|
SERVICES › SAMPLES › AI GOVERNANCE › A5-1
AI impact assessment AIA-01: Helpdesk reply drafting
Part of the AI governance review worked example for Quillmarrow Supplies Pte. Ltd. (fictional). Read it here in full or download the files.
Fictional organisation, illustrative only, not a client case study « Back to the AI governance review pack
|
|
|
|
| A5-1 · AI impact assessment AIA-01: Helpdesk reply drafting |
| Organisation | Quillmarrow Supplies Pte. Ltd. (fictional) |
|---|
| Document | A5-1 · AI impact assessment AIA-01: Helpdesk reply drafting |
|---|
| Version | 1.0 |
|---|
| Date | 30 October 2026 (illustrative) |
|---|
| Prepared by | Alex Lew, reviewer and sign-off |
|---|
| Client approver (role) | Chief Operating Officer |
|---|
| Label | Fictional organisation, illustrative only, not a client case study |
|---|
1. Summary
| Item |
Assessment |
| AI use |
AI-02: Helpdesk reply drafting |
| Owner |
Head of Customer Service |
| Tool |
Fictional Helpdesk AI feature, supplied by V-01 |
| Tier |
Tier 2 Medium, a working business category rather than a statutory classification |
| Oversight |
Human-in-the-loop: a Customer Service user checks and approves every reply before it is sent |
| Recommendation |
Approve with conditions |
| Decision needed |
COO to decide whether to approve the use with the conditions in section 8. Approval and implementation have not been evidenced. |
This assessment is illustrative. The agreed register identifies AIA-01 but records the evidence for AI-02 as partial: settings, data flow and review evidence have not been supplied. The recommendation is therefore conditional. It is not evidence that the feature is configured correctly or that checks have taken place.
2. Description of the use
The feature assists Customer Service staff in preparing replies to customer enquiries. It is intended to draft, not to make the final decision or send a reply without a user’s approval.
The described workflow is:
- A customer enquiry arrives in the helpdesk and is recorded as a ticket.
- The feature drafts a proposed reply using the ticket history and relevant knowledge-base material.
- A Customer Service user reads the enquiry and draft, checks the facts and edits the response where necessary.
- The user approves and sends the reply.
The workflow description does not establish which ticket-history fields or knowledge sources are available to the feature, or whether the approval gate is enforced in Quillmarrow’s configuration. Those points need confirmation.
The data used may include the customer’s enquiry, contact details, order and delivery information, previous correspondence and relevant knowledge-base content. The precise fields and any additional metadata remain to be confirmed. Customers may be affected by inaccurate or inappropriate replies or disclosure of their information. Customer Service staff may be affected by unclear checking responsibilities, misleading drafts or excessive reliance on the feature.
The customer web-chat widget is a separate use, AI-12. It is not yet live and may be piloted only after its controls are in place. This assessment does not approve customer-facing use of that widget.
3. Data and personal data
| Matter |
Assessment and evidence needed |
| Categories |
Enquiries and correspondence may contain names, contact details, order references, delivery addresses and other information customers choose to include. The actual fields available to the feature have not been confirmed. Staff identifiers and service metadata may also be processed; confirm the fields. |
| Source |
The described sources are the customer’s ticket, ticket history and helpdesk knowledge base. Confirm whether attachments, linked records or other sources are included. |
| Processing location |
V-01 says primary tenant data is hosted in Singapore, with backups in a named regional location, and supplied a general hosting-location schedule. The applicable location and backup arrangements for Quillmarrow’s configuration, including logs and any support access, remain unconfirmed. |
| Retention and deletion |
V-01 supplied a general retention-settings guide and says administrators can set retention and request deletion. Quillmarrow’s settings, applicable periods and deletion process have not been evidenced. |
| Vendor training |
V-01 says customer prompts and replies are not used for general model training and supplied a DPA schedule and a sample setting screenshot. Whether that position and setting apply to Quillmarrow’s service and configuration remains unconfirmed. |
| Malaysia-related data |
Enquiries may concern Malaysia orders fulfilled by Tanjung Contoh Logistics Sdn. Bhd. The data shared with the partner, processing locations and access arrangements for this workflow have not been supplied. |
The Head of Operations, as DPO, should review the confirmed data flow and proposed handling with the service owner. Questions about applicable legal requirements, including cross-border arrangements, should be referred to a Singapore-qualified lawyer. No legal conclusion is made here.
4. What could go wrong
Likelihood and impact are illustrative Low, Medium or High judgements for prioritisation, not measured results. Residual ratings are provisional, assuming the proposed controls operate as intended; implementation and effectiveness have not been verified. “Existing control” means the required or described control, not proof that it operates.
| ID |
Risk |
Likelihood |
Impact |
Existing control |
Proposed control |
Provisional residual rating |
Owner |
| R1 |
Draft gives the wrong order status or delivery information. |
M |
M |
User must check before sending. |
Check current order and delivery records against the draft; correct uncertainty before replying. |
Low |
Head of Customer Service |
| R2 |
Draft uses inappropriate, insensitive or misleading tone. |
M |
M |
User must review the reply. |
Provide examples of suitable tone and escalate complaints or sensitive cases. |
Low |
Head of Customer Service |
| R3 |
Draft discloses another customer’s details. |
M |
H |
Human approval is required; access settings are not supplied. |
Restrict source access, check recipient and all details, and test for cross-ticket disclosure before continued use. |
Medium |
IT Manager |
| R4 |
Draft invents or promises a refund, replacement or other remedy. |
M |
H |
User must approve the reply. |
Verify the applicable policy and obtain any required authority before making a commitment. |
Medium |
Head of Customer Service |
| R5 |
Staff over-rely on plausible drafts and fail to check facts. |
M |
M |
Human-in-the-loop is required. |
Train users, sample sent replies and give feedback; pause use if checks are routinely missed. |
Medium |
Head of Customer Service |
| R6 |
A vendor or tenant setting changes, widening data access or altering the approval workflow. |
M |
H |
V-01 describes a change-notice process; tenant settings are missing. |
Record an approved baseline, assign change review and retest approval and access after material changes. |
Medium |
IT Manager |
| R7 |
Malaysia-related customer data is exposed or handled through an unconfirmed partner or support-access route. |
M |
H |
Partner fulfilment is known; this helpdesk data flow is not supplied. |
Map fields, recipients, locations and access; obtain DPO review and legal advice where needed. |
Medium |
Head of Operations |
| R8 |
Staff cannot reconstruct why an incorrect reply was sent or detect a recurring problem. |
M |
M |
V-01 supplied a sample audit-log export; Quillmarrow’s records are not supplied. |
Confirm available logs, access and retention; record review findings and corrective actions. |
Medium |
Head of Customer Service |
If a suspected personal-data incident occurs, staff should stop the affected activity, preserve relevant evidence and escalate promptly to the DPO under Quillmarrow’s incident process. Whether notification is required should be assessed and, where needed, confirmed with a Singapore-qualified lawyer. A notifiable data breach must be notified to the PDPC as soon as practicable and no later than three calendar days after the organisation assesses it as notifiable.
5. Human oversight
AI-02 is human-in-the-loop. The Customer Service user is responsible for the reply and must check and approve it before sending. A plausible draft is not evidence that its contents are correct.
Before sending, the user should check:
- The recipient and ticket are correct, and the reply contains no other customer’s information.
- Order, delivery and account details match the current authorised records.
- Any proposed refund, replacement or other commitment follows the applicable policy and has the necessary authority.
- The reply answers the enquiry, uses an appropriate tone and does not present uncertain information as fact.
- Any sensitive or unresolved issue is escalated rather than guessed at.
As a proposed monitoring measure, the Head of Customer Service should review 20 sent AI-assisted replies a week for the first eight weeks, where that many are available. Record the number available and reviewed, errors found, corrections and follow-up. Include a mix of routine and more difficult enquiries; do not treat this sample as proof that all replies are correct. The owner should then decide and record an appropriate continuing review frequency.
Intervention triggers include disclosure or suspected disclosure of another customer’s data, an unauthorised commitment, repeated factual errors, a bypassed approval gate, an unexpected setting change or evidence that users are not checking drafts. The Head of Customer Service should pause the affected use where necessary, correct or contain the issue, notify the IT Manager and DPO as relevant, and record who authorises resumption after checks.
6. Vendor evidence status
The vendor questionnaire (A8) records the following supplied material and claims. General vendor material is not evidence of Quillmarrow’s tenant settings or operating effectiveness.
| Topic |
Recorded as provided by V-01 |
Missing or unconfirmed for AI-02 |
| Training use |
DPA schedule and sample setting screenshot; vendor says prompts and replies are excluded from general model training. |
Applicability to Quillmarrow’s service, actual tenant setting and any other improvement use. |
| Data flow |
General data-flow diagram and field inventory. |
Quillmarrow-specific inputs, sources, recipients and permitted fields. |
| Locations |
General hosting-location schedule. |
Applicable primary, backup and log locations, support access and Malaysia-related data routes. |
| Retention |
General retention-settings guide. |
Actual retention settings, deletion arrangements and evidence of operation. |
| Security |
Access-control guidance and a 2026 penetration-test executive summary. |
Quillmarrow’s access configuration, approval gate and remediation status relevant to this use. |
| Human approval |
Workflow diagrams and sample admin-setting screenshots. |
Demonstration that each AI-02 reply requires a Customer Service user’s approval before sending. |
| Logging |
Redacted sample audit-log export. |
Quillmarrow-specific log fields, availability, retention and review records. |
| Testing |
General reply-drafting testing overview. |
Results for Quillmarrow’s configuration, including disclosure and incorrect-promise scenarios. |
7. Framework mapping
This mapping identifies relevant governance questions and process checks. It does not establish conformity or legal compliance. No AI Verify technical tests were run.
| Framework |
Relevant areas, dimensions or principles |
Application to AI-02 |
| Model AI Governance Framework, Second Edition (PDPC and IMDA, January 2020) |
Internal governance structures and measures; determining the level of human involvement; operations management; stakeholder interaction and communication. |
Assign an owner, require human approval, document controls and establish an appropriate route for customer concerns. |
| Model AI Governance Framework for Generative AI (IMDA and AI Verify Foundation, May 2024) |
Accountability; data; trusted development and deployment; incident reporting; testing and assurance; security. |
Confirm responsibilities, data handling, approval settings, testing, incident escalation and change review. |
| AI Verify testing framework (AI Verify Foundation) |
Transparency; explainability; safety; security; robustness; data governance; accountability; human agency and oversight. |
Document the drafting workflow, its limitations, source controls, human checks and monitoring as process checks only. |
| PDPC Advisory Guidelines on Use of Personal Data in AI Recommendation and Decision Systems (March 2024) |
Relevant to considering personal-data handling where the described use falls within the guidelines’ scope. |
The DPO should review the confirmed data flow and purposes and seek Singapore-qualified legal advice where necessary. Applicability has not been determined here. |
8. Conditions of approval
- Confirm and document the AI-02 data flow, including fields, sources, recipients, locations and support access.
- Obtain and review Quillmarrow-specific retention, deletion and vendor-training settings and the applicable contract terms.
- Demonstrate that every draft requires a Customer Service user’s approval before sending; retain the test evidence.
- Restrict access to the data and knowledge sources needed for the role, and test for cross-customer disclosure.
- Provide staff with the checking and escalation instructions in section 5, including controls over refund promises.
- Test representative incorrect-information, inappropriate-tone and disclosure scenarios before relying on the workflow; record findings and remediation.
- Establish the proposed initial sample review, intervention triggers and incident escalation route.
- Record a settings baseline and a process for reviewing material vendor or configuration changes.
- Have the DPO review the confirmed data flow, including Malaysia-related handling, and obtain legal advice where needed.
- Obtain the COO’s documented decision. Until the conditions are evidenced, the recommendation must not be treated as proof of approval or control operation.
9. Unresolved decisions
No decision outcomes or completion dates have been supplied. The dates below are proposed decision targets, not records that a decision was made.
| Decision or outstanding point |
Owner |
Proposed target (days from report acceptance) |
| Decide whether to approve AI-02 with the conditions above. |
COO |
Day 14 |
| Confirm the source fields, approval gate and access settings to be tested. |
IT Manager |
Day 30 |
| Decide the checking instructions, sample-review approach and pause/resumption process. |
Head of Customer Service |
Day 30 |
| Review the confirmed data flow, including Malaysia-related handling, and decide whether legal advice is needed. |
Head of Operations, as DPO |
Day 30 |
| Confirm the applicable vendor-training position, retention settings and required evidence with V-01. |
IT Manager |
Day 30 |
10. Sign-off
| Capacity |
Role |
Signature |
Date |
| Prepared by |
The reviewer |
|
|
| Reviewed by |
Head of Customer Service |
|
|
| DPO comment |
Head of Operations, as DPO |
|
|
| Approved by |
COO |
|
|
About this sampleFictional organisation, illustrative only, not a client case study. Quillmarrow Supplies Pte. Ltd., its people, vendors, findings and dates are invented. Nothing in this document reports a client result. I personally review and sign off every deliverable. You appoint your own DPO. I support them. This is practical compliance support, not legal advice. For legal advice, use a Singapore-qualified lawyer. What's not included - Legal advice, legal opinions on vendor terms or regulatory representation.
- AI Verify technical testing, model benchmarking, penetration testing or security engineering, unless the relevant work is separately scoped.
- Building an AI system, implementing integrations or changing live permissions and configurations.
- Certification, a compliance guarantee or an assurance that a model cannot fail.
- An organisation-wide ISO/IEC 42001 readiness engagement unless expressly included in the quote.
- Healthcare-sector work or on-site delivery.
ISO/IEC 42001 work is a readiness check only, never certification. Questions about this sample: DM @alewyl1 on X, or ask for a free intro call (coffee chat).
|
|
|
|
|
|