Staying ahead of fraud can feel like you're waging a battle under constantly changing conditions. The speed with which fraudsters evolve their tools and techniques can quickly make static anti-fraud measures less effective at protecting your business and your customers.
That's why our fraud and risk management solutions, Decision Manager and Fraud Management Essentials, use artificial intelligence (AI) and machine learning (ML) to automate fraud screening. Powered by Visa's AI platform and access to 141 billion global VisaNet transactions1, our ML model generates a highly accurate risk score for every transaction in less than a second—helping detect and prevent fraud faster than a human ever could.
Positive + negative behavior models—a unique combination
We've made our ML model even better by integrating Identity Behavior Analysis (IBA), which is built on a positive behavior model. Previously, the risk score was based on fraudulent activity (negative behavior) alone.
IBA is designed to recognize:
- Good repeat customers
- Genuine new customers who have no history with you
By supplementing our ML model's fraud recognition capability with training to identify good transactions, IBA enables even more accurate risk scores. This combination of positive and negative behavior models makes our ML model unique in the market and adds a new dimension of trust to our solutions.
What this means for your business
With IBA incorporated into our ML model, Decision Manager and Fraud Management Essentials users may notice increased risk score accuracy. We know the hardest transactions to make a decision about are those with mid-range scores. Going forward, you should find that more transactions receive risk scores at:
- The bottom end (0–9)—indicating low risk of fraud (the case for the majority of transactions)
- The top end (95–99)—indicating high risk of fraud
Risk teams should be able to make quicker and easier decisions about order acceptance based on the risk score, and you'll be able to increase your reliance on automated screening and more confidently accept good (low-risk) orders. You should also see a reduction in false positives, meaning fewer good customers will be impacted by manual reviews or other risk controls that could negatively affect their experience.
IBA also combines reinforced good patterns from repeat customers with anomaly-detecting features to better understand if:
- A bad actor is trying to pass themselves off as a good customer
- A good customer is doing something different during a transaction (such as purchasing from a new device or unfamiliar location)
Trust the risk score
With IBA incorporated into our ML model, you can trust the risk score more than ever and confidently base your fraud strategy off it. The model will learn something new with every transaction screened, so it will inevitably be more accurate than complex rules that can't be updated as quickly.
Disclaimer: Results are not claimed to represent typical results, experience, and are not intended to represent or guarantee that everyone will achieve the same or similar results.
1 VisaNet transaction volume based on 2022 fiscal year. Domestically routed transactions may not hit VisaNet.