How PolariseMe helped a leading European financial institution automate payment investigations on Appian—combining AI, deterministic rules and human oversight without giving up control.
CASE STUDY | FINANCIAL SERVICES
Turning 3,000 Manual Investigations a Month Into a Governed, AI-Assisted Process
At a Glance
~3,000
payment-investigation emails / month
Financial Services
Payment Investigations
Appian
AI + process orchestration
Pilot complete
production launch preparation underway
The Challenge
Every month, a leading European financial institution receives roughly 3,000 payment-investigation emails. Investigators had to determine whether requests were in scope, extract payment details, identify the underlying transaction, assess missing information and maintain continuity as follow-up correspondence arrived.
The objective was to automate that journey without trading away governance: eligible investigations should move straight through, while exceptions and cases missing information remain under human control.
The institution wanted AI in this process, but it wasn’t willing to trade away governance to get it.
Every significant action still had to be controlled, explainable and auditable.
The Solution
A Governed, AI-Assisted Investigation Process
Use AI where judgment is needed
AI Agents and AI Skills interpret unstructured emails and attachments where contextual judgment is required.
Use rules where outcomes should be predictable
As the build matured, bounded activities such as transaction search and validation moved toward task-specific AI Skills and deterministic rules.
Keep people in control
Investigators can review, accept, reject or override AI output, with every action captured as part of the case record.
Results & Business Impact
Straight-through processing demonstrated
Eligible investigations can move from email intake through to case creation automatically.
Exceptions stay under human control
Cases needing judgment or missing information are routed to investigators rather than forced through automatically.
More efficient as the architecture matured
Processing time / investigation
24 min → 4–5 min
Token usage / investigation
800k → 100k
Figures are per investigation and reflect solution maturation during the pilot, not a validated comparison with the prior manual process.
Get the Full Case Study
See how the architecture evolved as the team worked out where AI belonged — and where it didn’t. The full case study covers the detailed solution architecture, human-review and audit model, production-readiness design, and how AI Agents, AI Skills and deterministic rules were matched to the right activities.