Every card transaction flows through the payment network before it reaches the issuer. The authorisation request arrives at the network, which routes it to the issuing bank for a decision. In that routing step, the network has a window — measured in milliseconds — to add its own assessment of the transaction’s risk before the issuer makes the authorisation call.
That assessment is the network fraud score. It is not a decision. The network does not approve or decline transactions. It provides a signal that the issuer incorporates into their own authorisation logic alongside the signals they can generate from their own data. The issuer knows the cardholder’s account history, credit limit, KYC profile, and behavioural baseline. The network knows none of that. What the network knows is everything that has happened across the scheme involving this card, this merchant, this acquiring bank, and this pattern of activity — across all issuers simultaneously.
Those two views are complementary. The issuer’s model is deep on the individual cardholder. The network’s model is broad across the entire participant ecosystem. The combination produces better precision than either alone.
What the network sees that no issuer can
The network’s fraud signal draws on data that is structurally inaccessible to any individual issuer. When the same card has been used at three different merchants in two different countries in the previous hour, the issuer of that card sees one of those transactions when they receive the authorisation request. The network sees all three. The velocity signal that should inform the authorisation decision is fully visible only at network level.
When a new fraud typology begins appearing — a specific pattern of merchant category, transaction amount, and geography that has been generating disputes at multiple issuers over the previous 48 hours — no individual issuer has enough data points to recognise the pattern. The network, seeing the same pattern emerging across dozens of issuers simultaneously, identifies the typology in hours rather than weeks and incorporates it into its scoring before most individual issuers have accumulated enough cases to detect it in their own data.
When a specific acquiring bank’s portfolio begins showing elevated fraud rates across multiple card schemes simultaneously — suggesting a compromised merchant or a facilitated fraud operation — no individual issuer can see the cross-scheme signal. The network sees it as a participant-level pattern and adjusts the risk weighting of transactions through that acquirer accordingly.
These cross-participant signals are the network’s unique contribution to the authorisation decision. They are not available from any other source. The issuer cannot acquire them by improving their own data infrastructure. They exist only because the network sits at the intersection of every transaction across every participant.
The precision argument
The false decline rate across major card networks averages between 1 and 2 percent of total authorised volume. At an average transaction value of fifty dollars, a 1 percent false decline rate on 10 billion annual transactions represents $5 billion in legitimate sales that did not complete. That cost flows to merchants through abandoned purchases, to issuers through cardholder friction and attrition, and to the network through lost transaction economics. It exceeds the value of the fraud it was intended to prevent.
A better network fraud score reduces both fraud and false declines simultaneously by improving precision: more genuine fraud above the score threshold, more legitimate transactions below it. The improvement does not require the issuer to rebuild their authorisation logic. It requires the network score they receive to be more accurate than the one they are currently using. The incremental precision gain from a better network signal translates directly into authorisation quality improvement at the issuer, without the issuer needing to see the signals that produced it.
The commercial case for network fraud scoring as a value-added service rests on that precision improvement. An issuer that demonstrates lower fraud loss rates and lower false decline rates after adopting a better network score has a measurable, attributable improvement in their portfolio economics. That attribution is the evidence base for the service relationship.
The architecture of the network score
The network fraud score is built from transaction characteristics that are available within the authorisation interchange: the card BIN, the merchant identifier and category, the transaction amount and currency, the geography of the merchant, the time of the transaction, and the recent transaction history of the card across the network. Customer account details and KYC data are not available to the network and are not inputs to the score. The score is a function of observable transaction behaviour, not of customer identity or financial profile.
The model architecture that produces the best network-level fraud signal uses these signals in combination: individual card velocity within defined time windows, merchant-level fraud concentration signals, cross-participant emerging typology detection, and acquiring bank portfolio quality indicators. Each of these is a network-level observation that complements rather than duplicates the issuer’s own behavioural baseline model.
The score is returned to the issuer as a numeric value within the authorisation response. The issuer decides how to incorporate it into their authorisation logic — as a direct input, as a trigger for step-up authentication, or as a threshold for additional verification. The network provides the intelligence. The issuer makes the call.
What success looks like
The metrics are fraud detection rate improvement attributable to the network score, false decline rate reduction attributable to the network score, and issuer adoption rate for the enhanced scoring service. Measuring attribution requires a controlled deployment methodology — issuers using the enhanced score against a baseline that uses the standard routing signal — with the improvement in both fraud and false decline rates tracked separately. The combined precision improvement, expressed in dollar terms against the issuer’s transaction volume, is the commercial value of the service. That number is the conversation the network has with issuers about adoption, and it is the metric that drives renewal.
