Financial services · Enterprise AI
Risk & Control Self-Assessment
AI can raise the risk.
It can't own the decision.
I led the product and interaction design of dbRCSA: connecting risk signals, AI recommendations and accountable assessor decisions.
Built, released and successfully adopted.
The decision in the delivered workflow
A connected risk-assessment product. Delivered.
A change without context isn't enough.
Business or data changes could affect an inherent-risk rating. The assessor needed to see what had shifted, what informed the new judgement and whether to accept it.
Make the reasoning inspectable.
I designed connected views for portfolio-level signals, assessment work, previous and indicative ratings, contributing factors, explicit human decisions and the audit trail.
One connected assessment journey.
Portfolio monitoring → work queue → investigation → human decision → audit record. Each view keeps the context needed for the next action.
I led the end-to-end product and interaction design, including screens, workflow logic and stakeholder management. My design remit was to translate the rating methodology and underlying data into a journey assessors could navigate, inspect and act on.
Show what changed before asking someone to open it.
I surfaced the previous and indicative ratings in the queue so assessors could judge the direction of change before opening a record. Keeping the comparison at triage level preserves the reason for prioritising an assessment alongside its desk and trigger.
02Assessment queue. Previous versus indicative and direction of change, visible before opening a record.
Assessment 1234-01 · Inherent risk
The design decision
Keep the comparison at the point of triage.
- Previous position: the assessor retains a visible baseline.
- Indicative proposal: the changed rating and direction are explicit.
- Context retained: the same assessment connects the comparison to deeper investigation.
Inherent risk is the exposure before controls; control effectiveness describes the safeguards; residual risk is what remains after them.
Compare the change before judging it.
I kept the previous rating, Important, beside the indicative Critical proposal, with likelihood, severity and rationale attached to each. The assessor can compare the evidence without losing the baseline; Critical remains provisional until they decide.
03Rating comparison. Keep the previous state, the AI proposal and their explanations in one view.
Trace the recommendation back to its drivers.
The next question is why the assessment moved from Important to Critical. I grouped the contributing factors and separated likelihood from severity and impact, allowing the assessor to investigate the basis of the recommendation while retaining the assessment context.
Product
Complexity of government bonds
Client
Retail exposure and susceptibility
Advice
Change in advised activity
Distribution
Sales route and risk context
04AI reasoning. Product, client, advice and distribution factors underpin the proposed likelihood and severity.
The interaction
Now make the decision.
The real product was shipped. Explore its decision logic with fictional sample data: accept Critical, reconsider with sample information, override the rating or mark the assessment not applicable. Every route works through button clicks.
Demonstration only. All rationales and outcomes use fictional sample data. The reconsideration response is scripted, not generated by an AI model. No Deutsche Bank system is connected; no information is transmitted, uploaded or saved.
Make the approval explicit. Keep its history.
I kept agreement, disagreement and application to similar tasks as separate actions. The decision record then retains the rating, explanation and human action so the next reviewer can reconstruct what happened.
One decision. A defined scope.
Approving one assessment does not silently approve similar tasks. Disagreement is a supported route with its own rationale.
05Human decision. Agree, disagree or act on similar tasks as distinct choices.
A record another person can inspect.
The current rating, explanation and event history remain together after the assessment is closed.
06Decision record. A traceable close-out, including rationale and audit events.
A connected assessment experience, in operational use.
dbRCSA was built, released and successfully adopted at Deutsche Bank. It connected portfolio-level risk signals with individual assessments, evidence, human decisions and their records.
Product and interaction design leadership
I led the screens, workflow logic and stakeholder management across the connected assessment journey.
Keep context at each decision point
- Surface previous and indicative ratings in the work queue.
- Attach the reasoning to the recommendation.
- Separate individual approval from action on similar tasks.
- Retain the human decision and its rationale in the record.
From a signal to a traceable decision
Assessors could move from changing risk information through investigation to an explicit decision, with a record available to the next reviewer.