Mastercard
real-time transaction risk scoring
In 2024, Mastercard launched Decision Intelligence Pro, a generative model that scores each transaction in less than 50 ms; its internal modeling reports +20% fraud detection on average (up to +300%) and more than 85% fewer false positives.
Key points
- Real-time fraud risk scoring on every transaction on the network.
- Decision Intelligence Pro, transformer-type generative model, score refined in under 50 ms.
- +20% detection on average (up to +300%), false positives reduced by more than 85%.
- Evidence B, confirmed status: Mastercard internal modeling, picked up by CNBC and PYMNTS.
Objective
Help issuing banks distinguish legitimate transactions from fraudulent ones in real time, to block more fraud while reducing false declines that degrade experience and lose revenue.
The deployment
Decision Intelligence assigns a risk score to each transaction on the network. In February 2024, Mastercard launched Decision Intelligence Pro, a version that relies on a generative model to assess the relationships between the entities around a transaction (merchant, terminal, behavior). The model scans a trillion data points to predict whether a transaction is likely authentic, and refines the Decision Intelligence score in less than 50 milliseconds. According to Mastercard's initial modeling, the improvement raises the fraud detection rate by 20% on average, up to 300% in some cases, and reduces false positives by more than 85%.
Results Proof B
Figures from Mastercard's official press release (internal analysis and modeling, interested source) and picked up by name by CNBC and PYMNTS. These are initial modeling results, not audited financial KPIs; the +20% and +300% are presented by Mastercard as projections, which the record preserves.
How it works
Documented architectureThe stack in detail
- plateforme Mastercard Decision Intelligence real-time risk scoring on every transaction on the network, passed to issuing banks at authorization
- llm Decision Intelligence Pro transformer-type generative model (launched in February 2024) that assesses the relationships between entities (merchant, terminal, behavior) and refines the score in less than 50 ms
- infra Donnees du reseau Mastercard a trillion data points scanned; the confirmed-fraud labels serve for retraining
- infra Moteur de decision de la banque emettrice consumes the score in the authorization rules; the final decision stays with the issuer, supervisable and contestable (GDPR, AI Act)
How it runs, concretely
For ops teams-
1Transaction pass-through site_app / network
A transaction arrives for authorization on the Mastercard network.
-
2Scoring AI
The model assesses the relationships between entities and scans the context to assign a risk score in less than 50 ms.
-
3Enrichment of the DI score AI
Decision Intelligence Pro refines the overall score passed to the bank.
-
4Decision issuing bank
The issuing bank approves or declines the transaction based on the score and its own rules.
Transaction behavior and the relationships between entities (merchant, terminal, history). If the confirmed-fraud label signal degrades, the model drifts and accuracy drops.
How your customers perceive this type of use
Sourced studiesC'est la famille la moins acceptee : 68% des Americains jugent inacceptable un score financier personnel calcule par algorithme et 67% l'analyse video automatisee d'entretiens d'embauche (Pew Research, 2018). La demande d'explication et de recours est massive : 83% veulent savoir quelles donnees l'IA utilise et 91% veulent pouvoir corriger des donnees erronees (Consumer Reports, 2024). A l'echelle mondiale, seuls 46% se disent prets a faire confiance aux systemes d'IA et 70% jugent une regulation necessaire (KPMG / Universite de Melbourne, 2025).
Acceptance conditions
- Transparence sur les donnees utilisees : 83% des Americains la reclament (Consumer Reports 2024)
- Droit de correction des donnees erronees : 91% le demandent (Consumer Reports 2024)
- Explication de la logique de decision : 44% des consommateurs sont plus enclins a utiliser un agent IA si sa logique est clairement expliquee (Salesforce 2024)
- L'acceptabilite depend du contexte de la decision : 50% des Americains jugent equitable un score de risque criminel pour la liberation conditionnelle, contre 32% pour un score financier applique aux consommateurs (Pew Research 2018)
Red lines
- La decision opaque et sans recours sur l'emploi, le credit ou le logement : 45% tres mal a l'aise pour l'embauche, 39% pour le pret, 39% pour le logement (Consumer Reports 2024)
- Le scoring des personnes a partir de donnees comportementales : 68% le jugent inacceptable pour les offres financieres (Pew Research 2018)
Sources: Pew Research Center 2018 · Consumer Reports 2024 · KPMG / Universite de Melbourne 2025 · Salesforce 2024
How to replicate
Inference, not sourcedData prerequisites
- Real-time transaction feed
- History of confirmed fraud for learning
- Relational data between entities (merchant, terminal, device)
Org prerequisites
- Ability to integrate an external score into the authorization decision engine
- Risk team to calibrate the thresholds
Possible stack
- Mastercard Decision Intelligence (network activation)
- third-party anti-fraud scoring engines (Feedzai, Featurespace)
- custom model if internal volume and data are sufficient
The plan, step by step
- Step 1Frame with the network the activation of the score on the relevant card portfolioDeliverable: Activation scope + access to the score in the authorization flow
- Step 2Plug the score into the decision engine, first in observation mode without blockingDeliverable: Score consumed in production, logged on real traffic
- Step 3Calibrate the thresholds by segment with the risk team, on history and observed trafficDeliverable: Tested threshold matrix, trade-off between fraud detected and false declines
- Step 4Move to decision mode on a pilot segment and compare against a control periodDeliverable: Measurement of detection rate, false positives and approval rate
- Step 5Generalize to the portfolio and close the retraining loop with confirmed-fraud labelsDeliverable: Detection, false-positive and approval dashboard tracked continuously
First step: For an issuer: plug the existing network score into the authorization rules and measure the effect on false declines and fraud.
The same pattern, proven in other industries
Sources
- S1 Mastercard Says New AI Model Ups Fraud Detection by 20% Established press archive pending
- S2 Mastercard supercharges consumer protection with gen AI Primary archive pending
- S3 Mastercard jumps into generative AI race with model it says can boost fraud detection by up to 300% Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.