<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ahmed Ossama on Medium]]></title>
        <description><![CDATA[Stories by Ahmed Ossama on Medium]]></description>
        <link>https://medium.com/@ahmed.ossama109?source=rss-fb49a753659a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*Oqw7huFUM7D-XAbu</url>
            <title>Stories by Ahmed Ossama on Medium</title>
            <link>https://medium.com/@ahmed.ossama109?source=rss-fb49a753659a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 18:54:10 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ahmed.ossama109/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Don’t build a Signal Graveyard: Flash’s Framework for Fraud Detection in Software Products]]></title>
            <link>https://medium.com/useflash-app/dont-build-a-signal-graveyard-flash-s-framework-for-fraud-detection-in-software-products-94bb7dd534b8?source=rss-fb49a753659a------2</link>
            <guid isPermaLink="false">https://medium.com/p/94bb7dd534b8</guid>
            <category><![CDATA[fraud-detection]]></category>
            <category><![CDATA[fraud]]></category>
            <category><![CDATA[fraud-prevention]]></category>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Ahmed Ossama]]></dc:creator>
            <pubDate>Sun, 19 Apr 2026 12:45:18 GMT</pubDate>
            <atom:updated>2026-04-19T12:45:18.263Z</atom:updated>
            <content:encoded><![CDATA[<p>Fraud teams love a good signal.</p><p>You spot a pattern. Bad actors cluster on the same IP. They create too many accounts from a specific device model. They use repeated names.</p><p>So you do the obvious thing: turn the signal into a rule.</p><p>Block users with a fake name and using any of those niche high traffic devices (e.g. Mr. <strong>Lplp Lpll </strong>using<strong> Oppo A5s </strong>with Android 8 in 2026<strong>)</strong></p><p>But there’s a hidden cost no one talks about enough:</p><p><strong>The moment you start hardening a signal into a prevention rule, you are often going to lose any value out of it.</strong></p><p>That is the <em>signal graveyard</em>.</p><h3>A true story: The shared IP rule</h3><p>Your analytics team notices a super high churn rate for users acquired through the new voucher the growth team worked on.</p><p>This triggered further investigation.</p><p>&gt; Oh, most of these users have very weird names, this doesn’t make sense.</p><p>Let’s dig deeper.</p><p>&gt; 70% of the churned users through this voucher came from a small fixed range of IPs.</p><p>Wait, you can’t be serious.</p><p>&gt; 60% of the acquired users are using HUAWEI phones.</p><p>Ok, I get it. This is definitely voucher abuse through fake account creation.</p><p>That’s around 60% of the money we spent on these vouchers, being wasted on fake accounts.</p><h4>Now what?</h4><p>Alright, we have to stop this somehow.</p><p>&gt; We can’t add a rule to block users with weird names, it is hard to predict.</p><p>&gt; We also can’t block HUAWEI phones completely. This will include many false positives.</p><p>Okay, let’s add a rule to limit registration from the same IP.</p><p>We’ll only allow up to X users to register from the same IP everyday. At least we will limit the fake accounts they are creating.</p><p>PR is out. Reviewed. Deployed. ✅</p><p>Dashboards look great. Registrations are now evenly distributed across IPs.</p><p>We can now see the failed attempts to register because of sharing the same IP.</p><h4>A few days later</h4><p>The growth team sends the following message:</p><p>“Hey, churn on that voucher campaign is still abnormally high.”</p><p>We pull the data again.</p><p>The fake accounts are still there. Same weird names. Same HUAWEI devices. Same behavioral patterns.</p><p>But now? They’re spread across <strong>hundreds of different IPs</strong>.</p><p>The fraudster didn’t stop. They must have figured out a way to bypass our registration block rule based on similar IPs.</p><p>A <a href="https://medium.com/@datajournal/what-are-residential-proxies-67023101f356">residential proxy</a> pool costs a few dollars a month. You mass-deployed a rule, and all it took was a $5/month subscription to bypass it.</p><h4>The real damage</h4><p>Aside from wasted development time, we lost an important signal that can reliably tell the fraud users.</p><p>Fraudsters have evolved their flow, and they are now harder to differentiate from the real users because they use unique IPs.</p><p>The registration IP signal is now dead.</p><p>It doesn’t stop the fraud actors, and it doesn’t help us even track those actors.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gfegiR_WC5GsKNgf0p-P2A.png" /></figure><h4>Wisdom: Measurement vs Prevention</h4><p>A more solid way to think about fraud detection is to split it into two separate efforts:</p><ul><li><strong>Measurement</strong></li></ul><p>You can never eliminate fraud completely, you wanna make the best effort to measure how much fraud is <strong><em>currently happening</em></strong> in your system.</p><ul><li><strong>Prevention</strong></li></ul><p>Enforce rules into your system to detect a fraudulent attempt <strong><em>before it concludes</em></strong> and prevent the damage.</p><p>Remember: <strong>“If you can’t measure it, you can’t manage it</strong>.”</p><p>Without proper measurement in place, you can’t really drive prevention efforts and assess their efficiency.</p><p>The key point is that using a signal for prevention often degrades its value for measurement, so treat them as competing uses and choose deliberately.</p><p>In our story above, we could have used the <strong>registration IP signal as</strong> <strong>measurement signal</strong> to reliably flag fraud accounts.</p><p>Better focus on <strong>finding another signal for prevention, </strong>that is more reliable (and also measurable that now IPs can help us tell)<strong>.</strong></p><h4>Framework principles</h4><ul><li><strong>Measurement first</strong></li></ul><p>Never rush to preventing fraud, unless you really can measure it.</p><p>Take a few rounds/cycles/sprints to build a measurement framework first to tell how much fraud does your system currently has.</p><p>A nice tip here is that measurement can tolerate false positives.</p><p>It is meant to show you trends, new abnormal patterns that need to be investigated. But not meant to be bullet proof accurate like prevention.</p><ul><li><strong>Trial and error for prevention</strong></li></ul><p>Some signals are hard to tell if they are easy to bypass or not (especially with teams with no prior experience in this area).</p><p>The recommendation here is to try the signal for prevention, yet ensure that you can detect when your rule is bypassed (through a good set of measurement rules).</p><ul><li><strong>Cleanup the signal graveyard</strong></li></ul><p>If a prevention rule becomes obsolete, i.e. fraudsters were able to bypass it, have the courage to clean up the rule and move it back to your measurement arsenal.</p><ul><li><strong>Acknowledge tricky fraud patterns</strong></li></ul><p>It is <strong>completely fine</strong> to have a specific flow of fraud, that <strong>you know how to measure, but don’t know how to completely prevent yet</strong>.</p><p>Prevention rules often need deep investigations, a lot of logging, experimentation, and asking around to know what can effectively cut the flow.</p><h4>We’re together on this</h4><p>At Flash, we’re continuously investing in making our fraud detection smarter.</p><p>This is an evolving challenge, and we believe the best solutions come from shared learnings across the industry.</p><p>If you’re tackling similar problems, especially in the Egyptian market, we’d love to exchange ideas. Reach out at <a href="/service/mailto:engineering@useflash.app">engineering@useflash.app</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=94bb7dd534b8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/useflash-app/dont-build-a-signal-graveyard-flash-s-framework-for-fraud-detection-in-software-products-94bb7dd534b8">Don’t build a Signal Graveyard: Flash’s Framework for Fraud Detection in Software Products</a> was originally published in <a href="https://medium.com/useflash-app">useflash.app</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>