evidoria

← Back to browse

Good practice Imported

MyICSA Analysis Engine — Papua New Guinea's AI-Powered Visa Processing System

Papua New Guinea · Port Moresby · See the Papua New Guinea profile · See the Port Moresby profile

Evidence: Observational / pre–post Top 75% 47/100 · Ask Evidence Copilot about this practice

Papua New Guinea's immigration authority, working with AWS, built a generative-AI engine that cuts short-term visa processing from up to two weeks to under four minutes and auto-processes 95% of applications — a rare Pacific example of AI in frontline government service delivery.

14000+
Applications processed (first three months) (2025)
54000+
Supporting documents analysed (first three months) (2025)
up to 2 weeks
Typical processing time before deployment
<4 minutes
Typical processing time after deployment
95 %
Share of applications auto-processed
100000
Historical cases used for pre-deployment validation
1000 applications/day
Designed daily processing capacity
MyICSA Analysis Engine — Papua New Guinea's AI-Powered Visa Processing System

Details

Maturity
Scaling
Promoter
Immigration and Citizenship Services Authority (ICSA), Papua New Guinea, with AWS and NiuPay
Period
2024–present (proof-of-concept 2024; national go-live 2025)
Keywords
immigration, visa processing, generative AI, document verification, e-government

Context

Papua New Guinea's Immigration and Citizenship Services Authority (ICSA) worked with Amazon Web Services and local technology partner NiuPay to build the MyICSA Analysis Engine (MAE), a generative-AI system that reviews visa applications and supporting documents automatically. After a proof-of-concept in 2024, the system went live in 2025 and is now fully integrated into PNG's eVisa platform, handling short-term visa applications around the clock.

Objectives

Automate visa document review to cut processing times and manual workload for short-term visa applications, as one of the first uses of generative AI for a government workload in the Pacific Islands.

Activities

The model was validated against 100,000 historical sample cases before deployment, and the build took roughly 10 months from project start to national go-live. The system is designed to handle up to 1,000 applications a day, operating continuously as part of the eVisa platform.

Results

In its first three months of operation MAE processed more than 14,000 applications and analysed over 54,000 supporting documents, according to AWS's technical case study. The system cut typical short-term visa processing time from up to two weeks down to under four minutes, and now auto-processes about 95% of short-term applications without manual review — a shift AWS and ICSA describe as up to a 99% improvement in operational efficiency.

Conclusions

Independent PNG business and trade press have confirmed the system is embedded in the country's eVisa workflow and describe it as one of the first uses of generative AI for a government workload in the Pacific Islands. That said, every hard metric in this record traces back to the vendor (AWS) or the implementing team; no independent government audit, third-party accuracy assessment, or published error/appeal rate was found. Cost and budget figures were not disclosed in any available source, and no data-protection or algorithmic-transparency documentation was located — points any adopting government should probe before replicating.

Implementation
Indicative cost
Medium (€50k–€500k) — Cost and budget figures were not publicly disclosed by ICSA, AWS or NiuPay in any source reviewed.
Time to results
Short (< 1 year) — Proof-of-concept in 2024, followed by a roughly 10-month build from project start to national go-live in 2025.
Staffing & skills
Immigration and Citizenship Services Authority (ICSA), Papua New Guinea, Amazon Web Services (AWS) — cloud/AI technical partner, NiuPay — local technology partner

Conditions for success

  • Cloud AI vendor partnership (AWS) combined with a local implementation partner (NiuPay)
  • Pre-deployment validation against a large historical case set (100,000 cases)
  • Integration with an existing national eVisa digital platform

Common failure modes

  • No independent audit or third-party accuracy assessment of the AI decision engine
  • No published error or appeal rate
  • No disclosed data-protection or algorithmic-transparency documentation — all flagged as risks for replication

Where it fits

Governance type
national immigration authority with a private technology-vendor delivery partnership
Scale
national (up to 1,000 applications/day)
Income level
lower-middle-income (Papua New Guinea)

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful