Client · 2025–2026 · Agentic AI
Up to 350 questions per review. Night shifts. Review process redesigned.
4×
Throughput per analyst
~80%
Reduction in administrative work
350
Questions analysed and redesigned
Event Driven Reviews: triggers activated a manual review process of up to 350 questions. Depending on which triggers fired, an analyst answered anywhere from around 15 to 350 questions. Process redesigned with self-service customer intake, prefilling from existing systems and a single unified platform. Questions without risk coverage scrapped, related questions merged. Around 80% reduction in administrative work. Management started a formal project for implementation and broader rollout.
Process map
Hover over a component to see what was going on, and what changed.
Old situation
Trigger-driven reviews, fully manual
Risk triggers (events)
Event Driven Review
Customer behaviour or new data activates risk triggers. The review counts up to 350 questions. Depending on which triggers fired, an analyst answered anywhere from around 15 to 350 questions manually. Volume grew faster than capacity.
Manual intake
350 questions, no prefilling
Every review: 350 questions answered manually. No reuse of previously collected data. Starting from scratch every time.
Progress tracking
Excel, email
Status in separate spreadsheets and email chains. No visibility on capacity or throughput time. Every escalation via the manager.
Capacity pressure
Night shifts, high turnover
Analysts working night shifts to keep up with the backlog. The regulator wasn't waiting. Every day of delay: new reputational risk.
Redesign and AI agents
Azure AI Foundry
New platform
Faster, explainable, scalable
Self-service intake
One overarching platform
All source systems integrated. Analysts work in one environment. Status, backlog and capacity directly visible.
Prefilling engine
~80% automatable
From 350 manual questions to around 70 real decision points. The rest filled in automatically based on data from previous reviews and external sources.
Question redesign
Scrapped & merged
Questions without demonstrable risk coverage scrapped. Related questions merged into logical units. Remaining questions grouped logically per risk category.
Result
4× throughput per analyst
Analysts processing four times as many reviews in the same time. No night shifts. Backlog cleared. Regulator satisfied.
What was replaced
What's there now
Click or hover for context per component
Stack
PowerApps
Power BI
Python
Databricks
Approach
Process redesign
Question rationalisation
Self-service intake
Prefilling from source systems
Platform consolidation
Workflow automation