Government
An AI position that survived the governance committee.
A provincial government department
Illustrative engagement
- Evidence label
- Illustrative engagement
- Duration
- 7 weeks
- Diagnostic hierarchy
- Levels 1 and 3
- Frameworks applied
- GoTo AI Navigator™, GoTo Trust Framework™
01
Evidence statement
What this page claims, and what it does not
Structurally representative of the firm’s work in this sector. Client identity withheld. Figures shown are the kind of measures agreed at the Define stage, not reported results.
02
The brief and the diagnosis
02The presenting request
Advise on adopting an AI assistant for public enquiry handling.
“We have been shown a chatbot. Executive wants to know whether to buy it.”
What we found
The department could not answer four questions the governance committee would certainly ask: who is accountable for a wrong answer, what data leaves the department, what happens when the assistant is wrong in public, and how the decision gets reviewed. The tool was not the decision. The governance position was.
Diagnostic hierarchy: Levels 1 and 3
03
Approach
03How the work was run
Approach.
- 01AI Readiness Assessment scoring maturity, opportunity and risk separately.
- 02One narrow use case prioritised: repeat public enquiries with answers that already exist in approved published material.
- 03Architecture specified as retrieval over the approved corpus with citations on every answer, rather than fine-tuning.
- 04Refusal behaviour written into the acceptance criteria and tested before sign-off.
- 05Supplier register entry recording tool, provider, data location and no-training commitment. POPIA role established and recorded.
- 06A named human owner for the assistant’s output, and a stated escalation route for public error.
Delivered
AI Readiness Assessment with maturity score and risk register
AI governance framework and acceptable use policy
Committee submission pack
Pilot assistant over the approved corpus
Staff awareness session and training record
04
Measures
04Success measures
Specific, sourced and dated. Agreed at the Define stage.
The figures below are illustrative targets of the kind agreed at Define. They are not reported results, and this page does not present them as such.
| Measure | Baseline | Target | Source |
|---|---|---|---|
| Repeat enquiries resolved without staff handling | Not measured | 35 percent of the identified enquiry set at 90 days | Enquiry log and assistant query log |
| Answers traceable to an approved source | Not applicable | 100 percent, tested at acceptance | Acceptance test record |
| Refusal on out-of-corpus queries | Not applicable | 100 percent, tested at acceptance | Acceptance test record |
Handover
Governance framework, supplier register entry and training record handed to the department. Named owner recorded.
05
Related
Frameworks applied
Services involved
Book a Discovery Session
The most useful first step is a discovery conversation.
A structured session that produces a written problem statement, whether or not it leads to an engagement. If the problem is not one we should be paid to solve, we will say so.
Enquiries acknowledged within one business day, always.