PrerequisiteAn approved IL 101 Industry AI Project Charter is required
0 / 2 completed
Learning objectives
State the core question IL 201 answers: is this AI solution valuable and feasible enough to justify further investment?
Walk through the Sprint 0-5 rhythm, and understand how Proof of Value differs from Proof of Concept.
Write a falsifiable solution hypothesis, not an untestable hope.
Proof of Value, not just a demo
60 min
Answering "is it worth the next investment?" with evidence
IL 201's core question is: is this AI solution valuable and feasible enough to justify further investment? Success is not "we built an AI demo": it is "we produced enough evidence to make a better investment/adoption decision." A well-supported No-Go is as successful a learning outcome as a Scale decision.
Minimum conditions before a project is admitted to IL 201
Item
What it covers
Executive sponsor / business-process owner
A confirmed industry sponsor and process owner
Approved problem statement / baseline / target outcome
An approved problem statement, baseline, and target outcome
Defined evaluation metric
A clearly defined evaluation metric
Feasible evidence/data plan
A feasible plan for obtaining evidence/data
Acceptable privacy/security scope
An acceptable privacy and security scope
1Sprint 0: Kickoff Confirm the charter, roles, and way of working.
6Sprint 5: Decision The sponsor and process owner make an Evidence Review decision: Proceed to Readiness, Iterate, Hold, or No-Go (Module 6). A positive decision moves the project into the IL 202 Readiness Review, not directly into a pilot.
Info
Human, AI, Human: the sponsor and cross-domain advisors jointly define the problem and evaluation standard; AI helps build the prototype and analyze results; the final Go/No-Go decision, including the sponsor review, stays with people.
Key takeaway: Proof of Value answers "is it worth investing further?" with checkable evidence, not just a demo you can show.
60 min
Write a falsifiable hypothesis
A solution hypothesis is not an untestable hope like "AI should help." It is a specific statement that could turn out to be wrong: it names the scenario, what the AI does, the expected outcome, and how you would measure it.
Hypothesis v1 (not falsifiable)
Hypothesis: introducing AI will improve the quality of customer-service replies.
Hypothesis v2 (falsifiable)
Scenario
Demo Service Company A's customer-service knowledge Q&A cases (SYN-CS-001), focused on the normal and missing-information categories.
What the AI does
The AI drafts a reply from the knowledge base with cited sources; the agent verifies it before sending.
Expected outcome
On the acceptance set, the AI-assisted end-to-end handling time drops by at least 25% versus baseline, with a critical-error rate no higher than the current manual process.
Falsification condition
If the critical-error rate on the acceptance set is higher than the current process, or handling time does not drop, the hypothesis is not supported and an Iterate or No-Go recommendation follows.
Warning
Common mistake: writing scope as if it were a hypothesis (e.g. "we will build a customer-service assistant"). Scope is not a hypothesis; a hypothesis must include a specific expected outcome that data could disprove.
Info
This hypothesis builds on the already-approved IL 101 charter and the shared synthetic case SYN-CS-001. Even if your IL 101 project has a real industry sponsor, this module's initial usage scope still needs a fresh confirmation from the sponsor and process owner; the IL 101 approval does not substitute for it.
1Step 1 Pull the target outcome and success metrics from your approved charter.
2Step 2 Write a hypothesis card with four fields: scenario, what the AI does, expected outcome, and falsification condition.
3Step 3 Confirm the initial usage scope (which case categories, which data is usable) with the sponsor/process owner, and capture the approval in writing or meeting minutes.
Check yourself
What is the main problem with "introducing AI will improve the quality of customer-service replies" as a solution hypothesis?
Key takeaway: A good hypothesis must be specific enough that data could disprove it, otherwise the Evidence Review has no way to judge whether it held up.
