Workbook plus evidence · Internal evaluation

Harbour Lane customer-support assistant

See how a workbook and six fictional evidence records support a conditional internal-evaluation decision.

Start a free assessment Compare all worked examples

Context and problem

A specific task with a visible human boundary.

A fictional Australian retailer wants to improve customer-support triage and reply drafting while keeping people responsible for every message and sensitive decision.

Manual triage and drafting are slow and inconsistent, and enquiries are often moved between queues.

Proposed AI task: Classify enquiries, suggest priority and routing, retrieve approved information and draft a source-linked reply.
Human decision boundary: A support agent decides whether and what to send. Authorised people make all refund, account and sensitive decisions.

Starting point

The input changes the strength of the report.

A completed workbook and six fictional supporting evidence records, including a baseline, operating boundaries, a synthetic test set and an evaluation plan.

Material governance questions

Are privacy, security, supplier, retention and access controls adequate for the proposed evaluation?
Do routing, retrieval and mandatory-escalation tests meet approved thresholds?
Are vulnerable-customer, accessibility, correction and complaint pathways ready?

Representative findings

A measurable fictional service baseline and human-review boundary are recorded.
The evidence supports an internal evaluation, not customer exposure or production use.
Senior governance and incident ownership still need confirmation.

What remains uncertain

No live system, customer data or production control was tested.
The fictional evidence cannot establish performance in the organisation's real environment.

Controls and conditions

Use synthetic and approved de-identified examples only.
Keep outbound replies, refunds and account decisions under authorised human control.
Approve privacy, security, monitoring, escalation and pause controls before evaluation.

Recommended next step

The report supports the next human decision.

Seek a human decision on a time-boxed internal evaluation after the blocking conditions are accepted.

Download the complete fictional assessment report (PDF)Open the complete worked example in AI Governance (sign-in required)Download the fictional Harbour Lane Starter Pack (sign-in required)Open the fictional Harbour Lane evidence pack (sign-in required)

Start with your own short description.

One eligible account receives one free assessment. You do not need a workbook to begin.

Start a free assessment