Back to course overview
Module 1AI products are different 15 min

Workshop: The opportunity

Pick and frame your course-long AI product opportunity: user problem first, the three questions answered in writing, and the error-cost map that shapes everything after.

Each module's workshop builds one artifact toward the capstone AI PRD. This one picks the product. Default case: Dana's next bet — extending Harbor Helper from answering questions to proactively resolving delivery issues (it notices the delayed-package pattern and offers resolution before the customer asks). Better: your own — an AI feature your actual product or company could ship. The workshop's standard applies either way:

  1. 1Start from a user problem with evidence: write the problem statement with receipts — support-ticket themes, user quotes, funnel data ('18% of Harbor Lane tickets are where-is-my-order; median resolution 4 hours; CSAT on these: lowest of any category'). An opportunity that starts 'now that models can…' gets rewritten or discarded; the workshop's first discipline is problem-first framing under AI temptation.
  2. 2Answer the three door questions in writing: wrong-answer cost (mapped to user-felt severity — a proactive message about the wrong order is worse than silence; write why), how-we'll-know-it's-good (sketch the measurable definition — Module 3 will harden it), and why-AI-beats-boring (what's the lookup-table version? the human version? why do they lose? — if they don't lose, congratulations on the cheaper roadmap item).
  3. 3Draw the error-cost map: list the feature's 5-8 plausible failure types, place each on a 2×2 of likelihood × user-felt cost, and mark which quadrant each mitigation strategy serves (forgiveness design for frequent-cheap; oversight gates for rare-expensive; scope cuts for frequent-expensive — that quadrant usually means the feature is too ambitious as scoped).
  4. 4Write the one-page opportunity brief: problem + evidence, the three answers, the error map, and the deliberately-unsolved questions (data availability? feasibility? — Module 2's job). Get one adversarial read (a colleague or your AI assistant as a skeptical head of product: 'what's the strongest case against building this?') and append the strongest objection unrebutted — living with an open objection is PM cardio.
Problem set 1

Six feature pitches to triage with the three questions — two are solid, two are tech demos in feature costumes ('AI-powered dashboard insights' with no decision attached), one has an unmeasurable quality definition, and one fails the boring-alternative test to a well-designed form. Write the kill memos kindly; you'll be writing real ones within a month of taking this job seriously.