WORKSHOP: Cost Per Task, Not Cost Per Token: Unit Economics of Frontier LLM Workloads
Cost Per Task, Not Cost Per Token: Unit Economics of Frontier LLM Workloads
11 September | 14:00-15:00
Tech Open, Stockholm
Led by DoiT
Token prices are plummeting across every major provider. Yet total AI infrastructure bills keep climbing. Measuring GenAI efficiency purely by cost per token creates a false sense of efficiency while hiding compounding multi-step calls, agentic loops, retries, and retrieval overhead.
In production, cost per task is the only metric that directly ties your AI spend to tangible business ROI.
Join the DoiT in a session where we unpack the realities of production LLM costs. We will share actionable frameworks for building reliable unit economics models, eliminating invisible spend leaks, and making sustainable architectural decisions.
What we'll cover
The Token Illusion: Why falling model API prices aren’t lowering overall enterprise AI spend.
Cost-per-Token vs. Cost-per-Task: Breaking down how compound AI systems, tool use, and multi-turn workflows warp traditional cost tracking.
Hands-on Unit Economics Modeling: A step-by-step framework to attribute compute, caching, retrieval, and guardrail overhead directly back to business tasks.
Engineering & FinOps Alignment: Building shared visibility, guardrails, and feedback loops between technical decisions and gross margin targets.
Optimization Playbooks in Practice: When and how to implement prompt caching, semantic routing, small-model distillation, and task-level batching.
Benchmarking Real-World Workloads: Case studies and architectural lessons from scaling LLM features to paying users.
You'll walk away with
A Cost Framework: Transition your tracking from abstract tokens to measurable task ROI.
Production Forecasting Templates: Practical architectural blueprints to model the true cost of compound AI workflows.
Targeted Optimization Playbooks: Strategies to immediately eliminate spend leaks and defend feature margins.
Actionable Peer Benchmarks: Tradeoff analysis between model size, task complexity, and infrastructure spend from production teams.
Who should attend?
Builders and decision-makers responsible for building, monetizing, and governing production AI workloads, including:
CTOs, Heads of Engineering & Technical Leads
AI Product Owners & Product Managers
FinOps Practitioners & Cloud Cost Owners
Founders & Technical Executives
Who are the speakers?
Philipp Heinrich, Engineering Manager at DoiT Labs, will lead the session, joined by Max Sautter (Manager of Sales) and Helmer Baas (Senior Account Executive).
How the session works
We will start promptly at 14:00. Please arrive a few minutes early to get settled.
Places are limited.
Location: Workshop Tent, located at the entrance to the Tech Open venue, in the tent closest to the fountain.
*This workshop is part of Tech Open during Nordic Tech Week in Stockholm. For questions, contact events@nordictechweek.org.