accountability-patterns

AI Accountability Matrix

This example is generic and illustrative. It does not describe a real production system.

1. System Metadata

Field Value
System / model name Support Routing Assistant
Version 0.4.0
Use case classify support tickets and recommend routing path
Risk tier Medium
Environment Staging
Review date 2026-04-26

2. Named Ownership

Role Named owner Accountable for Backup owner
Business owner Example Support Lead user/business outcome, acceptable use, benefit-risk tradeoff Example Operations Lead
Technical owner Example ML Owner model behavior, validation, reliability Example Platform Owner
Data owner Example Data Steward data quality, lineage, retention, permitted use Example Privacy Analyst
Risk / governance owner Example Risk Owner risk controls, policy adherence, escalation rules Example Compliance Lead
Operations owner Example Operations Owner monitoring, incident handling, runbook execution Example Support Manager

3. Decision Rights

Decision Decision owner Required evidence Escalation trigger
Approve development start Business owner intake form, initial risk screen unclear owner or high-risk use case
Approve staging release Technical owner + governance owner validation report, model card draft, risk assessment failed required gate
Approve production release Product owner + governance owner release gate review, monitoring plan, rollback plan unresolved high-impact risk
Override AI output Human reviewer reason code, human review note repeated overrides or high-risk case
Retire or suspend system Business owner + operations owner monitoring evidence, incident record, business need review severe incident or sustained degradation

4. Human Oversight and Escalation

Trigger Required human action Owner SLA Evidence retained
Low confidence review and approve routing decision Support Lead 1 business day task ID, confidence, reviewer decision
High-risk output immediate human review Risk Owner 4 hours ticket category, escalation note, final action
User appeal human review of original decision and supporting evidence Support Manager 5 business days appeal record, reviewer rationale, outcome
Safety, privacy, legal, or fairness concern escalate to governance owner Risk Owner same day incident record, severity rating, remediation action

5. Redress and Appeal Pathway

Step Description Owner Target time
Intake affected user or operator submits challenge Support Operations same day
Triage classify severity and assign reviewer Support Manager 1 business day
Review human reviews evidence and system output Human Reviewer 3 business days
Decision uphold, correct, reverse, or escalate Support Manager 5 business days
Feedback loop update controls, training, UX, or monitoring if needed Risk Owner next governance review

6. Audit Trail Requirements

Minimum records to retain:

7. Accountability Review

Review question Answer
Is every high-impact decision assigned to a named owner? Yes
Is there a documented override path? Yes
Is there a documented appeal path? Yes
Are escalation SLAs defined? Partially, production SLAs need final confirmation
Are audit records sufficient to reconstruct a decision? Yes for staging, retention policy requires approval before production
Are unresolved accountability gaps tracked as risks? Yes

8. Open Accountability Gaps

Gap Severity Owner Required action Due date
Production retention policy not finalized Medium Data owner confirm log retention and masking policy 2026-05-03
Appeal process not tested with realistic cases Medium Support Manager run tabletop review with sample appeal cases 2026-05-10
High-risk escalation SLA needs governance approval High Risk Owner approve final escalation SLA before production gate 2026-05-05