In-Product Feedback & Contestability Framework

(Issue Reporting, Escalation & Response SLAs)

---

1. Purpose

This framework describes how users and organisations can report concerns, challenge outputs, and seek review or correction of askKira responses.

It ensures that:

• AI outputs can be questioned, reviewed, and improved

• Harmful or inappropriate responses can be escalated quickly

• Responsibility remains clearly human-led and accountable

---

2. Principles

askKira’s approach to feedback and contestability is guided by the following principles:

• Accessibility – reporting concerns must be simple and visible

• Proportionality – response speed reflects severity and risk

• Human oversight – all contested issues are reviewed by people

• Transparency – users understand what will happen next

• Continuous improvement – feedback informs system refinement

---

3. In-Product Issue Reporting

Reporting mechanism:
Users are provided with an in-product option to flag or report an issue directly from an askKira interaction.

Issue categories (minimum set):

• ☐ Factual inaccuracy or misleading information

• ☐ Safeguarding concern

• ☐ Bias or discriminatory content

• ☐ Harmful, inappropriate, or offensive output

• ☐ Data protection or privacy concern

• ☐ Security or misuse concern

• ☐ Other / general feedback

Information captured (proportionate):

• Prompt and system output (context-limited)

• Issue category

• Optional user description

• Organisational identifier (automated)

Gaps / to confirm:

☐ Exact UI placement and wording
☐ Ability to submit anonymously within an organisation
☐ User acknowledgement message

---

4. Contestability & Human Review

What can be contested:

• The accuracy, appropriateness, or safety of an AI output

• The suitability of askKira’s response for a specific professional context

• Potential bias or unfair treatment

• Handling of sensitive or safeguarding-related scenarios

Review process:

• All contested issues are reviewed by a human reviewer.

• Automated responses are never treated as final in contested cases.

• Safeguarding and high-risk issues bypass standard queues.

Outcomes may include:

• Clarification or correction

• Acknowledgement of limitation

• Prompt or guardrail adjustment

• Escalation to organisational safeguarding procedures

• System improvement action

Gaps / to confirm:

☐ Reviewer role definitions
☐ Use of specialist reviewers (e.g. safeguarding)
☐ Documentation of review decisions

---

5. Escalation Pathways

Tiered escalation model:

Level 1 – General issues

Examples:

• Minor factual errors

• Clarity or usefulness concerns

Handled by:

• Standard support review

Level 2 – Sensitive issues

Examples:

• Bias concerns

• Repeated inaccuracies

• Inappropriate tone

Handled by:

• Senior reviewer

• Pattern analysis for recurrence

Level 3 – High-risk / safeguarding issues

Examples:

• Safeguarding-related outputs

• Serious harm or risk indicators

• Data protection incidents

Handled by:

• Immediate human review

• Organisational safeguarding lead notified (where appropriate)

• Formal incident process initiated

Gaps / to confirm:

☐ Criteria for tier classification
☐ External authority notification thresholds
☐ Incident logging format

---

6. Response SLAs (Indicative)

Notes:

• SLAs apply during normal UK business hours.

• Safeguarding issues take precedence over all other work.

Gaps / to confirm:

☐ Public vs contractual SLA commitments
☐ Weekend / out-of-hours handling
☐ Customer notification format

---

7. Organisational Visibility & Oversight

For organisations:

• Designated admin users may receive:

  • Aggregated issue summaries

  • High-risk alerts relevant to their organisation

• Supports Trust-level governance and audit.

Gaps / to confirm:

☐ Admin dashboard availability
☐ Exportable audit logs
☐ Reporting cadence

---

8. Learning & Continuous Improvement

Feedback and contested issues are used to:

• Refine prompts and guardrails

• Improve safety filters and guidance

• Update onboarding and training materials

• Inform periodic risk reviews

No individual user is penalised for raising concerns.

Gaps / to confirm:

☐ Formal feedback-to-change loop
☐ Change log transparency
☐ Communication of improvements to customers

---

9. Relationship to Safeguarding & Data Protection

• askKira does not replace organisational safeguarding or data protection processes.

• Where a safeguarding concern arises, users are reminded to follow local safeguarding procedures immediately.

• Data protection incidents follow UK GDPR breach management processes.

---

10. Summary for Buyers

askKira provides:

• Clear, accessible routes to report, challenge, and escalate AI outputs

• Human-led review of all contested issues

• Proportionate response SLAs aligned to risk

• Organisational oversight suitable for public-sector assurance

This framework supports the UK Government’s expectations on accountability, transparency, and safety, while remaining practical for real school and Trust environments.