Curriculum

Prove it paid off

Outcomes and metrics for AI is 62 percent and, tellingly, outranks evals (53 percent), yet the report says course metric content is dominated by eval mechanics: this is a deepen theme and a named gap. It asks to separate product and busines

1. 09.01

   Business outcomes

   You separate the business outcomes leadership funds from model eval scores, and connect the two so a passing eval maps to a result you can defend.
2. 09.02

   Build the metric tree

   You build a named metric tree and define activation as a measured threshold for a product whose value is time saved, without letting engagement proxies stand in for usefulness.
3. 09.03

   Pull the number yourself

   You write the query yourself: fix the cohort window and the denominator, build the funnel, and agree definitions with whoever owns the pipeline.
4. 09.04

   Diagnose a drop when the treatment is nondeter

   You decompose a metric drop by segment, surface and cohort before hypothesizing, then rule out the three confounds unique to a nondeterministic system nobody recoded.
5. 09.05

   Cost per successful task

   You price a feature by cost per successful task, commit a per-surface latency target, and defend gross margin by pulling the levers in order, cheapest first, before you touch the model.
6. 09.06

   Price it

   You choose seat, usage, hybrid or outcome pricing against a value metric and a cost metric, with gross margin as the constraint and freemium math that runs opposite to SaaS.
7. 09.07

   The business case

   You assemble a business case that states value net of what would have happened anyway and the cost to run it forever, then hold the exact ship-or-no-ship number you pre-registered in the PRD, before you saw the result, and defend the call to leadership.
8. 09.08

   INTEGRATION CASE

   Defend one impact claim by showing the metric tree, the denominator, the owners, and how strong your evidence really is.