Scoping and Delivering AI Projects
Land with a client, find the problem worth solving, scope what can ship, prove it with a pilot and hand it over so it keeps working.
3 modules · 9 lessons~20 hoursUpdated 2026-11-13
- Lessons9
- Quizzes9
- Sources59
- Of study~20h
About this course
About this course
The consulting half of the forward-deployed job. You can already build, connect, evaluate and run an agent; this course covers the conversations and documents that decide whether one ships. Discovery that finds where time and errors actually go, choosing a problem the model can solve and the business will pay for, success criteria the client signs before you build, data access and readiness, a scoping document with phases and exclusions, a shadow-then-assisted pilot that ends in a decision, trust and adoption, handover, and the post-pilot memo. Every lesson is hands-on: Claude role-plays the sceptical operations lead, the IT manager and the sponsor, and the exercises produce the real artefacts of an engagement.
About the running example
The examples follow Harrowfield Facilities Services, a fictional mid-size company that runs cleaning, maintenance and security contracts for office buildings. Its 14-person operations team lives in a shared inbox of supplier emails, subcontractor invoices and site incident reports, answered from a patchy knowledge base. You are a forward-deployed engineer on a 12-week engagement, and the project becomes an agent that triages the inbox and drafts replies, with a human approving anything that commits money. The course ships synthetic interview transcripts, an inbox sample, a pilot log and inventories; every company, person and message is invented.
What you'll learn.
- 1Run discovery at a client: interview and shadow a team, map its workflow, documents and systems, and write a one-page brief with evidence counts
- 2Score candidate problems on value and feasibility, place each on the capability-to-risk ladder, and say no to the ones that fail
- 3Turn a business outcome into operationalised success criteria, a measured baseline and an acceptance eval the client signs before you build
- 4Write a scoping document with phases, exclusions, assumptions, a pricing choice and a time-boxed pilot with go/no-go criteria
- 5Run a shadow-then-assisted pilot, read its results against the baseline, bring sceptics and users with you, and hand over a runbook and a monitoring digest the client can operate
- 6Write the post-pilot decision memo and handle expectation-setting, scope creep and honest failure reports
3 modules, 9 lessons.
11h 45m of lessons
Module 1 · 3 lessons · 4h 10m
Discovery and the right problem
Land with the client, learn how the work actually happens, choose a problem worth solving that a model can solve, and define success as a measured baseline the client signs.
Module 2 · 2 lessons · 2h 45m
Scoping the engagement
Turn the chosen problem into something you can commit to: data access and readiness, the approval boundary, phases, assumptions, exclusions, pricing and a time-boxed pilot.
Module 3 · 4 lessons · 4h 50m
Pilot, people and handover
Run the pilot so it produces a decision, bring sceptics and users with you, hand the system over so it keeps working, and have the conversations that decide whether it ships.
The Harrowfield engagement pack
Run the whole engagement on paper, from six discovery transcripts to a post-pilot decision memo, producing every artefact a client would sign, read or operate, and defend it to a steering committee that Claude plays.