AI Teams
Social CRM
Government & Integration
Solutions
Resources
1,000,000 applications. One complete history for every person.
Housing subsidies are a major social welfare policy run jointly by central and local government. Subsidized households grew from about 150,000 in 2020 to over 800,000 in 2026, and the program expanded from one subsidy type to three. The system has to serve the public nationwide, case officers in 22 cities and counties, partner banks, and the central competent authority at the same time.
- Client
- National Land Management Agency, Ministry of the Interior
- Contract period
- 12 months from October 2026 (in progress)
- Security level
- Medium
- Scope
- Feature expansion, external integration and operations, AI adoption, on-site support and help desk, nationwide training
Scattered data, caseloads doubling year after year.
An applicant’s identity, income and assets, land records and subsidy history all sit in systems run by different agencies. Officers had to look each one up and cross-check by hand, and there was no way to see one person’s applications, resubmissions, payouts and recoveries across the years in one place.
As caseloads kept doubling, attachments were still checked one image at a time, and a single round of document requests often took more than a month. Decision logic lived in the code, so every change to a regulation or policy parameter meant changing code, retesting and redeploying, which made it hard to keep three subsidy types in sync.
Data from many agencies, organized around each person
Household registration, land registry, tax and finance (via T-Road), securities and other social benefit data are brought together, with APIs for cities’ and counties’ own systems and for partner banks. Addresses use the Ministry of the Interior’s address identifier, so cross-database matching moves from text comparison to precise location.
- Every applicant has a complete history across years and subsidy types
- Household registration data exchange uses tokenized personal data, file encryption and data fragmentation
- Every data change keeps its history and can be traced record by record for audits and administrative disputes
AI prepares, the officer decides
For city and county reviewers, AI first identifies attachment types, collapses images that need no review and flags what is missing, then drafts a plain-language request for the missing documents. The applicant is notified only after the officer confirms.
- Central staff can describe a query in plain words and get statistics across cities and counties, using read-only accounts and predefined queries, with every request logged for audit
- Public-facing and officer-facing AI customer service share one knowledge base; answers cite the legal source, and when the basis is not enough, it does not answer
- Subsidy amounts and eligibility are always decided by rules, never by AI; inference data stays in the country and is never used for training
Hello, your rent subsidy application is missing a copy of your household register. Please upload it within 14 days (by Nov 25), or bring it to your district office. We’ll continue the review once it’s in.
Policy changes take effect the same day
Policy logic is pulled out of the code into five layers of parameters: regulation, policy, city/county, workflow and display. The central government sets the legal and policy baseline; each city and county sets its own intake period, required attachments, resubmission deadline, notice wording and local top-ups within its authority.
- Every parameter has an effective date, a version and a change history
- Each case always uses the version in force on its acceptance date
- All three subsidy types update together, with no code changes
No slowdowns at peak, and you know right away when a source drops
Reads and writes are separated, so queries at peak season never slow down intake. Each connected source shows its last successful sync, record count and status, and timeouts or repeated failures automatically alert the source’s officer and the help desk.
- Automated regression tests run on every release, with zero regressions in existing features
- On-site staff and help desk close at hand
- Training for all 22 cities and counties
More efficient reviews, faster answers for the public.
| Area | Results |
|---|---|
| Scale | Supports 1,000,000 applications to date; reads and writes are separated, so peak-season queries never slow down intake |
| Data integration | Records from multiple agencies are linked to each person, so officers see an applicant’s full history on one screen instead of searching system by system |
| Review efficiency | Attachments are pre-tagged and document requests are drafted automatically, so notices go out as soon as a case is opened, not after a full read-through; target accuracy for document type identification is 95% |
| Management decisions | Cross-criteria statistics go from exporting several reports and comparing by hand to a single query; the target is to cut hours down to minutes |
| Policy agility | Policy changes are now parameter settings instead of code changes, and all three subsidy types update together; cities and counties maintain their own local rules |
| Reliable operations | Automated regression tests run on every release, with zero regressions in existing features |
Maritime and Port Bureau, MOTC (Taiwan)PMO and technical advisory for the maritime personnel digital upgrade
Department of Medical Affairs, Ministry of Health and WelfareMedical Data Integration Hub
Taoyuan City Government Economic Development BureauIndustry analytics platform
"Warm Companion" projectAI volunteer care platform for rural communities
Start your AI transformation today.
