Advanced analytics and machine learning help Poste Italiane identify and stop fraud in real time while enhancing customer experience

Italy’s largest service distribution network relies on predictive analytics from SAS to detect fraud with greater precision and reduce losses.

When you have 35 million customers and €586 billion in total financial assets, it’s imperative to have an end-to-end fraud detection and prevention framework in place. That’s why Poste Italiane partnered with SAS to guard itself and its customers from fraud, abuse and improper payments.

Poste Italiane is Italy’s largest service distribution network, offering logistics, letter and parcel delivery, financial services, insurance and telecommunications services. Customers can mail a package, make online payments, open a savings account, take out an insurance policy, or top off mobile phone minutes. In fact, 11 million customers interact with Poste Italiane every day. The 160-year-old organization plays an essential role in the country, helping bring a sustainable, digital-forward approach to everyday activities. Its impact on Italy’s gross domestic product is estimated at €12.5 billion.

We spoke with Raffaele Panico, Head of Fraud Management and Security Intelligence at Poste Italiane, to learn how the company uses advanced analytics and artificial intelligence (AI) technologies to detect fraud with greater precision and reduce losses.

Poste Italiane is a complex entity with an impressive amount of data. How do you manage this wealth of information to prevent and manage fraud?

Panico: Data is vital to offer the diverse services that are part of our group. It’s not just raw data. We need information to leverage appropriate analysis models.

Information is necessary to define the right and appropriate business proposition. It’s like Netflix. We want to propose the service or product that is most in line with customers’ needs. We can do this from studying behavior patterns.

This may seem far removed from the topic of anti-fraud, but it is not. Even in fraud management and prevention, data analysis, particularly modeling behaviors and habits, becomes the knowledge base for intercepting and preventing as well as stopping fraudulent actions.

How can fraudulent behavior be detected and prevented today? What are the enabling technologies?

Panico: First, it is necessary to have a holistic view of customers and their interactions with your enterprise. Through a single view and an advanced system to analyze large masses of data from multiple sources and channels, you can identify and understand customers’ usual behaviors, then bring out anomalous deviations that could be indicative of fraud.

You need big data, advanced analytics, AI and machine learning, as well as substantial computing power to achieve predictive analytics. These are all ingredients that were not there just a few years ago, other than in the theoretical or experimental level.

Detecting anomalies in customer behaviors and associating or not associating those anomalies with fraud requires great processing and analytical capacity on huge amounts of data. In the past, analysis was based solely on pre-established deterministic rules that did not allow intercepting and responding to changes in a timely manner. Behaviors and deviations had to be defined before analyzing the data; that is, analysts had to detail the anomalies to be searched for prior to actually searching for them. This gave rise to many false positives.

Thanks to machine learning and AI, we can observe and analyze customer behavior. It’s important that the data about each product or service that the customer uses be merged, as opposed to being kept in silos. This leads to accurate analysis and forecasting.

How important is it to have a real-time approach in detecting fraudulent behavior?

Panico: Without a real-time approach, it would not even make sense to talk about fraud prevention. Let's think of a person who wants to perform an online transaction by logging in from their smartphone. Real-time analysis allows me to understand whether that transaction is related to habitual behavior or similar to other behavior that has occurred in the past, or whether it is unusual behavior that needs to be verified. All of this has to happen in real time.

With real-time analytics, it is possible to automatically block the transaction, even if only temporarily, to make the verifications. This might take only a few seconds, with a text message check and a quick interaction with the customer to verify that that transaction is wanted.

This is only possible if there is a unique, centralized layer of information for individual customers, a single analytical environment, a platform like SAS that allows you to look at the customer and what they do, not only from the point of view of their activity on our channels but also in their daily life and habits when they use our services.

These days, it’s not just fraud management – you must think in terms of fraud intelligence because the key lies in the ability to make predictions thanks to real-time analytics. It’s about predicting customer behavior and identifying those who deviate from habits. Machine learning techniques provide that layer of intelligence that allows us to identify risk situations, analyze them very quickly and intervene when necessary. We have reduced false positives by 40% and increased our ability to handle anomalies by more than 20%.

Not only do we see enormous benefits for business but also for our customers in terms of prevention and security as well as a positive customer experience. Protected customers are satisfied, so what better way to retain them?

What are the next challenges? How will the fraud management and prevention system at Poste Italiane evolve?

Panico: We started in banking and then extended it to e-money, where we have achieved amazing results. While we saw a continuous increase in transactions, we recorded a decrease in fraud. On the e-money side, in the last three months, the fraud ratio has dropped by 50%. This is an astonishing figure, especially when we consider that in the last two years fraud has increased by 90% worldwide.

We are further extending the system to the insurance sector. The telephony sector remains outside, as it integrates its own native anti-fraud system, but the goal going forward is to shape, even in fraud prevention, a true real-time omnichannel approach, thus extending it to postal services and delivery centers where the fraud risk is lower. In each case we have managed to achieve an 80% reduction in fraud and risk by locking down a number of services and products.

The real challenge – and future goal – is to take this system outside; that is, to offer it to the market. We would be, in this case, the first customers to have tested it and be able to bring the best references.