Indian state Assam builds shared digital layer to improve welfare targeting

Getting the right benefits to the right people, quickly and transparently, is a challenge for any state at scale. 

From old-age pensions, health insurance, housing and education loans, the Assam state government is working to deliver support more efficiently and accurately to its 36.8 million residents. 

The Indian state is building a common digital layer that links identity, beneficiary and scheme information. This allows eligibility to be assessed across programmes, rather than scheme by scheme. 

Speaking to GovInsider, Rashsul Hussain, the project manager for direct benefit transfers (DBT) at the finance department with the Government of Assam, says that welfare schemes have traditionally developed their own databases, application processes, eligibility checks and payment workflows.  

While a department may verify its beneficiary list accurately, it does not identify duplication or conflicting eligibilities across other schemes. 

He manages an initiative known as the Digital Infrastructure for DBT Schemes (DIDS), which aims to tackle this problem by enabling information to be assessed across schemes. 

Connecting identity, data exchange and payments 

Digital public infrastructure (DPI) for welfare delivery brings together three capabilities:  

1. Digital identity to establish who a beneficiary is,  

2. Interoperable data to assess eligibility across programmes, and  

3. Government-to-person (G2P) payment infrastructure to deliver benefits to the intended recipient. 

DIDS serves as the eligibility and registry layer within this broader welfare system, supported by payment and lifecycle-assurance mechanisms. 

“The shift is from ‘How do we digitise each scheme?’ to ‘How do we build common rails on which every scheme can deliver better?’” Hussain says. 

In Assam, Aadhaar-based identity, the Aadhaar Vault and bank-account validation provide the identity and authentication layer, while DIDS brings together citizen, beneficiary and scheme registries.  

This allows authorised eligibility rules and validation checks to operate against a common database. 

“Identity tells us who the person is; interoperability allows us to understand the person’s relevant relationship with government programmes,” Hussain says. 

However, interoperability does not mean giving every department access to all available information.  

Instead, it involves exchanging relevant data for a defined purpose under appropriate governance and controls. 

Identifying discrepancies across welfare schemes 

A unified registry helps governments move from scheme-by-scheme verification to cross-scheme targeting. 

Hussain distinguishes between three types of problems: duplicate beneficiaries who may be counted more than once; “ghost” records that do not correspond to a genuine, reachable person; and ineligible beneficiaries who do not meet a particular scheme’s criteria. 

Each requires a different form of validation. 

By matching beneficiaries within and across schemes, the system can identify potential duplication or conflicting benefits.  

Other government datasets also help validate eligibility criteria such as age, employment, or receipt of another benefit. 

Hussain gives the example of matching beneficiary records from a scheme targeting vulnerable women with those from another scheme supporting widows. 

When the schemes maintain separate beneficiary lists, each record appears valid in isolation.  

“Some targeting errors are invisible within a scheme; they become visible only when you connect to the schemes,” he says. 

Such a match, however, does not automatically establish fraud or ineligibility.  

It signals a case that requires further verification against the applicable scheme rules and the individual’s circumstances. 

Why human oversight remains essential 

Distinguishing a potentially fraudulent record from a genuine beneficiary whose information is incomplete, outdated, or inconsistent is a critical challenge in digital welfare delivery. 

Hussain says the system uses multiple validation checks, including identity, bank account details, cross-scheme matching, and relevant government datasets.  

These checks help determine whether a discrepancy reflects an eligibility issue, duplication, poor data quality, or information that needs updating. 

Frontline staff remain responsible for verifying cases and making decisions in accordance with the relevant scheme rules.  

The digital system should provide them with the evidence needed to make informed decisions, rather than automatically determine whether someone receives a benefit. 

This is particularly important in a country as large and diverse as India, where the same data discrepancy can have different explanations depending on an individual’s circumstances and location. 

Hussain describes the guiding principle as “flag first, verify before deciding”. 

Once an anomaly is detected, the case is referred to the relevant department or field functionary, alongside the information that triggered the flag.  

Officials then check the case against eligibility criteria and, where necessary, conduct field verification or consult other authoritative sources. 

The decision to correct a record, retain a beneficiary or include or exclude someone remains with the authorised official. 

The system also needs audit trails, mechanisms for updating information, and grievance-redress processes, so that decisions can be reviewed and corrected. 

“The machine can identify where to look; the human authority must decide what it means,” Hussain says. 

Where AI could add value 

According to Hussain, artificial intelligence (AI) could help identify unusual combinations across large and diverse datasets and prioritise flagged cases for human review. 

But AI-generated risk signals should not be treated as definitive evidence of fraud.  

Before such signals influence access to public benefits, models need to be validated, their performance and false-positive rates understood, and safeguards established around data quality, bias, explainability, auditability and human oversight. 

Even a small error rate can affect a substantial number of genuine beneficiaries when systems operate at the scale of millions of people. 

“In welfare delivery, an AI risk score should open a case for human review, not close a citizen’s case,” Hussain says. 

Governance is as important as technology 

“Technology is rarely the hardest part of interoperability; governance is,” Hussain says. 

Connecting systems through application programming interfaces (APIs) is only one part of the process.  

Departments must also agree on data definitions, ownership, access permissions, information-update processes and accountability. 

Data quality also requires continuous attention. A registry must be able to keep information current, identify inconsistencies, and provide ways to correct errors. 

Hussain says welfare targeting should be treated as a lifecycle encompassing inclusion, verification, updating, de-duplication, exit and audit feedback, rather than a one-time decision made during enrolment. 

Accountability must also remain clear when analytics or AI influence decisions.  

There should be an auditable record of what triggered a flag, what evidence was considered, who made the decision, and how it can be challenged or corrected. 

Citizen trust depends on meaningful grievance-redress and correction mechanisms. Digital systems should make government more consistent and responsive, not make it harder for eligible people to resolve errors in their records. 

“Digital systems can automate a decision process, but they cannot automate accountability,” Hussain says. 

Lessons for other states 

Hussain identifies several lessons for states seeking to use DPI to improve welfare targeting. 

First, governments should treat targeting as a system-wide capability rather than a feature of individual schemes.  

Shared infrastructure for identity, interoperability, eligibility, payments, and lifecycle assurance can allow multiple programmes to use the same capabilities. 

Second, systems should be designed to accommodate future schemes, not just existing ones.  

New programmes should be able to use established identity-resolution, data-validation, and payment infrastructure, instead of rebuilding these capabilities independently. 

Third, governments should focus on exclusion errors as well as inclusion errors.

Identifying potentially ineligible beneficiaries is only one part of effective targeting; ensuring that eligible people receive their entitlements is equally important.  

Failed payments, grievances, and other signals from the delivery process can help identify people who are missing in these schemes.

Finally, technology must be supported by institutional ownership, data governance, continuous data quality management, empowered frontline staff, and mechanisms for correction and grievance redress. 

“The goal is not simply to find the wrong beneficiaries. It is to build a system that is continuously better at finding the right beneficiaries — and ensuring they are reached,” Hussain says.