How to Track Anthropic Claude Costs by Customer (Same Problem, Different Dashboard)

March 15, 2026

We added Claude for document analysis in Q2. By Q3, our Anthropic bill had overtaken OpenAI. We had no idea which customers were driving it. The Anthropic Console showed total spend and usage by model. That was it. Same problem as OpenAI — different dashboard. If you're trying to get Claude cost attribution by customer, you've hit the same wall we did.

The Anthropic API lets you pass metadata — user ID, custom attributes. Sounds great. We tried it. Six months later, 35% of our Claude calls had no customer ID. Legacy features we hadn't touched. A third-party RAG wrapper that stripped metadata. Partial visibility turned out to be worse than none — we were making decisions on half the picture. A Claude customer spending breakdown that's missing a third of your calls is worse than no breakdown at all.

Why metadata alone isn't enough for Anthropic cost per user

Every engineer, on every feature, has to pass the right ID every time. New features get tagged. Old ones get forgotten. Third-party tools don't support it. The gaps compound. We thought we had Claude cost attribution figured out. We didn't. Our "cost per customer" view was missing the heaviest users — the ones who used features we hadn't updated. The Anthropic API cost per customer only works if tagging is consistent. Ours wasn't.

Don't build a separate pipeline for Claude

We almost did. Log every Claude call, store it, join to customers. Build our own pipeline. Then we realised: we'd already spent three months on the OpenAI pipeline. Doing it again for Anthropic meant another 2–3 months. And we'd still have two separate views — GPT costs here, Claude costs there. Never the full picture. No single view of which customers cost the most across both providers.

What we needed was one view that pulled from both. Connect our Anthropic and OpenAI accounts to something that maps usage to customer IDs and surfaces the combined breakdown. No pipelines. No data warehouse. Just the numbers we actually needed for pricing and product decisions. Claude cost attribution and OpenAI cost attribution in one place.

The view that changed how we priced

Once we had cost per customer across both providers, we could see who was heavy on Claude vs GPT. One enterprise account was using Claude for document analysis at $400/month — we were charging them $199. We'd never have spotted that from the provider dashboards. Another customer was 80% Claude, 20% GPT. We raised their price and they stayed. A third was burning through Claude on a feature we thought was low-volume. We gated it behind a higher tier.

The Claude customer spending breakdown we finally had — combined with OpenAI — gave us the full picture. We stopped subsidising heavy users. We priced in proportion to cost. That's what proper Anthropic cost per customer visibility enables.

PerUnit connects to Anthropic, OpenAI, and Google — and Stripe for revenue. Cost attribution by customer, feature, and tier, without building any infrastructure. Our free AI cost calculator helps you estimate Claude spend before you dig into attribution.