, Vantage | What did our infra cost? |
| 5 | AI gateway / provider cost | Portkey, LiteLLM | What did each call cost? |
| 6 | Pricing intelligence | Are we profitable, and what should we charge? |
Most AI companies have layers one through three. Growing teams add four and five. Almost none have layer six. The first five layers produce data. The sixth produces decisions.
The Equation
The revenue side is precise. The cost side is instrumented. The equation that connects them is still a spreadsheet at most AI companies. Someone exports revenue. Someone cross-references the provider dashboard. Someone estimates infrastructure allocation. The result takes days to build and is stale by the time it reaches the board.
The teams that close this gap build pricing on data: which customers are profitable, which features drive costs, which tiers need adjustment, and what happens to margins if model costs drop 90% next quarter. The teams that operate without it build pricing on intuition and discover the gap the same way Microsoft did.
Bear Lumen is the pricing intelligence layer that connects your existing stack. See how unit economics work for AI products.