Track 09 · Systems & unit costLesson 05 of 11
Cost per resolved case
The model economics the CFO will ask about, in their units.
- The FDE path · Stage 6 of 6
- 35 min read
- 75 min hands-on
- Pro
What you'll be able to do
01- Explain each component of a model bill and which ones your workload pays.
- Derive cost per request, per task and per resolved case, and defend why only the last goes to the CFO.
- Diagnose agent cost growth from context growth, cache misses, retries and escalations.
- Decompose turn latency and predict what moves p50 versus p95.
- Design a tiering or cascade strategy and justify it with a simulation and sensitivity table built with finance and operations.
Before you start
- Tokens, context windows and why answers go wrong and Choosing a model you can defend.
- Reasoning models and what thinking costs, which owns reasoning effort as a lever; here we only price it.
- Context on a budget: what to keep, compress and cut.
- The agent loop, written by hand and Retries, fallbacks and failing gracefully.
- Handing off to a person, well: escalations are a cost line.
- Basic probability (expected value, percentiles).
The short version
A price sheet gives the cost of a token; the customer pays for outcomes. Between the two sit the number of model requests per task, the size of each request (context grows every loop step), the share served from a prompt cache, retries, and how often a human finishes the job. The unit that matters is the resolved case: AI spend on every case plus human cost on the ones the agent did not finish, divided by cases genuinely resolved. Latency works the same way: the user feels the whole turn, and p95 is set by the slowest slice of traffic.
- A model request costs about a cent, a task tens of cents, a resolved case dollars. Show the CFO the last.
- Agent loops re-send growing context, so input tokens grow roughly with the square of steps; caching is the main defence.
- The cheapest model per token is often the most expensive per resolved case, because its failures cost human minutes.
- Build the model with finance and operations, and lead with the sensitivity table.
Pro lesson
The rest of this lesson is in Pro.
The mental model, worked example, field note, failure table, exercise and self-check, with the voice tutor.
Pro opens this stage
Stage 6 · Ship, run and hand over- 8 lessons, with the voice tutor
- Lab · Check a trace export before you trust the dashboard, graded here
- Halden week 7 · The champion moves
- Halden week 8 · The portal goes down before cut-off
- Halden week 9 · Prices up, and a new rule
- Halden week 10 · A second site, and handover
- The lab kit, to run Halden weeks on your machine
Also in Pro
- Every stage and specialisation, in full
- Every lab and every role project
- Halden weeks 2–10, on your record
- Every drill, live in the playground
- All solutions, plus the server judge
- Lab Assistant and every Teach with AI