Case Study / Financial Services
Onboarding
that reviews
itself.
For a financial services company in the USA, every new customer's documents needed a human reviewer before an account could open. EdgeWeb built a document AI pipeline that reads, validates and scores each submission automatically — routing only the genuinely uncertain ones to a person. Try the exact decision logic below.
The Problem & The Approach
The problem
Every new account required a reviewer to manually check an ID photo against the application, confirm the document hadn't expired, and catch mismatches — a queue that grew faster than reviewers could clear it, and slowed down the honest majority of applicants to catch a small minority of problem ones.
The approach
Build a pipeline that runs the same checks a reviewer would, in seconds: extract the data, validate it against a rule set, score the confidence, and only hand a person the submissions that actually need judgment — with the specific issue already highlighted, not a generic file to re-review from scratch.
Core System Logic
Five stages, every submission.
Document Ingestion & OCR
ID photos and address proof are parsed the moment they're uploaded, extracting names, numbers and dates without a human retyping anything from a scan.
Automated Validation Rules
Extracted data runs against a rule set — does the ID match the name entered, is the document expired, does the address proof match — the exact first pass a reviewer used to do by eye.
Confidence Scoring
Every check contributes to one confidence score per submission, instead of a binary pass/fail that hides how close a borderline case actually was.
Risk-Based Routing
High-confidence submissions auto-approve in seconds. Borderline or flagged ones route straight to a reviewer with the specific failed check pre-highlighted — the live demo below shows exactly what that reviewer sees.
Full Audit Trail
Every automated decision keeps the evidence and the rule that triggered it, since a compliance review needs to reconstruct exactly why an account was approved months later.
Try It
Run the onboarding pipeline.
This runs the same validation and routing logic as the production system. Submit a clean document to see the auto-approve path, or a mismatched one to see exactly what gets flagged and why.
Live onboarding pipeline
0 submissions processed this sessionArchitecture
Chosen for auditability, not just speed.
Outcomes
FAQ
Does this loosen compliance to move faster?
No — every check a reviewer used to run manually still runs, just automatically. The system only speeds up the submissions that clearly pass every check; anything uncertain still reaches a human.
What happens to the submissions that get flagged?
They land in a reviewer's queue with the specific failed check already highlighted — a name mismatch, an expired document — instead of a generic file the reviewer has to re-check from scratch.
Can the validation rules be changed later?
Yes — rules and confidence thresholds are configured, not hardcoded, so compliance can tighten or relax specific checks without a code change.
How long did a project like this take?
An initial pipeline covering the core document types typically takes 10–14 weeks, with the rules engine and audit trail built alongside it from day one rather than added later.
Start a Conversation