RST Software
Editorial Team
Ross Krawczyk
Reviewed by a tech expert

What mobility startups can learn from Uber’s fraud detection and prevention technology

Read this articles in:

The largely faceless nature of the internet makes it rife with all sorts of con artists trying to take advantage of vulnerabilities in various systems. Oftentimes it’s not a full-frontal attack but rather a pretense of legitimate action underpinned by ill intentions designed to benefit the scammer. Fraud management appears to be indispensable for a lot of companies which become targets of such exploitation attempts if they want to not only stop but also prevent any possible bleeding in the future. In this article, we’ll talk about the mobility industry in particular and see what can be learned from one of its major players in Uber.

In the realm of mobility startups, the challenge of fraud management looms large, as there’s a constant threat of misuse of their digital systems. According to Melanie Ensign, Uber's security spokeswoman, “People are literally trying to defraud the platform every single day.” Safeguarding against scam is essential for companies to protect their operations and build trust among their user base.

By examining Uber’s approach to tackling fraud, we'll aim to uncover valuable insights and lessons that can benefit any emerging mobility startups but before we do, let's cover some essentials.

What is a fraud management system

In short, a fraud management system is a comprehensive set of tools, processes, and technologies designed to detect, prevent, and mitigate fraudulent activities that may occur within a company's operations. And occur they will, especially past a certain point of growth.

The primary goal of such a system is to safeguard the integrity and security of the organization's services, while also ensuring a seamless and trustworthy experience for customers. By implementing anti-fraud systems, companies can identify and respond to potentially deceitful activities in real-time, reducing financial losses, protecting sensitive data, and maintaining their reputation in the market.

These systems utilize advanced algorithms, machine learning techniques, and data analysis to detect patterns and anomalies across various touchpoints. By monitoring transactions, user activities, and other relevant data, fraud management systems can raise red flags pertaining to unusual payment patterns, suspicious user behavior, or unauthorized access attempts.

Importantly, fraud detection systems aim to strike a balance between prevention and minimization of false positives. While it is crucial to identify and stop fraudulent activities, it is equally important to avoid inconveniencing genuine customers with unnecessary security measures (more on this later in the post). Therefore, these systems employ intelligent risk assessment and adaptive fraud detection mechanisms to enhance accuracy and reduce misestimations.

​​Fraud problems in the mobility industry

The mobility industry's rise to popularity transformed the way we travel and commute. Unfortunately, many of the innovative services and cutting-edge tech haven’t been immune to deceptive practices. Addressing these issues requires a vigilant approach and the implementation of effective fraud detection systems.

One of the prevalent challenges in this sphere is account takeover. This occurs when unauthorized individuals gain access to users' accounts, compromising their personal information and exploiting it for fraudulent activities. With a solid fraud detection system in place, suspicious login attempts, unusual user behaviors, and IP address anomalies can be promptly identified and flagged, stopping unauthorized access and use.

Another area of concern is financial fraud. As the mobility industry heavily relies on online payment gateways, fraudsters attempt to exploit vulnerabilities in these systems. They may employ techniques such as stolen credit card information, fraudulent chargebacks, or fake identities to deceive service providers. By leveraging sophisticated fraud detection systems, suspicious payment patterns, inconsistent user information, and atypical transaction activities can be detected and investigated in real-time, thwarting potential fraud attempts.

Behavioral analytics from a scammer using Uber. Source: Uber

The mobility industry is also affected by identity theft and fraudulent account creation. Tricksters may set up fake user profiles using stolen identities or fabricated information to misuse services, engage in illegal activities, or exploit promotional offers. A well-designed fraud detection system employs identity verification checks, anomaly detection algorithms, and user behavior analysis to identify and block fraudulent account creation, thus preserving the integrity of the platform.

Loyalty program fraud poses a significant threat to the mobility industry as well. Fraudsters may attempt to exploit loyalty rewards and promotional offers by engaging in activities such as point manipulation, fake referrals, location spoofing or abusing reward redemption mechanisms. An effective fraud detection system can analyze loyalty program data as well as vehicle telematics data to identify suspicious behavior patterns, and implement measures to prevent fraudulent activities, safeguarding the loyalty program's value and integrity.

Uber’s fraud detection and prevention tech

Before we get into what the world’s perhaps most recognized mobility brand does to mitigate fraud-related risks, let’s talk some numbers first, to get a better sense of the scale of its operations. With this much business going on, scammers will naturally poke around for places susceptible to deceitful actions.

In 2022, Uber generated $31.8 billion in revenue, its drivers completed 7.6 billion trips, and about 131 million people have used the app. There are a lot of opportunities to try and game the system.

Unsurprisingly then, Uber has in place a number of fraud detection schemes designed to prevent tricksters from taking advantage of the platform. Let’s go ahead and dissect them now.

Uber’s RADAR Detect

RADAR Detect system is a tool they've developed to detect and combat fraud in their marketplace. It's an AI-powered system that keeps a close eye on different aspects of Uber's operations like countries, cities, payment methods, and device types. Using smart anomaly detection algorithms, it can quickly spot any suspicious activity that could indicate a fraud attack. Once a potential attack is identified, the system generates a rule to stop it in its tracks. But here's where the human touch comes in. The rule is then sent to fraud analysts, who carefully review and approve it before it goes into action. Once approved, the rule is applied to the marketplace, blocking or limiting any fraudulent transactions.

RADAR Detect is designed to be swift, accurate, and transparent. To meet these criteria, it uses synthetic data and Apache Spark to scrutinize the activity time-series from Uber's payment platform. By converting this data into understandable human actions and predictions, they can better anticipate and prevent fraud. The system also utilizes Uber’s proprietary probabilistic forecasting framework called Orbit 1.1 to enhance the accuracy of anomaly detection. To make things more transparent, Uber gets actual people involved in the process to create human-readable rules, which allows them to review past decisions, ensure accountability, and keep a complete record of every decision made.

Uber understands the importance of protecting their global marketplace from payment scams which can not only lead to significant financial losses, but also disrupts the experience for their users and partners. That's why they've developed RADAR Detect to combine the power of AI with human expertise. Working together, they aim to create a reliable solution for exposing and responding to fraud as early as possible.

Source: Uber

Uber’s RADAR Protect

RADAR Protect system, which is a key component of their RADAR Detect system, plays a crucial role in identifying attack patterns and helping to effectively put a stop to them. The way it works is it utilizes an advanced pattern-mining algorithm called FP-Growth which analyzes a slew of factors like device type, payment method, and location to uncover patterns that are associated with fraudulent activity. Once they're identified, the system generates rules based on them. But, as it was with the RADAR Detect, before they go into action, they're sent to human fraud analysts for a thorough review and approval. Since the rules are tailored to specific actions and attacks, they typically have a short lifespan.

Uber has designed its system to be adaptive, scalable, and highly efficient. To achieve this, they rely on cutting-edge technologies like Apache Spark and Peloton which process the synthetic data and run the pattern-mining algorithm. The system also employs probabilistic programming models, which help the RADAR engineers quickly adapt to changes in the data generation process and attack patterns. This adaptability is crucial in staying ahead of fraudsters. Additionally, Uber's rule engine platform plays a significant role in the RADAR Protect system. It enables fast rule deployment and evaluation, allowing for a swift and effective response to emerging threats.

RADAR's high-level design. Source: Uber

Uber’s Orbit

Orbit model is a Python package designed for Bayesian time series forecasting and inference. It offers an intuitive interface for tasks such as initialization, fitting, and prediction. Its flexibility, interpretability, and high performance allow Orbit to effectively handle various types of time series data and problems pertaining to potentially fraudulent activities.

Orbit serves as Uber's versatile interface for Bayesian time series modeling. It uses probabilistic programming packages like PyStan and Uber's Pyro, allowing for easy model specification and analysis without limitations on the number of models. The tool supports popular forecasting models such as Exponential Smoothing (ETS), Local-Global Trend (LGT), and Damped Local Trend (DLT). In its latest version (as of writing), Orbit v1.1 introduces new features, including a refined class design, forecaster, and kernel-based time-varying regression (KTR), enabling dynamic pattern exploration and detection within time series data.

Uber’s Relational Graph Convolutional Networks (RGCN)

Another one of Uber's anti-fraud measures that has been making waves are Relational Graph Convolutional Networks (RGCN). These deep learning models, rooted in neural networks, operate on graph data structures that represent entities (known as nodes) and their relationships (called edges). What sets RGCN apart is their ability to handle multi-relational graph data, allowing them to discern between various types of edges and employ them as valuable learning features.

An illustrated graph of connected drivers and riders. Red nodes represent fraud users, blue nodes represent legitimate users. Users are connected via shared information. Source: Uber

One key application of the RGCN model is the detection of collusion, a form of cooperative fraud involving users who engage in activities like fake trips using stolen credit cards. To tackle this issue head-on, the RGCN model employs a user graph that comprises two types of nodes: drivers and riders. Through shared information such as device, location, or payment method, drivers and riders become interconnected. The model then learns a unique vector representation, or embedding, for each node, effectively encoding user properties and their immediate community dynamics. Leveraging these embeddings, the RGCN model can accurately predict whether a user is engaging in fraudulent behavior or not.

Yet again, Uber tries to be adaptable with its solution, with users having the freedom to customize model components and priors to suit diverse use cases and data characteristics. Additionally, the RGCN model supports a range of sampling and optimization methods for effective model estimation and inference, including Markov-Chain Monte Carlo (MCMC), Maximum a Posteriori (MAP), and Variational Inference (VI). To further enhance its anomaly detection accuracy, the RGCN model also harnesses the power of the previously mentioned Orbit 1.1, a cutting-edge probabilistic forecasting framework.

Uber’s behavioral and spatial analytics against GPS-spoofing

One prevalent scam Uber has to deal with involves GPS spoofing, where tricksters use two phones – one as a new rider and the other as a driver. By simulating a trip without physically moving, using GPS-spoofing apps, scammers exploit Uber as a means for money laundering.

They quickly exhaust the stolen card's value and even take advantage of Uber's incentive programs based on the number of trips a driver has completed, earning undeserved bonuses. In the end, greed is the fraudsters’ downfall, as Uber is able to identify their scam by catching abnormal altitudes and unusually high speeds, which are due to them wanting to clock in as many rides as fast as possible.

The difference between a GPS-spoofed ride and a real ride's altitudes. The fake ride is in red, and hundreds of feet in the air. Source: Uber

Now that you know more about the ride hailing giant’s fraud prevention measures, you may also be interested in learning how to build an app like Uber.

Issues with automated anti-fraud systems

When discussing fraud management within the mobility industry, it's important to examine the challenges posed by automated systems. We should keep in mind that however well they may appear to be designed, they shouldn’t be considered infallible arbiters.

In recent times, a number of cases have emerged, shedding light on the flaws within Uber's automated fraud detection system and the failure of human review. These cases involve drivers who were unfairly dismissed after being flagged by Uber's Hybrid Real Time ID (RTID) system, leading to significant consequences not only for their access to the platform but also for their livelihoods.

One prominent issue revolves around the facial recognition component of RTID, which utilizes Microsoft's FACE API. This technology, designed to verify drivers' identities through live selfies, has come under scrutiny due to its failure to accurately match faces. Concerns have been raised regarding the efficacy of FACE API. In response, Microsoft highlighted the distinct roles and responsibilities of the tech provider and the system operator in ensuring fair and appropriate operation.

Beyond facial recognition, RTID is associated with complex location profiling systems employed by Uber to combat fraud. These systems trigger location detection when RTID selfies are submitted, prompting Uber's algorithms to review the GPS locations of devices linked to driver accounts. The majority of deactivations arise from accusations of account-sharing based on geolocation checks that detect account access from devices located significant distances apart within a short timeframe.

The vague description of “account access” leaves drivers uncertain and insecure, as they face the burden of appealing the decision, leading to mounting expenses and months of lost work.

Furthermore, the process of deactivation and subsequent review by Uber lacks transparency. Drivers are often contacted after dismissal and asked about unusual events or suspicious behavior without clear explanations or outcomes. The absence of comprehensive information initially provided to drivers raises questions about the objectives of human review and the complexities overlooked by algorithmic systems. These failures, both technological and human, result in the shifting of costs, risks, and the burden of proof onto the workers. Algorithmic management subjects drivers to constant evaluation against unknown classifications of fraud, improper use, or irregular activity.

Lessons learned and what’s next

Fraud management is a complex issue requiring businesses to take preventive steps in order to not incur financial and reputational hits.

The ACES (Autonomous, Connected, Electric, and Shared vehicles) sector of the mobility industry, which Uber is a part of, is growing, securing over $80 billion in investments since 2019. This means there’s going to be a lot more opportunities and attempts made by scammers to cut themselves a piece of this pie.

Circling back to the question posed in the article’s title, what we can learn from Uber’s fraud detection and prevention practices is that:

  • the bigger you get, the more attacks you’ll be subject to
  • if you have a unique business model, fraudsters will come up with creative ways to abuse it
  • you’ll most likely need multiple systems or measures to counter any possible attacks
  • measures taken to prevent fraud have to be perhaps even more sophisticated than the scammers’ attempts, although the latter get creative and evolve quickly
  • however amazing it may be, fraud detection and prevention technology may lack sufficiently deep understanding of nuances when making its assessments
  • thus, it must be designed in a way which makes human review useful and effective
  • you may not be in a position to anticipate every approach a scammer may take, so the ability to adapt quickly will be crucial

Now, what can you do about all this?

If you’re only just getting started, ditching your idea because of the threat of a scam somewhere down the line isn’t an option. If you’re already in business, perhaps it’s time to audit the fraud prevention measures you have in place to make sure they’re up to par.

In any of these cases, you may want to consider consulting the matter with a trusted tech partner like RST Software, who can help you navigate this dicey area and put up protective measures safeguarding your operations.

Feel free to contact us if this is something you’d like to discuss in more detail.

People also ask

No items found.
Want more posts from the author?
Read more

Want to read more?


10 German mobility startups and companies shaping the future of urban transportation

Explore the future of urban transportation with these 10 innovative German mobility startups and companies. Stay ahead of the curve.

Electric Vehicle Routing - the ultimate guide to the best EV navigation providers

Navigate electric vehicle routes with confidence. Explore the best EV navigation providers in this comprehensive guide.

Automotive HMI design and development: how to build a digital cockpit

Dive into the future of automotive HMI. Create a digital cockpit centered on safety, user engagement, and cutting-edge tech.
No results found.
There are no results with this criteria. Try changing your search.