Reward integrity

# Pay what's owed. Hold what doesn't add up.

Reward leakage is money paid that wasn't owed, and money owed that wasn't paid. Ankasum checks for it at three points, inside your own environment, with deterministic checks you can reproduce and explain to an examiner.

[At design](#design) [Before payment](#before) [After payment](#after)

Daily run / **Held payout**Held

Member …4471

Spring checking bonus, version 3

Due May 9

Rule engine

$400.00

Independent check

$300.00

The two disagree by $100.00Treated as a possible data or software defect, not as abuse. Paid when both agree or a named reviewer releases it.

SOSam OkaforOperations, reviewerAssigned

Illustrative, sample data.

1

## At design: know the worst case.

Ankasum calculates each offer's worst-case payout per participant and in total. Every offer needs a hard bound: a budget pool, a participant cap or, for an offer to existing customers, your own customer count. If a required limit is missing, approval is blocked.

Named loophole checks run before the offer can be approved, such as deposit, qualify and withdraw the same day, or unbounded stacking with other offers.

Summer savings bonus / **Exposure**Blocked

Worst case per member$600

Worst case in totalNo bound

- No per-member limit or budgetAdd a budget pool or participant cap to continue.
- Same-day deposit and withdrawal excluded
- Stacks with Spring checking bonusSet a stacking group or mark it exclusive.

Illustrative, sample data.

2

## Before payment: every amount computed twice.

Every payout is recomputed by a second, independent interpreter that reads the approved offer definition, not the engine's compiled form. The approval readback names any construct the second interpreter doesn't cover, and those payouts are flagged rather than trusted.

Amounts must agree to the cent. A mismatch is treated as a possible defect in our software or the data mapping, not as abuse: the payout is held, both sides see why, and it's paid when the two agree or a named person at your institution releases it.

Daily run / **Independent check**

Checked

414

Agree to the cent

412

Held

2

Held payouts go to a named reviewerReview holds

Illustrative, sample data.

3

## After payment: paid against owed.

A leakage ledger compares what was paid with what was owed for every offer. Per-offer reports show accounts drained or closed soon after the bonus, so you can tighten the next offer's terms.

These reports never affect a payout.

Spring checking bonus / **Leakage ledger**

Owed to date$1,086,300

Paid to date$1,086,300

Difference$0

| After the bonus                   | Accounts |
|-----------------------------------|----------|
| Closed within 30 days             | 41       |
| Balance under $25 within 30 days | 96       |
| Clawback recovered                | 38       |

Illustrative, sample data.

## People at your institution decide.

Ankasum releases or holds. It never refuses a reward on its own.

- **Ankasum never rejects a payout.** Automated outcomes are release or a time-limited hold, and both follow a policy your institution approved.
- **A refusal needs a named person.** A rejection, forfeiture or clawback happens only when an authenticated reviewer at your institution chooses it and cites the disclosed clause it relies on.
- **A score is never grounds to refuse.** A pattern alone never refuses a reward, and behavior the offer's terms allow never holds a payout.
- **Holds don't expire into rejections.** Every hold has a due date tied to the offer's disclosed pay-by date, and escalates to a named approver.

- ## Fair by design

  Integrity checks never use protected characteristics or their listed proxies. Signals such as a shared address or phone number can only send a payout to human review; they never hold or reject it on their own. Household grouping uses your own household IDs by default.

- ## Where AI fits

  Language models never decide a payout, and no model runs in the reward-integrity checks. Eligibility, amounts and holds are decided by deterministic rules you can reproduce line by line.

## What reward integrity is not.

Clear limits, so you know what to keep running alongside it.

- ### Not an AML system

  It looks for reward abuse, not money laundering. Confirmed cases export with their evidence for your BSA/AML or FINTRAC team; any reporting decision is yours.

- ### Not cross-institution

  Ankasum sees behavior only within one institution. It can't see a customer collecting the same bonus at several institutions, and patterns are never shared between institutions.

- ### Not a guarantee

  No system detects all abuse. What Ankasum does is evaluate, limit, check, recompute and write payout files that conform to the offer definition your institution approved. Exposure and leakage figures are estimates.

## See one of your own offers in Ankasum.

Bring a bonus you run today. In a 30-minute demo we build it in Ankasum and show you the readback, the checks and the payout file it produces.

[Book a demo](https://ankasum.com/contact/form/) [Other ways to reach us](https://ankasum.com/contact/)

- No customer data needed for the demo
- We can sign your NDA first
- Bring marketing, compliance and operations together

---

Source: https://ankasum.com/reward-integrity/  
Updated: 2026-11-12