GenAI-enabled medical claims straight-through processing
Medical claims require a human to read every document before payment. AIA instead built a three-tier GenAI stack (OCR intake, report summarization, auto-adjudication plus FWA anomaly detection) so most claims complete without human touch.
Insurance · Four main Asia-Pacific markets (incl. Korea and Hong Kong; company-reported)
The compounding gains came from rebuilding the whole intake-to-payment pipe (OCR, summarization, auto-adjudication, e-payment), not from dropping a model into one step; and the company itself is careful to credit digital adoption and process redesign alongside GenAI, which is the honest way to report these numbers.
AI value · End-to-end claims straight-through processing rate
Company-reported
73% (December 2024)
Company-reported in a single company-authored partner-content article; the source itself attributes the gain to digital adoption and process redesign, with GenAI as one component.
Before
Submit physical forms and paper medical documentation → Manually key, validate, and route documents (multilingual, handwritten, non-standard formats) → Manual fraud, waste, and abuse checks → Process payment manually
After
Photograph invoices and documents with a phone → Extract key information from non-standard, multilingual, handwritten documents → Score claims against 20+ risk indicators with encoder-decoder anomaly-adjusted thresholds → Pay approved claims electronically (99% reported adoption)
Human boundary
AI holds adjudication authority for minor claims that meet auto-adjudication rules (company-reported 75% auto-adjudication). Complex claims remain with human assessors. The auto-approval threshold and denial authority are not disclosed in any retrieved source; that gap is recorded, not filled.
Why it matters
Medical claims require a human to read every document before payment.
How the work changed
Before
How the work ran before the change.
Step 1 of 4
Customer
Submit physical forms and paper medical documentation
ControlManual completeness check
Step 2 of 4
Claims intake staff
Manually key, validate, and route documents (multilingual, handwritten, non-standard formats)
ControlManual data entry QA
Step 3 of 4
Fraud team
Manual fraud, waste, and abuse checks
ControlRule-of-thumb manual review
Step 4 of 4
Payment operations
Process payment manually
ControlManual payment run
What changed
Medical claims require a human to read every document before payment.
Decision rightAI completes a bounded lane; humans own the residual
Process model built from the published workflow evidence for AIA Group. Every step, actor, and control appears in full below.Every step, actor, and control
Exception path
Poor-quality images are flagged and the customer is prompted to retake photos; complex claims route to human assessors with GenAI summaries; high FWA-risk claims route to investigation.
Decision authority
AI holds adjudication authority for minor claims that meet auto-adjudication rules (company-reported 75% auto-adjudication). Complex claims remain with human assessors. The auto-approval threshold and denial authority are not disclosed in any retrieved source; that gap is recorded, not filled.
Before
#
Actor
Action
Control
01
Customer
Submit physical forms and paper medical documentation
Manual completeness check
02
Claims intake staff
Manually key, validate, and route documents (multilingual, handwritten, non-standard formats)
Manual data entry QA
03
Claims assessor
Read every laboratory result, scan, and medical report end-to-end to adjudicate
Claims staff manually keyed and validated paper documents; assessors read every medical document in full; fraud checks and payments were manual.
Human role after
Assessors review GenAI-produced summaries of complex reports (reported review time halved) and concentrate on judgment cases; minor claims, data entry, routine fraud screening, and payment execution no longer require human touch.
AI role
GenAI OCR extracts and validates claim documents; auto-adjudication engine straight-through processes minor claims; GenAI summarizes complex medical reports for human assessors; an encoder-decoder anomaly model applies 20+ indicators for fraud, waste, and abuse.
June 2020 to December 2024 · Company-reported across four main markets; whether the figure is group-wide or market-subset is not specified in the source
Company-reported in a single company-authored partner-content article; the source itself attributes the gain to digital adoption and process redesign, with GenAI as one component. Not independently corroborated; AIA's FY2024 annual results corroborate the broader trajectory (92% of service transactions straight-through processed; back-office unit costs -43% from 2020 to 2024) but not this claims-specific figure.
Auto-adjudication rate
Company-reported
41%→75%
June 2020 to December 2024 · Company-reported, four main markets
Company-reported; single source; not independently corroborated.
Claim processing time (Korea)
Company-reported
Two days or more→Less than 25 minutes
As of article publication (July 2025) · Korea market; claim volume not disclosed
Company-reported; statistic type (median, mean, or best case) and volume not disclosed.
Assessor review time per complex medical report
Company-reported
Pre-GenAI review time (absolute value not disclosed)→Cut in half
As of article publication (July 2025) · Complex medical report reviews in deployed markets
Relative reduction only; absolute baseline not disclosed; company-reported.
What leaders can reuse
Anti-pattern
Attributing multi-year straight-through processing gains to GenAI alone when the driver is a combined digital adoption, process redesign, and AI program.
Questions
01Where does our auto-approval authority actually sit, and would we disclose the threshold as AIA declined to do?
02Which of our intake metrics combine process redesign with AI, and do our disclosures say so?
03What is our equivalent of the Korea 25-minute claim, and is it a median or a best case?
Portability conditions
High-volume, document-heavy claims or intake workflow
Willingness to rebuild data and platform foundations first (AIA cites a US$800m technology rebuild with 90%+ of compute in cloud)
In-house capability to handle non-standard, multilingual documents
A defined auto-adjudication rule set with an exception path to human judgment
Reputation risk
low
Evidence and authority
What the public record supports.
Current · updated
2 primary; publication outcomes are verified and reported.
Bundle 1.0.0 · reviewed 2026-09-06 · stable ID 3e04a1512c4ea48e
