PrerequisiteCompleted Module 3 (Value and Measurement)
0 / 2 completed
Learning objectives
Write an explicit in-scope/out-of-scope list to prevent scope from quietly expanding mid-project.
List the project’s key dependencies (data, systems, people, approval processes).
Define the AI/human responsibility split and complete an initial data-check list.
Scope, Dependencies, and the AI/Human Split
30 min
Define Scope and Dependencies
In scope (Demo Service Company A’s knowledge assistant)
Public policy Q&A on returns, shipping, warranty, billing cycles, and similar topics
Recognizing and correctly citing documents when one is expired or contradicts another
Detecting situations that must hand off to a human (amount, legal language, rising frustration)
Out of scope (not addressed at this IL 101-201 stage)
Real-time individual order-status lookup (requires integrating the order system, a later integration topic)
Implementation details of access control for internal restricted data (cost sheet, internal escalation matrix)
Human agents’ performance review or scheduling system
Info
Common mistake: writing "this system might also need to do X eventually" into the in-scope list, which lets the PoV (IL 201) scope spiral out of control. The scoping principle is "get one thing right first, then expand", not "cover every related capability at once."
Example dependency list
Dependency
Description
Impact if missing
Knowledge-document versioning and expiry
Needs a queryable status field (current/expired/superseded)
Cannot recognize an outdated source; the assistant would give wrong answers
Formal approval of the complaint-escalation rules
Needs governance/legal sign-off on escalation thresholds (amount, wording)
The assistant might fail to hand off a case that should have gone to a human
Identity-verification process for data requests
Needs to integrate with the existing account system or a manual verification process
Could violate KD-014’s personal-data request procedure
Key takeaway: Scoping is not "a list you file away"; it is a tool you can point to every time someone wants to add a feature and say "that is not in the current scope."
30 min
AI/Human Split and an Initial Data Check
Example AI/human responsibility split
Scenario
AI’s role
Human’s role
General policy Q&A (normal)
Retrieves the current document and drafts a reply
Samples reply quality before customer-facing launch
Missing information (missing_context)
Proactively asks a clarifying question
Designs the list of key fields that must be clarified
Restricted access (restricted_access)
Recognizes and declines to disclose, explaining the correct channel
Defines and periodically reviews which documents count as restricted
No answer (no_answer)
Honestly states no information was found and refers the customer elsewhere
Decides the correct referral channel and follow-up
Warning
Human-AI-Human: a human defines the task and the rules (which documents are restricted, when to hand off to a person), AI executes the retrieval and drafting, and a human keeps final responsibility (sampling review, periodic rule review). It must not be designed so that whatever the AI decides is sent straight to the customer with no human check point.
1Initial data check (exercise) List the data sources your project will use, and mark: does each have a version/expiry field? a clear access classification? does it contain personal data?
2Compare with SYN-CS-001 Look at knowledge_documents.csv’s column design (doc_id, status, role_scope, contradicts_doc_id) and check whether your own data sources have similarly traceable fields.
Check yourself
A stakeholder asks the knowledge assistant to "eventually also look up a customer’s real-time order status." What is the best way to handle this request during IL 101 chartering?
Key takeaway: An initial data check is not a full data-governance review (that is IL 202’s job), but it must be able to answer "who can see this data, how stale is it, and could it contradict something else?"
