Credit Card Fraud Detection

Here's how Stripe detects frauds with a 99.9% accuracy in 100 milliseconds (that too by checking over 1000 parameters for one transaction) Fraud detection in online payments isn’t just about stopping bad transactions it’s about doing it fast, at scale, and without blocking legitimate users. Stripe’s fraud prevention system, Radar, evaluates 1,000+ signals within 100 milliseconds to make decisions. Here’s how it works and why it’s so effective: 1. ML Models That Learn and Scale Stripe started with simple ML models (logistic regression) but quickly scaled to hybrid architectures combining: –XGBoost for memorization (catching known patterns). –Deep Neural Networks (DNNs) for generalization (handling unseen patterns). –Key Problem: XGBoost couldn’t scale or integrate modern ML techniques like transfer learning and embeddings. –The Solution: Stripe moved to a multi-branch DNN-only architecture inspired by ResNeXt. This setup allowed it to memorize patterns while staying scalable. It reduced training times by 85%, enabling multiple experiments in a single day instead of overnight runs. 2. Learning From Real Fraud Patterns Radar doesn’t just rely on static rules, it learns from data across Stripe’s network. –Engineers analyze fraud attacks in detail, e.g., patterns of disposable emails or repeated card testing. –Features like IP clustering and velocity checks were added to detect suspicious activity. –Fraud insights are shared across the network, so lessons learned from one business protect others automatically. Example: Analyzing IP patterns helped detect high-volume attacks where fraudsters used multiple stolen cards from the same source. 3. Scaling With More Data, Not Just Smarter Models Stripe realized that more training data could unlock better performance, similar to modern LLMs like GPT models. It tested scaling datasets by 10x and 100x. Result? Performance kept improving, confirming that larger datasets and faster training cycles work better than complex rules alone. Key Insight: Bigger datasets help uncover rare fraud cases, even if they occur in only 0.1% of transactions. 4. Explaining Fraud Decisions Clearly Fraud systems often act like black boxes, leaving businesses guessing why a payment failed. Stripe built Risk Insights to provide clear explanations: –Shows features contributing to fraud scores like mismatched billing and shipping addresses. –Displays maps and transaction histories for visual context. –Enables custom rules to fine-tune fraud checks for specific business needs. Result: Businesses trust Radar’s decisions because they can see why a payment was flagged. 5. Constant Adaptation to Stay Ahead Fraud patterns evolve, so Stripe built Radar to adapt in real time: Uses transfer learning and multi-task learning to generalize better. Incorporates insights from the dark web and emerging fraud tactics. Continuously retrains models without disrupting performance.

Fraud wasn’t supposed to be a core product challenge. But for most businesses operating online today, it has staunchly become one. In 2024, Indian businesses lost ₹22,842 crore to cybercrime. That’s a 206% increase over the previous year. The first few months of 2025 have already added another ₹7,000 crore in losses. This isn't just a compliance or security concern anymore. It shows up as frozen accounts, locked working capital, rising chargebacks, and misuse through stolen cards, fake UPI payments, and promo abuse. What surprised us most was how quickly chargebacks became part of the everyday reality for merchants: 1. More than half involve deliberate abuse 2. Smaller businesses aren’t spared - around 30 percent of Indian SMEs now report direct losses from fraud, with revenue hits of up to 5 percent. The nature of fraud has changed. Attacks are faster, more coordinated, and more sophisticated. The usual playbook of reacting after the damage doesn't hold up anymore. We decided to rebuild our approach from first principles. RiskShield is what came out of it. It’s a fraud detection engine that runs within the payment flow. It scores every transaction in real time using machine learning, detects fraud rings using graph intelligence, syncs with government risk data like I4C, DoT blacklist, NCRB, and blocks bad actors mid-transaction. It also flags early signs of promo abuse, card testing, and UPI manipulation. So far, RiskShield has helped block over ₹1,700 crore in fraud attempts. It has flagged 2 crore high-risk signals and protected more than 6,600 merchants. The system operates quietly in the background, with an F1 score of 87 percent which is a measure that balances precision (how often fraud alerts are correct) and recall (how much fraud we actually catch) and recall close to 95 percent. Most issues are prevented before anyone files a complaint. There’s still more work to do, but one thing is clear to us now: Fraud cannot be treated as an after-effect. It has to be designed against from the beginning. PS. Here's the flow we have built ⬇️

🚨 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐏𝐚𝐲𝐦𝐞𝐧𝐭𝐬 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 𝐢𝐧 𝐌𝐨𝐭𝐢𝐨𝐧 — 𝐅𝐫𝐚𝐮𝐝 𝐏𝐫𝐞𝐯𝐞𝐧𝐭𝐢𝐨𝐧 by DEUNA Traditional, static fraud rules often fall short — tightening controls so much that they block good customers, or leaving gaps that allow fraud to slip through. Agentic intelligence changes this paradigm. By leveraging historic transaction data and strategic signals (PSPs, payment methods, geographies, behavioral trends), it dynamically recommends risk controls tailored to each scenario. — 𝐃𝐞𝐞𝐩 𝐃𝐚𝐭𝐚 𝐂𝐨𝐧𝐭𝐞𝐱𝐭 Historic transaction patterns and behavioral signals are integrated with granular specifics like BIN, card franchise, and geography. This allows the system to distinguish between legitimate customers and potential fraud with precision. → The Walt Disney Company leverages historical subscription behavior data to differentiate genuine recurring payments from suspicious account takeovers, reducing false declines. — 𝐋𝐨𝐰 𝐑𝐢𝐬𝐤 𝐯𝐬 𝐇𝐢𝐠𝐡 𝐑𝐢𝐬𝐤 𝐓𝐫𝐚𝐧𝐬𝐚𝐜𝐭𝐢𝐨𝐧𝐬 Low-risk transactions flow seamlessly with minimal friction, boosting conversion and improving customer satisfaction. High-risk transactions are dynamically routed through targeted fraud prevention layers — activating the most relevant PSPs and antifraud providers at the right moment. → Uber adapts fraud checks by geography, applying stronger measures in regions with high fraud incidence while keeping repeat riders’ payments frictionless. — 𝐏𝐫𝐨𝐯𝐢𝐝𝐞𝐫 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 𝐰𝐢𝐭𝐡 𝐅𝐫𝐚𝐮𝐝 𝐂𝐨𝐧𝐭𝐞𝐱𝐭 Risk scoring is factored into provider and PSP selection to balance approval rates, cost efficiency, and security. → Airbnb leverages intelligence to dynamically adjust fraud controls by market and traveler profile — applying stronger authentication in high-risk regions or for first-time guests, while allowing frictionless payments for trusted, repeat customers. — 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐞𝐝 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐚𝐭 𝐒𝐜𝐚𝐥𝐞 Fraud tools are embedded directly into the orchestration layer, enabling smarter allocation: fraud detection where it is most impactful, and seamless flows where customers have already proven trustworthy. → Worldline merchants leverage adaptive authentication, activating 3DS selectively when intelligence identifies elevated risk — enabling smoother experiences for low-risk customers. — The Result → Intelligent Growth with Protection ✅ Higher approval rates without compromising safety ✅ Smarter allocation of fraud tools where they matter most ✅ Frictionless checkout experiences for trusted customers — This is proactive fraud prevention in motion — moving beyond rigid rules into an era of intelligent orchestration, where every payment decision optimizes both security and customer satisfaction at scale. — Source: DEUNA ► Subscribe to The Payments Brews: https://lnkd.in/g5cDhnjC ► Connecting the dots in payments... | Marcel van Oost

I never thought it would happen to me. One day, I noticed a spike in chargebacks. I knew something was wrong, but I didn’t know what. I started by investigating the types of fraud we were experiencing. From fake accounts to transaction fraud, it was overwhelming. Here’s how to detect and prevent fraud at every stage of the customer journey: Stage 1: Data Collection Data is your first line of defense. • Gather as much user data as possible. • Track device information, IP addresses, and user behavior. • Monitor changes in user activity. Understanding user patterns helps in identifying anomalies early. Stage 2: Basic Risk Scoring Identify low-hanging fruit. • Use simple rules to score transactions. • Look for mismatched billing and shipping addresses. • Flag unusual purchasing behaviors. This stage catches the most obvious fraud attempts. Stage 3: Dynamic Friction Balance security and user experience. • Implement step-up authentication for suspicious activities. • Use dynamic risk based routing • Introduce verification processes at critical points. Dynamic friction helps reduce fraud without hurting conversion rates. Stage 4: Advanced Analytics Deep dive into data for insights. • Use machine learning to detect patterns. • Analyze transaction histories and behaviors. • Integrate third-party data sources for enhanced detection. Advanced analytics provide a comprehensive view of potential threats. Stage 5: Continuous Optimization Stay ahead of evolving threats. • Regularly update your fraud detection rules. • A/B Test and refine your strategies. • Stay informed about new fraud techniques and trends. Continuous testing ensures your not two steps behind fraudsters. A comprehensive fraud strategy requires a layered approach.

Let's build a Real Time ML System to fraud. Step by step 🧵↓ 𝗧𝗵𝗲 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 💼 Every time your credit card is used online by someone (hopefully you), your card issuer (for example Visa, Mastercard or PayPal) has to verify if it is you the person trying to pay with the card. Otherwise, the transaction is blocked. Now the question is: ““𝗛𝗼𝘄 𝗱𝗼𝗲𝘀 𝗩𝗶𝘀𝗮 𝗱𝗼 𝘁𝗵𝗮𝘁?”” And the answer is… a real time ML system! 𝗦𝘆𝘀𝘁𝗲𝗺 𝗱𝗲𝘀𝗶𝗴𝗻 📐 As any ML system that has existed, exists and will exist, this one can be broken down into 3 types pipelines 1️⃣ Feature pipelines 2️⃣ Training pipeline 3️⃣ Inference pipeline Let's go one by one 1️⃣ 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀 💾 The feature pipelines are the Python services that produce the inputs (aka features) our ML model needs to generate its predictions. In our case, we have (and I bet Visa has) at least 3 feature pipelines: ▣ 𝗥𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 feature pipeline from recent transactional data. - runs 24/7 - consumes incoming data from an internal message bus (like Kafka, Redpanda) - transforms this data on-the-fly using a real-time data processing engine - saves the the final features in a feature store, like Hopsworks. ▣ 𝗕𝗮𝘁𝗰𝗵 pipeline from historical features in the data warehouse. - runs daily - reads data from the data warehouse/lake, and - saves it into another feature group in our feature store, so it can be consumed by our ML model really fast. ▣ 𝗟𝗮𝗯𝗲𝗹𝘀 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲, so the ML model can be trained with supervised ML. Each completed transaction that is not claimed by the card owner within 6 months can be safely called non-fraudulent (class=0). We call it fraudulent (class=1) otherwise. Once we have these 3 feature pipelines up and running, we will start collecting valuable data, that we can use to train ML models. 2️⃣ 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲 🏋🏽 We can use a supervised ML model (a boosting tree model like XGBoost does the job in most cases) to uncover any patterns between > the features available in your Feature Store, and > the transaction class: 0 = non-fraudulent, 1 = fraudulent. The final model is pushed to the model registry (like MLflow, Comet or Weights & Biases), so it can be loaded and used by our deployed model. And this is precisely what the last pipeline in our design does. 3️⃣ 𝗜𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲 🔮 The inference pipeline is a Python streaming application, that at start up loads the model from the registry into memory and for every incoming transaction > loads the freshest features from the store for that card_id, > feeds them to the model, and > outputs the predictions These fraud scores can be then consumed by downstream services, to > Block the card, and > Send an SMS alert to the card owner, for example. --- If your company is looking for an AI consultant to accelerate business growth, or if you are looking to build your career in AI, send me a message and I can help

🧠 Let’s talk about fraudster psychology. If I handed you a $500 debit card, but told you: ▪️ You can only use it for a single purchase. ▪️ You’re competing with 99 other people racing to drain it first. ⏰ What would you do in the next 10 minutes? It’s a thought experiment worth sitting with – because it’s exactly the headspace fraudsters live in. And under these conditions, your behavior wouldn’t look anything like a normal customer’s. 🔍 If a fraud fighter had perfect visibility into your actions, here’s what they’d see: ▪️ You blitz through the shopping flow and ship fast. The product needs to leave the warehouse before the order gets flagged. ▪️ You sort searches by highest price. Maximum value per transaction is the only game. ▪️ You go for gift cards if you can – somewhere to stash the value, privately. ▪️ You skip anything that slows you down. Promo codes? Loyalty signups? Not your problem. ▪️ You don’t browse. You don’t compare. You don’t optimize. You’re not looking for value, you’re looking for extraction. That behavioral fingerprint – the urgency, the tactics, the intent – is how fraudsters give themselves away. And that’s what Spec sees: the psychology of the fraudster behind their behavior. We capture the entire customer journey: every page, every click, every hesitation, every tool. We don’t wait for checkout. We detect fraud in motion, before it becomes a loss. Fraudsters are trying to beat the system. 🧠🔍 We built a system that sees how they think.

Mastercard's recent integration of GenAI into its Fraud platform, Decision Intelligence Pro, has caught my attention. The results are impressive and shows the potential of “GenAI in Advanced Business Applications”. As someone who follows AI advancements in Fraud across the FSI industry, this news is genuinely exciting. The transformative capabilities of GenAI in fortifying consumer protection against evolving financial fraud threats showcase the potential impact of this integration for improving the robustness of AI models detecting fraud. The financial services sector faces an escalating threat from fraud, including evolving cyber threats that pose significant challenges. A recent study by Juniper Research forecasts global cumulative merchant losses exceeding $343 billion due to online payment fraud between 2023 and 2027. Mastercard's groundbreaking approach to fraud prevention with GenAI integrated Decision Intelligence Pro is revolutionary. - Processing a staggering 143 billion transactions annually, DI Pro conducts real-time scrutiny of an unprecedented one trillion data points, enabling rapid fraud detection in just 50 milliseconds. - This innovation results in an average 20% increase in fraud detection rates, reaching up to 300% improvement in specific instances. As we consider strategic imperatives for AI advancement in fraud, this news suggests what future AI models must prioritize: - Rapid analysis of vast datasets in real-time, maintain agility to counter emerging fraudulent tactics effectively, and assess relationships between entities in a transaction. - By adopting a proactive approach, AI systems should anticipate and deflect potential fraudulent events, evolving and learning from emerging threats to bolster security. - Addressing the challenge of false positives by evolving AI models capable of accurately distinguishing legitimate transactions from fraudulent ones is vital to enhancing overall security accuracy. - Committing to continuous innovation embracing AI is essential to maintaining a secure and trustworthy financial ecosystem. #artificialintelligence #technology #innovation

How Stripe's "Money-GPT" Slashed Fraud 𝗔𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻 𝗶𝘀 𝗔𝗹𝗹 𝗣𝗮𝘆𝗺𝗲𝗻𝘁𝘀 𝗡𝗲𝗲𝗱 Payments are tricky. Fraud? Trickier. Stripe just flipped the script. Instead of manually picking apart payments, feature by tedious feature (BIN, zip, payment method : you get the drift), they went big. Transformers big. Think ChatGPT - but make it money smart. Turns out, payments speak their own kind of language. A bit of syntax here, some semantics there. Transactions cluster together, humming in harmony. Cards from the same issuer? Neighbors. Same bank? Even tighter. Shared email? Practically twins. Stripe's new foundation model spots fraud patterns, even subtle ones, real-time. Card testing detection jumped from 59% to 97%. Overnight. Money talks. Stripe listened. Link to the post in comments #Stripe #FraudDetection #GPT #PaymentSecurity #FraudPrevention #LLMs ---- 👉 Follow me, Robin Jose for daily insights on AI, leadership, and technology. ❤️ Like and share if you found this valuable!