Tech Guru’s Post

Machine learning presents a formidable array of tools to combat fraud in finance. To build a robust fraud detection system, some key considerations must be followed: - Understand your data - Choose the appropriate algorithms - Address imbalanced datasets - Continuously monitor and update models - Ensure data privacy and compliance These steps are crucial in leveraging machine learning effectively in the battle against financial fraud. #MachineLearning #FraudDetection #FinanceFraud #DataPrivacy #Compliance https://lnkd.in/e-ry2kJa

If my solutions are accurate, why do I have so many false positives? Must be performance degradation, right? 📉 🤔 Well, there could be another reason for it, one that is usually ignored. That is, the performance success criteria. Specifically, how do you measure it - absolutely or incrementally? Short explainer: Let’s say you’ve released a rule that utilizes an AI score and out of 100 hits, it blocks 90 fraud cases. Pretty good, right? For each 9 fraud cases blocked, you have 1 false positive. However, this ignores the incremental benefit of the new rule. Say that out of the 100 hits, 50 are already blocked by other means. All of them are fraud. Your effect on reality then, is that you block 40 fraud cases and create 10 false positives. Same rule, different measurements - 9:1 or 4:1 fraud blocked per false positive. 😵 As simple as the concept sounds like, even when being aware of it, it can be very challenging to simulate incremental performance against your live system. One reason can be that you simply don’t have the tooling features that allow you such tests. Other times, it’s complicated to simulate different solutions from different services. Imagine if I employ hard-coded score thresholds, a vendor’s rule engine and an internal tool I’ve built for keeping “block lists”. Is it possible for me to simulate the entire system end-to-end? As said, many times the answer is “no”. So when your team is deploying fresh solutions that just proved highly accurate in simulations, in effect you’re just hitting the same fraud cases while ramping up false positive cases. ❓ So how would you avoid this situation? ✔ As the saying goes - what gets measured gets managed. Solution policies and their success criteria should be based on incremental KPIs that represent the change to the system rather than the solution in a lab environment. Once these are the KPIs you are measured by, you’ll start finding the ways to uphold them. It’ll probably be hacky at first, and will likely involve some of the ugliest SQL known to man But that’s usually how a first step in the right direction looks like. The level of control that awaits at the end of this journey will be worth it.

Need to create a security newsletter or inforgraphic for your workforce but feeling uninspired or simply out of creative 'juice'? Here is a simple little AI prompt that can generate a draft for you to get you started. For the example below I use "protecting your financial accounts" but you can replace that with any topic. Don't like the output, find it too formal or boring? Follow-up with the prompt "do it again but make it sassier" or "do it again but make it more technical". PROMPT: I'm the security awareness officer for my company and I would like to create a security awareness newsletter. The content should apply to anyone who uses a computer. Please create a newsletter on how people can lock down and secure their online financial accounts, such as their checking, savings or investment accounts. Start the newsletter with a story of how someone falls victim to a scam and how their financial accounts are accessed and their money stolen. Then after the short story explain why protecting your financial accounts is so important and five steps people can take to protect their financial accounts. Keep the newsletter to 800 words or less." What I often do is combine elements of the "sassy" version with elements of the "formal" version to create my own version of the newsletter. #AI #securityawareness #securityculture

How do you secure data in hybrid cloud and safely leveraging #AI for business. The Treasury recently released its report (https://lnkd.in/gEbfGJcT) on managing AI and risks to security and resiliency, particularly for the financial services sector. Check it out

"#Doxing" (derived from "docs"/"dox", abbreviation of the English word "documents") refers to the collection and publication of personal #data on the Internet with various, sometimes criminal intentions. Even if #datatheft cannot be completely prevented, it is up to every user to make it more difficult for the perpetrator to access personal data and minimize the impact. https://lnkd.in/gwsUH6Ba #worlddatasummit #dataliteracy #customeranalytics #dataanalytics #bigdata #opensource #datascience #datainnovation #dataanalysis

🔍 Discover the Future of Finance: Synthetic Data Revolutionizing Risk Assessment and Fraud Detection 🔍 In the data-driven realm of finance, synthetic data emerges as a game-changer, balancing insights with privacy and innovation for proofs-of-concepts (PoCs). Follow-us on X @ https://lnkd.in/eXwhucA6 PS: Currently, our website is being upgraded, and we're adding original content. #SyntheticData #RiskAssessment #FraudDetection #USFintechNews #EuropeFintechNews #InnovationStreets

To automatically identify complex types of fraud or financial crime, you need to bring in data from all kinds of systems - transactions, finance, CRM, ERP, website analytics, contact centre and information from the device. You have to mine that data to look for complex fraud patterns in your datasets, and then put systems into place to identify when that class of fraud occurs in future. Though there are some great platforms such as Sardine that are flexible enough to catch complex scenarios, I don't think businesses can get away from the fact that they also need to analyse their own data using data science techniques, then deploy real time analytics and machine learning models to continually check when the situations of interest are occurring in future. To some businesses and fincrime teams, this sounds like a lot of work or somewhat intimidating. However, if you are still losing millions of dollars to fraud despite your best efforts, then it is worthwhile hiring data engineers, data scientists, and ML engineers and getting your hands dirty with your own data. The types of fraud risks are only going to get more complicated over time and data and analytics are a muscle that businesses of scale need to build. Ideally, this would integrate with modern and sophisticated fincrime tech to really reach the highest levels of detection and capability. Fortunately, this doesn't have to be disruptive to your existing technology stack. You can use these analytics to expose a few additional reports or dashboards, or even integrate the alerts back into your existing systems. Using modern cloud based tools, the investment in bespoke data and analytics doesn't have to be that high either.

Automated Classification benefits for eDiscovery GREATLY & is already included in all leading eDiscovery solutions. So why are you still not using it? Find out why and read up on the time saving and cost saving benefits here in this interview I had with Simon Wright at UC Today: https://vrt.as/3SCv7lt

Question of the Day “Share three things you would do to make the Office of State Auditor more effective?” Demand Letter State Auditor has subpoena authority to require information to be provided to assist in the investigation and identification of waste and fraud. Currently the State Auditor has no authority to pursue waste and fraud. Demand letter authority will enable the State Auditor to immediately pursue and recover monies that were fraudulently gained and stop the flow of monies that are being wasted. Common fund codes There are over 40 agencies/departments as well as 16 universities and 58 community colleges that purchase computers, software and other information technology (IT) related items under contracts through the state. It is estimated that the total annual IT spend by the state is approaching $3B. The problem, nobody knows for sure. Nobody knows if discounts are being maximized or even taken. Nobody knows how many duplicate products and services are being purchased. Departments use their own set of fund codes for entry into the chart of accounts in their general ledgers. We need to immediately implement a single common set of fund codes for IT related expenditures. This change will result in $30M to $50M in annual savings for tax payers. The savings can be accomplished through this simple administrative change. Technology Day one begin standing up a center focused on detecting and preventing fraud and waste through the use of advanced technologies. Will leverage AI, ML, and auditbots. Blockchain, sophisticated identity management and rules engine tools will be utilized to assist in pattern recognition and identification of fraud. Will establish a function similar to the Program Evaluation Division within OSA to leverage human skills of AI engineers, forensic auditors and data analytic scientists. I have four departments preliminarily identified that should result in returns of over $100M to tax payers.

Deepfakes threat should be taken seriously! How to respond: Awareness, Tools & Processes! Thanks CFO Dive for addressing this topic, and for the references! CC Medius https://lnkd.in/eSwNpNht