You Can't Price Per Outcome If You Don't Know Your Cost Per Outcome

Intercom charges $0.99 per resolution. Zendesk is moving to per-resolution pricing. Sierra charges per resolved conversation. Salesforce Agentforce is $2 per conversation.

The whole industry is converging on outcome-based pricing. Makes sense. The customer pays when the AI actually solves their problem, not when it exists on their plan.

But almost none of these teams actually know what an outcome costs them. They know their OpenAI bill. They know their AWS bill. They might know their Pinecone bill. "What does it cost us to resolve one customer inquiry through our Voice Agent" gets a long pause and then something like "we think around two bucks?"

A Voice Agent vs. a Chatbot vs. Escalations

We've been building tooling around this problem and the numbers are interesting.

A Voice Agent resolution in a typical customer service stack costs about $1.98. That touches Deepgram for speech, OpenAI for inference, Pinecone for retrieval, plus cloud infra underneath.

A Chatbot resolution on the same product costs $0.58. Lighter stack, fewer vendors, way better unit economics. An Escalation flow lands at $0.83 but generates less productivity value.

Same product. Three features. Three completely different cost profiles. Setting one price for "a resolution" means either leaving money on the table or losing it.

Customer-Level Economics

It gets more interesting at the customer level. An enterprise account doing 4,200 resolutions/month on Voice Agent + Chatbot runs $3,100 in AI spend. A mid-market account doing 2,600 resolutions mostly on Chatbot runs $1,400. Fundamentally different economics. Pricing them the same way doesn't work.

Beyond Per-Call Observability

The per-call observability tools (Helicone, Langfuse, Portkey) are good at what they do. But they show cost per API call. A single resolution might involve an intent router, a knowledge agent, an action agent, a response agent, multiple RAG calls, and a bunch of infra. The API call is one piece.

Outcome-based cost tracking means stitching all of that together: model costs across every agent in the pipeline, RAG pipeline costs (vector DB, embeddings, reranking, caching), infrastructure (compute, storage, networking), and data pipelines (ETL, transforms). Then mapping it to one resolution.

First-Mover Advantage

The companies that figure this out first will set the pricing standard for their category. The ones that don't will either underprice and bleed margin, or overprice and lose deals to someone who did the math.