AI Unit Economics: Why Cost-Per-Outcome Matters More Than Token Counts

Every AI team has a cloud bill. Few have a clear picture of what each workflow actually costs end-to-end.

The default approach, tallying OpenAI invoices and adding up AWS line items, gives you total spend. But it doesn't answer the question leadership is really asking: "Is this AI feature making us money or losing it?"

The Problem With Aggregate Spend

When your recommendation engine costs $25K/month in LLM calls, that number means nothing without context. What's the revenue per recommendation? How much compute and storage does the pipeline consume on top of the API spend? Without workflow-level attribution, you're flying blind.

What Cost-Per-Outcome Looks Like

Cost-per-outcome means mapping every resource a workflow touches (LLM calls, vector DB queries, data preprocessing, cloud compute) back to a single business result. A customer support ticket resolved. A document processed. A recommendation clicked.

This is what botanu does. We instrument your workflows with OpenTelemetry, pull billing data from your cloud and LLM providers, and correlate everything into one cost-per-outcome metric.

Why It Changes Decisions

Once you see cost-per-outcome, decisions get clearer:

• A chatbot costing $0.12 per resolution that saves $4.50 in support costs? Scale it.
• A summarization workflow at $2.30 per document when manual review costs $1.80? Optimize or sunset it.
• An internal tool burning $8K/month with 12 monthly users? That's a conversation worth having.

Getting Started

You don't need a 3-month project to start tracking cost-per-outcome. With botanu, most teams are instrumented and seeing their first workflow economics within a week.

The hardest part isn't the tooling. It's shifting from "how much do we spend on AI?" to "what does each AI workflow cost per outcome?" Once you make that shift, every optimization decision becomes data-driven.