India's direct benefit transfer programmes moved payments straight into beneficiaries' bank accounts. That solved one problem and exposed another: the quality of the list. A beneficiary list is usually built from several registers kept by different departments, with different identifiers, spellings and update cycles.
Why lists go wrong
- The same person appears in several registers with small differences
- Eligibility changes, but the list is refreshed in batches
- Payments fail for account reasons nobody is assigned to fix
- Field verification results live outside the system
What agents can do — and must not
Matching people across registers, checking eligibility rules and grouping payment failures by cause are exactly the kind of evidence-gathering agents do well. Deciding that someone is ineligible is not. In our designs every flag goes to an officer with the evidence assembled, and the agent has no ability to change the list.
Exhibit 1
Flag, verify, decide
Agent designs from our AI & Agentic Engineering practice
| Step | Who |
|---|---|
| Find likely duplicates and ineligible entries | Beneficiary Dedup Agent |
| Group payment failures and suggest fixes | Payment Failure Analyst |
| Verify in the field | Officer |
| Include, exclude or correct | Officer, with an audit trail |
Note: Designs, not delivered results.