The economics of recovery: success-aligned AI for the public sector
A vendor paid only from verified arrears collected carries the risk a department cannot. Field notes on how these contracts are structured, measured, and audited — and the clauses that decide whether they survive.
The arrears problem
Most Indian public bodies sit on a recoverable book they cannot work. A municipal corporation we assessed carried property-tax arrears of roughly ₹340 crore against a recovery cell of eleven staff. The constraint was not willingness to collect. It was capacity: reconciling a demand register maintained partly on paper, tracing owners across name changes and inheritance, and issuing notices that survive legal challenge. Manual follow-up reached perhaps two hundred accounts a month.
The standard remedy — procure software — stalls at the tender stage. A department cannot justify a large upfront licence against an uncertain recovery outcome, and the budget head for it often does not exist. Tender cycles add their own drag: a requirement drafted in April may award in December, by which time the arrears book has aged another year. L1 pricing then selects the vendor most willing to underquote, not the one most likely to collect.
Fees tied to verified recovery
The success-aligned alternative inverts the cash flow. The vendor builds, deploys, and operates the recovery system at its own cost, and is paid a fixed percentage of dues actually collected — typically above a historical baseline, so the fee applies only to money the department would not otherwise have seen. No recovery, no invoice. This changes vendor behaviour immediately. Deployment milestones stop mattering; collection does. A feature that does not move recovery does not get built.
The contract lives or dies on the definition of recovered. A workable clause is narrow: a payment credited to the designated treasury head, matched to a demand notice the system issued, within ninety days of that notice. Anything looser invites dispute. Consider the defaulter who pays at a ward office counter a week after a system-generated SMS. The contract must state in advance whether that payment counts, because the accounts officer certifying the fee will ask, and the answer has to hold up in audit.
The machinery behind the fee
Work starts with the ledger, not the defaulter. Arrears registers are usually wrong: duplicate demands, payments posted against the wrong property ID, account holders long deceased. In one water-charges register we reconciled, close to a fifth of listed defaulters had already paid through a bank channel the register never captured. Chasing them would have produced grievances, not revenue. Reconciliation against bank statements and treasury records comes first, and it is the least glamorous, highest-value step.
The follow-up loop is then mechanical and measured. A flagged account gets a notice within forty-eight hours, SMS and voice contact in the local language, and escalation to a named revenue officer after two failed attempts. The hand-off line is firm: the agent drafts and schedules; the officer signs and decides. Recovery is a statutory function, and the signature on a demand notice belongs to a public servant, not to a model.
Everything is logged because everything will be questioned. Each notice, call attempt, and payment match is written to an append-only record, exportable for departmental audit or a CAG query without a file requisition. When a resident disputes a demand, the full case history — who issued what, when, on what basis — is retrievable in under a minute. That trail is what allows an accounts officer to certify a success fee and defend it later.
Ownership, uptime, and exit
None of this works if the department ends the contract owning nothing. The data, the reconciled ledger, and the case histories sit in a database the department owns, hosted on NIC cloud or on its own premises, with vendor access that can be revoked in an afternoon. Source code goes into escrow or is handed over at term end. A success-fee arrangement that quietly becomes vendor lock-in has failed, whatever it collected.
Operations need the same discipline as pricing: a 99.5 per cent uptime commitment with penalty clauses attached, a monthly reconciliation certificate signed by both vendor and accounts officer, and a settlement calendar that pays the fee from the certificate rather than from a dashboard. In our public-sector work at XRISE, that signed certificate is the document that moves money; the dashboard is a convenience. The distinction sounds small and is not.
The economics, stated plainly: the department carries no upfront cost, the vendor is paid from money that would not otherwise have arrived, and every rupee of fee traces to a receipt an auditor can inspect. This is not a discount structure. It is a reallocation of risk to the party best placed to manage it, and it holds only where measurement is honest, hand-offs are defined, and the exit is written before the entry.
