ChequeMate


Deterring Fraud, not Clients

Introduction

In attempting to facilitate convenient payment methods for their clients
many businesses are faced with continued exposure to fraudulent cheque
transactions. Still, the extent of damage caused by rejecting valid cheques
is difficult to estimate: the immediate loss of revenue is
probably less significant than the offence taken by the customer.
Using optimization techniques based on past transactions and currently
supplied data ChequeMate facilitates a positive payment experience for
the client yet reduces the retailer's exposure to fraud. Unlike Procrustes,
ChequeMate won't shoehorn your business to match design limitations.

Configuration

ChequeMate is initially configured using a flexible rule-based system
with all conditions carrying tunable weightings.
A typical rule might state: If the Account number was supplied and it
passed the Check Data Verification test confirming that it is a
legitimate account number, then add a positive value towards immediate
cheque acceptance.
Another rule might state: if this is the third time an account number is
being used within a day, then add a negative value towards cheque
acceptance, perhaps resulting in a POS instruction to seek supervisor
intervention, or to decline the cheque outright.
Once transaction history grows, ChequeMate also makes use of neural
network technology to further optimize decision-making.

Features

Intelligent Credit Limit
The transaction history and information supplied by the POS allows
ChequeMate to determine a credit limit for each transaction.
For instance, should a particular account provide a positive payment
transaction history then, based on the value of those transactions
and a tuning factor, the credit limit is determined. This results in
customers being treated with individual "recognition" and rewards those
with positive records by pre-empting a routine, time-consuming and
sometimes embarrassing supervisor call to "clear" cheques above the
fixed floor limit.

ChequeMate Suggested Actions
Predicting cheque fraud will never be an exact science. For this reason
appropriate recommended actions are not always black and white
(accept or decline), so ChequeMate returns a value within a tunable
range of suggested actions to the POS. This value may result in a
suggested action from "Approve" to "Call Supervisor" to
"Contact Call-Center" to "Decline" etc.

Heuristic Engine Room
In keeping with this flexibility, each criterion used by ChequeMate is
assigned a tunable weighting to ensure a response appropriate to the
organization's culture and customer service ethos. As there are many
criteria, with the option of easily adding others, determining
appropriate values requires careful attention. To facilitate this task
ChequeMate provides a GUI interface with sliders representing each
criterion next to a sample of bar charts representing typical transaction
profiles (VIP, High Risk etc). As the sliders are adjusted the implied
results are immediately apparent as the suggested action shifts between
categories.
Using past transaction data the H.E.R. can be used as a modelling tool,
suggesting optimal slider values to minimize the acceptance of
potentially bad cheques. An additional slider is used to tune between
erring on the side of accepting a few too many bad cheques
(being more "customer forgiving") or to minimize immediate financial
risk by rejecting a larger number of "suspicious" cheques.

Artificial Neural Networks
Although the rule based criteria system does a good job of identifying a
significant class of fraudulent transactions it may flag an
unacceptably large number of false alarms. As a final option to calculate
the optimal recommended action the H.E.R. uses Neural Network technology.
Neural Networks are especially useful for problems which are tolerant of
some imprecision and which have lots of training data available,
but to which hard and fast rules cannot always be applied.
As the transaction history grows the Neural Network option becomes
more feasible and, indeed, powerful.
In its simplest terms, an artificial neural network is a model of
interconnected nodes inspired by the densely interconnected
structure of the brain. Each connection between nodes is assigned a
weight that adjusts as the model learns. These connection weights
store the knowledge necessary to solve problems. The distinction
between a neural network and regular computer models is the
ability to perceive new patterns and learn from data fed to the model.

Hierarchical System Architecture
ChequeMate works through a configurable hierarchy of servers that can be
tailored to an existing IT infrastructure. Each node within the
hierarchy can operate independently providing a resilient solution in
the event of network instability. Each intermediate node
(NT or Unix based) runs a server process requiring minimum system
resources. ChequeMate utilises industry standard protocols TCP/IP and XML.

The Retailer's Dilemma
IT solutions to business problems frequently involve a compromise:
customer friendliness is sacrificed for safety guarantees.
ChequeMate maximises fraud detection yet provides the business with a
mechanism to enhance and build up customer trust.