Scope: AI use cases in customer support and sales enablement.
Date: .
Executive summary
We assessed the client's AI use cases against five governance frameworks. Overall: solid foundations in security and data quality; gaps in GDPR documentation, AI ethics governance, and data sovereignty mapping. Three high-priority recommendations at the end of the report.
Findings by framework
GDPR: Partial
DPA in place with primary AI provider. Lawful basis documented for support chatbot.
Gaps
No ROPA entry for AI use cases.
Privacy notice does not mention AI processing.
No process documented for subject-access requests involving AI-held data.
Recommendations
Add AI use cases to Record of Processing Activities.
Update privacy notice; include AI in DPIAs for new use cases.
AI Ethics: Gaps
No formal accountability for AI decisions. Support responses not tested for bias. Sales scoring uses historical data that may under-represent newer segments.
Recommendations
Define AI decision owner (e.g. Head of Support, Head of Sales).
Run bias check on support response samples before scaling.
Add ethics criteria to AI procurement checklist.
Data Quality: Good
Support tickets and CRM data are well-structured. Some duplicate accounts in CRM; recommend deduplication before feeding to sales AI.
Recommendations
Run deduplication on CRM before scaling sales AI.
Define quality metrics and baseline before next phase.
Security: Good
Access control in place. Audit logs enabled. Provider has SOC 2. One gap: no prompt-injection testing documented.
Recommendations
Run OWASP LLM Top 10 checklist; document prompt-injection mitigations.
Sovereignty: Partial
Provider offers EU region; currently using US default. UK/EU customer data may be processed in US.
Recommendations
Switch to EU region for support and sales AI; document in data-residency matrix.
Review contract for exit and data-deletion clauses.
Prioritised next steps
Immediate. Switch AI provider region to EU for UK/EU personal data.
Within 4 weeks. Update ROPA and privacy notice for AI use cases.
Within 8 weeks. Assign AI accountability; run bias check on support samples.