Data flow mapping
Where personal data enters, is embedded, cached, logged and retained across the full AI chain.
All services
Governance & Trust
Privacy & Data Residency
Privacy decided at design time costs a fraction of privacy retrofitted.
- Every hop
- Prompt, index, cache and log mapped
- By region
- Residency enforced in architecture
- DPIA-ready
- Assessments and evidence prepared
The problem we solve
AI systems copy data into places privacy programmes never mapped: prompts, embeddings, caches, logs and evaluation sets. We map those flows, apply minimisation and retention at each hop, and enforce residency in the architecture itself.
What the work includes
Minimisation and redaction
PII detection and redaction at ingestion and at the prompt boundary, with reversible tokenisation where needed.
Residency architecture
Regional processing, model selection by region and contractual controls on sub-processors.
Retention and deletion
Retention rules applied to indexes, embeddings and logs, including verifiable deletion on request.
Assessments
DPIAs, legitimate interest assessments and records of processing written for AI-specific processing.
How the engagement runs
A sequence you can plan a quarter around.
- 01
Map
Discovery of AI data flows, sub-processors and current retention behaviour.
- 02
Design
Controls, residency architecture and assessment drafting with privacy counsel.
- 03
Implement
Technical controls delivered and evidenced, with ownership handover.
Common questions
Before you commit.
- Can we use US-hosted models under EU rules?
- Sometimes, with the right controls and contracts. Where it does not hold up, we design a regional or self-hosted path.
- How do you handle deletion from a vector index?
- Tenant and subject keys are carried through indexing so deletion is a targeted operation, not a rebuild.
Related practices
AI Governance & Compliance
Model registry, risk classification, documentation and audit trails mapped to the EU AI Act, ISO 42001 and internal policy.
ViewAI Security & Red Teaming
Prompt-injection defence, data exfiltration testing, agent abuse scenarios and adversarial evaluation before attackers get there.
ViewIndependent AI Assurance
Third-party review of models, vendors and controls — an objective read for boards and risk committees.
ViewTalk through privacy & data residency.
A 45-minute briefing with the people who would run the work — scope, timeline and a straight answer on whether it is the right next step.