How lenders can decode creditworthiness and fraud through device analytics

By Michele Tucci, Chief Strategy Officer & MD Americas in credolab

Our phones and computers are literal treasure troves of data, much of it highly personal in nature and very complicated to work with. But there is also a ton of data stored on these devices – the majority in fact – that is completely impersonal and anonymous. What can we learn from the way a person moves the computer mouse when applying, or how long ago the operating system on the applicant’s smartphone was updated? You may be surprised to know that we can learn a lot from this impersonal data.

As the digital landscape continues to evolve, the lending industry is grappling with significant challenges, such as data scarcity for assessing potential borrowers’ creditworthiness and the growing threat of fraud. This landscape necessitates the integration of new technologies to tackle these looming risks. For example, behavioural data harvested via AI and Machine Learning (ML)-driven technologies during customer interactions on web pages or mobile applications. These interactions offer a rich reserve of information for detailed analysis, including user device features, typing patterns, and session duration. Such a valuable data trove empowers institutions to bolster fraud detection mechanisms and reduce risk-associated costs. A critical element of this method is that the data is completely anonymized, ensuring no personal information is compromised.

Having amassed a comprehensive picture based on over 200 million digital footprints, I would like to share some of the most common behavioural patterns and what they can tell digital lenders about a borrower’s creditworthiness or potential fraud attempts.

What our devices can tell lenders about us

Let’s delve into what a smartphone can reveal about a user’s creditworthiness without relying on any personal data. Firstly, it’s evident that users of premium flagship devices from renowned brands are more likely to possess a stable financial profile and exhibit lower credit risk. This observation is quite straightforward. But what other valuable insights can lenders glean from metadata? A lot!

• Operating System: Keeping your phone’s operating system up-to-date indicates responsible device management, hinting at potential financial stability.
• Security Features: Adopting biometric authentication (e.g., fingerprint or facial recognition) and robust encryption tools suggest a priority for personal and financial data security, thereby painting a positive picture for lenders.
• Browser Preferences: A user’s chosen browser, its version, and installed plugins can provide critical insights into potential fraud. Uncommon or outdated browsers or manipulated browser settings can be red flags for fraudulent behaviour.
• Screen and Hardware Configuration: Screen size, resolution, colour depth combinations, language settings, and hardware configurations like CPU or GPU type and available memory can point to known fraud schemes or attempts to mask fraudulent activities.
• Apps installation patterns: Frequent downloads of productivity, professional networking, or educational apps implies a proactive approach towards personal and professional development, correlating with lower credit risk. A high frequency of gambling, gaming, or other addictive behaviour-related app installations may raise concerns about financial stability and credit risk.

Financial institutions can also detect potentially fraudulent activities by monitoring frequent app installations or deletions. For instance, quick installation and deletion of multiple loan or banking-related apps can hint at fraudulent attempts to use different identities.

Digital lenders often harness the benefits of verifying a user’s location using their IP address. So, a sudden shift in an IP address to a remote or high-risk region can activate additional security measures, including extra verification or account suspension. IP addresses associated with proxy or VPN services used to disguise identity can signal suspicious activity as well as an examination of the frequency and volume of loan applications from a specific IP address.

Even the mouse makes a difference

The exploration of users’ behaviours, such as keystroke patterns, mouse movements, and session duration, can serve as powerful tools to identify potential fraud attempts in loan applications. These elements can provide telling insights:

• Keystroke Dynamics Analysis. Typing patterns that deviate from usual user-keyboard interactions can hint at the use of automated scripts or bots. Clues may include unvarying keystroke duration, typing speed far from the average, or lack of typing rhythm variation. Uncommon “flight time” patterns, the time between releasing one key and pressing the next, may suggest data copying from external sources or scripted inputs. Frequent use of the backspace and delete keys could point to user uncertainty or attempts to hide mistakes or manipulate information.
• Mouse Movement Analysis. As human mouse movements generally follow curved or slightly jagged paths with variable speeds, consistently straight, sharp-angled paths at constant speeds could suggest script, bot, or remote access tool usage. Lack of idle or hovering time, or a high frequency of incorrect or outside-boundary taps, may imply a script or fraudulent sequence of actions.
• Session Duration Analysis. The time a user spends on each page or section of a loan application can reveal potential fraud. Rapid loan application completion might point to pre-filled forms, automated tools, or familiarity due to previous fraudulent attempts. Unusually long durations could indicate unfamiliarity with the data, cross-referencing from other sources, or seeking external help – all potential fraud signs. Users who linger on specific pages while quickly concluding others might be attempting to manipulate or fabricate certain data.
• Clipboard Usage Analysis. Frequently copying and pasting data into application fields could hint at using pre-compiled or stolen information. Users typically avoid copying sensitive data like national insurance numbers or bank details from external sources. Such behaviour might indicate fraud.
• Field Correction Analysis. Tracking modifications made by users in the loan application can expose potentially fraudulent activity. Frequent adjustments in name, date of birth, address fields, or sensitive financial information might imply manipulation or use of fraudulent data. A consistent pattern in corrections, such as alternating modifications or following a particular order, can suggest adherence to a pre-made script or a pre-fabricated set of false data. These analytical techniques help reveal red flags in loan applications, aiding institutions in their fight against fraud and ensuring the integrity of the lending process.

Spotting anomalies: Outliers, deviations and their business implications

The behavioural patterns discussed above do not encompass the entirety of valuable insights that can be derived from anonymized data. The deployment of such data is becoming increasingly critical for a range of entities, including credit bureaus, traditional banks and neobanks, Buy Now Pay Later (BNPL) services, digital lenders, ride-hailing and Earned Wage Access (EWA) apps, insurance companies, and crypto lenders.

With the advent of digital platforms such as Amazon, Grab, and other super apps, the lending function has evolved beyond its traditional confines within banks. These innovative platforms are revolutionising the financial landscape by integrating financial services with other industries like e-commerce, ride-hailing, and social media. In the face of this transformation, traditional financial institutions must adapt and innovate or risk obsolescence. Evidently, the future of lending will be significantly more integrated and reliant on data analysis than ever before.

In this evolving landscape, the ability to identify outliers and deviations holds paramount importance in lenders’ risk management strategies. A thorough examination of mobile metadata across loan applications can unveil crucial insights into potentially fraudulent activities. However, it is equally vital to champion a more responsible usage of behavioural data. Not all deviations necessarily indicate fraudulent activity or an inability to repay a loan. Certain users may exhibit unusual behavioural patterns due to valid, specific circumstances, underlining the necessity for a holistic, multi-layered approach to fraud detection.

By combining metadata analysis with diverse fraud detection techniques, lenders can diminish the occurrence of false positives, providing a more refined, accurate comprehension of user behaviour and bolstering the efficacy of risk management strategies. Essentially, identifying and understanding anomalies is not merely about pinpointing potential threats. It also enriches our broader understanding of consumer behaviour and credit risk. This comprehensive perspective is key to nurturing a more secure, reliable, and understanding financial ecosystem, fostering trust and precision in an increasingly interconnected world.

About the author

Michele Tucci is the Chief Strategy Officer and Managing Director in North and Latin America in credolab, one of the largest providers of bank-grade digital scorecards and data enrichment solutions. Prior to joining credolab in 2018 as Chief Product and Marketing Officer, Michele spent over 20 years working on international consulting assignments, product management and business development roles with Capital One, MasterCard, Intesa Sanpaolo Bank, and Telecom Italia Mobile.

DISCLAIMER: Biometric Update’s Industry Insights are submitted content. The views expressed in this post are that of the author, and don’t necessarily reflect the views of Biometric Update.

Article Topics

behavioral biometrics  |  biometrics  |  Credolab  |  financial services  |  fraud prevention  |  smartphones