AnantState
Industry · Retail and commerce

A hundred stores is a hundred live situations.

One compromised till is a local problem. The same pattern across a dozen stores points to organized fraud, and you cannot see the second if you look at stores one at a time.

In plain termsSpot patterns that only appear across stores, and price the response.

How one pattern spreads across stores

An illustration of the first four days, compared with how it is usually found.

Retail entities are independent in space and coupled in money. A store, a lane, a till, an online checkout and a member account each carry their own state, and a shared driver across them is the signal that no single store’s report contains.

One payment pattern across stores · illustrative

  1. Day 1: One till declines more cardsLooks local, and under every rule's limit
  2. Day 2: Three stores, one tender typeA shared driver shows in store state
  3. Day 3: Flagged as one patternRanked by money at stake
  4. Day 4: Action priced and appliedCost to act beside cost to wait
  5. Week 3: Finance finds itThe usual way: at reconciliation, after the money has moved
  6. Phase: Visible in store statefrom One till declines more cards to Flagged as one pattern
  7. Events are shown in order, not to scale.
A pattern that looks local on the first day is visible as a shared driver across stores within days, long before a reconciliation cycle would find it.

Why it is found late

  • It surfaces in reconciliation. Losses appear in a finance cycle weeks later, when the tender pattern has moved on and the money is gone.
  • Rules fire per transaction. Thousands of alerts a day bury the pattern across stores, because nobody is looking at state across a region.
  • The response cannot be priced. Nobody can say whether blocking a tender type recovered more than it cost in declined legitimate sales, so the policy stays as it is.

What moves in a store

StateWhat moves it (baseline)What moves it unexpectedly (residual)
Payment integrityTransaction volume, trading hours, basket mix, seasonalityTender manipulation, card testing, chargeback patterns
Refund pressureReturns policy, season, product mixRefund abuse, staff collusion
Gift-card exposurePromotions, holiday cyclesDraining, cloning, bulk activation
AvailabilityReplenishment schedule, demandSupplier failure, mis-picked distribution
Store postureFootfall, staffing, day of weekLocalized fraud clusters, equipment failure

Retail state is dominated by schedule and season, so the baseline is most of it. The learned residual exists to catch the part that is not scheduled, which is the part that costs money.

What you would see

  • Stores and accounts ranked by value at stake, with band and trend
  • One entity’s trajectory against its events
  • An intervention’s expected effect and its cost, side by side
  • The what-if, recomputed in your own browser

State dimensions and intervention costs are domain configuration, so a grocery chain, a fashion retailer and a fuel network each define their own. Currency is stated on every cost surface.

The OgMart payments domain as a ranked list of entities, each with its value pressure, value at stake and value band, ordered by pressure
The queue: every entity ranked, with the money at stake beside it, so a team can work it top-down.

A worked example

A hundred stores, a fifth of them acted on, and the arithmetic laid out.

Round, illustrative figures for stores: they show the arithmetic, leave out the subscription, and are for you to change in the calculator.

InputValueWhat it stands for
Stores in scope100Everything the team is responsible for
Value at stake per one$12,000Fraud and refund exposure at one store over the horizon
Cost of the action per one$400Blocking a tender type and reviewing the store’s accounts
Share acted on20%The stores ranked highest by value at stake: 20 of 100
Modeled, USDPessimisticExpectedOptimistic
Share of value at stake avoided10%25%40%
Cost to act, in total$8,000$8,000$8,000
Avoided loss$24,000$60,000$96,000
Net position$16,000$52,000$88,000

Break-even is 3.3%: each action costs $400 against $12,000 at stake, so it has to avoid that share of the loss to pay for itself.

Counterfactual theater: two forward projections of one OgMart entity, one with no action and one with early action, with the dollars at stake in each
Counterfactual: the same entity run forward twice, once ignored and once acted on early.

What exists today

OgMart, a 100-store demo retailer across 15 departments, is the live reference world. Retail is the industry with a running demo.

Watch a hundred stores at once

See the queue, the evidence and the priced action on the demo, then on your own region.