<?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 Pornpimol (Gym) Kaewphing on Medium]]></title>
        <description><![CDATA[Stories by Pornpimol (Gym) Kaewphing on Medium]]></description>
        <link>https://medium.com/@gympornpimol?source=rss-aea15c448e5b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*PyK2mNm0gOLBVwt-Nm1cUA.jpeg</url>
            <title>Stories by Pornpimol (Gym) Kaewphing on Medium</title>
            <link>https://medium.com/@gympornpimol?source=rss-aea15c448e5b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 03:56:10 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gympornpimol/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[Designing for Data Integrity — When Good Automation Fails Bad Data]]></title>
            <link>https://medium.com/@gympornpimol/designing-for-data-integrity-when-good-automation-fails-bad-data-a3a6263aec60?source=rss-aea15c448e5b------2</link>
            <guid isPermaLink="false">https://medium.com/p/a3a6263aec60</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[insurance]]></category>
            <dc:creator><![CDATA[Pornpimol (Gym) Kaewphing]]></dc:creator>
            <pubDate>Sun, 23 Nov 2025 03:25:31 GMT</pubDate>
            <atom:updated>2025-11-23T07:01:04.174Z</atom:updated>
            <content:encoded><![CDATA[<h3>Designing for Data Integrity — When Good Automation Fails Bad Data</h3><h4>Behind every broken dashboard, there’s usually a quite data mistake. Here’s how we found ours — buried inside a null field.</h4><p>In my previous post — “<a href="https://medium.com/@gympornpimol/the-real-power-of-data-isnt-in-dashboard-it-s-in-design-6797ba4c0a57">The Real Power of Data isn’t in Dashboard — It’s in design</a>.” I wrote about data doesn’t just describe behavior, it shapes it.</p><p>This story is the first real case behind that idea. It’s about how a simple missing phone number, one forgotten field , quietly broke an entire renewal system. And how fixing it didn’t just restore performance, but revealed a new way to think about data design altogether.</p><p>When the renewal campaign underperformed every thought the automation was broken. <br>We’d just launch a new flow to help customers renew their auto insurance through a simple SMS link — no calls, no paperwork. The idea was to free up the sales team for higher-value work while making renewal effortless for customer.</p><blockquote>Everything looked fine on paper.<br>The model worked.<br>The SMS system reported “SUCCESS.”<br>And yet, conversions were…flat.</blockquote><blockquote>So what went wrong? 💭</blockquote><h3><strong>🚨The First Red Flag: “I Never Got the Message.”</strong></h3><p>It started with a few unexpected calls to contact center.</p><blockquote>“My policy’s expiring next week. I never got any contact or any reminder to renew.”</blockquote><p>At first, the team suspected a delivery issue. Maybe a vendor glitch. But as more customers called, the pattern became too consistent to ignore.</p><p>The dev logs looked clean. The automation pipeline ran perfectly.<br>Then, buried in the outbound dataset, we found it: 🥶</p><blockquote><strong>Thousands of records had null phone numbers.</strong></blockquote><h3>🕵️‍♀️ The Investigation</h3><p>We traced the issue upstream, checking where the SMS list was being built.</p><p>It turned out the automation wasn’t wrong, the <strong>data model</strong> was.<br>Some customer phone number were stored in the wrong field during manual entry, so the system simply couldn’t find them.</p><p>This explained everything:</p><ul><li>The model ran fine.</li><li>The delivery logic worked.</li><li>But the dataset didn’t know <em>who to text.</em></li></ul><h3>💡The Pivot: Fix the Data, Not the Code</h3><p>Instead of waiting for a dev sprint, we looked for a faster way forward.</p><blockquote>“What if we fix the data model instead?”</blockquote><p>We created a new reference table that included verified phone numbers from other customer touchpoints such as claims calls or service updates where customers might have updated their contact info.</p><p>Then we adjusted the logic to always pull the most recent, verified number available:</p><blockquote><strong><em>Design Fix in One Line</em></strong><br>We introduced a verified reference phone source, the forced the model to fail-over intelligently using COALESCE():</blockquote><pre>COALESCE(t1.customer_phone_no, t2.customer_phone_no, t3.customer_phone_no) <br>  AS customer_phone_no<br>--------------------------------------------------------------------------<br>-- t1 = reference table where the customer updated phone number<br>-- t2 = customer profile table where they bought an auto policy<br>-- t3 = customer partnership table where the partnership input <br>-- customer information <br>--------------------------------------------------------------------------</pre><p>This simple change ensured that every customer in the renewal list had a valid, up-to-date phone number, even if their primary record was outdated or incomplete.</p><h3>🚀 The Impact: Three Lines That Changed Everything</h3><p>After rolling out the fix, the next SMS batch told a very different story.</p><ul><li>Take-up rate jumped to <strong>around 60%</strong>, up from single digits before.</li><li>The channel began generating <strong>over a million bath in monthly impact.</strong></li><li>And the best part, it required <strong>no new features, no extra manpower.</strong></li></ul><p>All it took was better data logic.</p><h3>🧠 The Lessen: Automation isn’t Smart Until the Data Is</h3><p>Here’s what we learned</p><blockquote><strong><em>“Sometimes the system isn’t wrong because the code fails.<br>It’s wrong because the data model didn’t know enough.”</em></strong></blockquote><p>Once we redesigned what the system <em>knew</em>, everything else, the logic, the results, the revenue followed naturally.</p><p>In the next post, I’ll breakdown:</p><ul><li>How we rebuilt our mapping logic to prevent future breaks</li><li>The validation layers that made our data model self-correcting</li><li>And how a small design change turned reactive fixes into a proactive system</li></ul><p><strong>😉<em>“Good data isn’t a report. It’s a design choice.”</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WsH0HZsfCqInzNECskFHkg.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a3a6263aec60" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Real Power of Data isn’t in Dashboard — It’s in design]]></title>
            <link>https://medium.com/@gympornpimol/the-real-power-of-data-isnt-in-dashboard-it-s-in-design-6797ba4c0a57?source=rss-aea15c448e5b------2</link>
            <guid isPermaLink="false">https://medium.com/p/6797ba4c0a57</guid>
            <category><![CDATA[insurance]]></category>
            <category><![CDATA[data-strategy]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[automation]]></category>
            <dc:creator><![CDATA[Pornpimol (Gym) Kaewphing]]></dc:creator>
            <pubDate>Sun, 23 Nov 2025 03:22:54 GMT</pubDate>
            <atom:updated>2025-11-23T06:58:02.201Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Real Power of Data isn’t in Dashboard — It’s in design</h3><h4>A real-world case of how fixing data flow quality, and system sync reshaped a legacy insurance operation.</h4><p>Most people think of data as something you look at a dashboard, a report, a metric. But in the real world of operations <strong>data isn’t just something you observe. It’s something that shapes how systems behave.</strong> In our attempt to automate a routine insurance renewal process, we expected the heavy lifting to happen in the code. But what we learned through three unexpected breakdowns was that the root cause wasn’t technical.</p><p>⚠️ It was data: missing links, wrong mapping, broken flows. And once we fixed the data, the systems followed. This is the story of how we used data not just to report what went wrong but to redesign what worked.</p><h3>What We Tried to Automate and Why?</h3><p>🚀<strong>Company Goals:</strong> Scale operations to automate insurance renewal.</p><p>❓<strong>Problems</strong>: (1) There was a limitation of capacity from sales team to call asking potential customers with 30–60 days motor policy expiring, the sales team did another tasks for supporting other product also. (2) There were too manual processes for collecting and inputing the data that led to the errors and costs.</p><p><strong>🌟Ideas:</strong> How to encourage customers to renew their motor policy by themself??</p><p><strong>🍀Concept:</strong> Sending SMS to notify customers with what do we offer them if they extend policy with us.</p><p>The business wanted move fast renewals for auto insurance had been handled manually for years — call centers, spreadsheets, human follow-up. But with growth on the horizon, leadership set a clear direction.</p><blockquote><strong>Can we remind, renew, and re-engage customers without manual touch?</strong></blockquote><p>The target was bold but simple</p><ul><li>Automatically remind customers before their policy insurance expired</li><li>Reduce call center workload</li><li>Increase self-renewals by offering personalized quotes via SMS</li><li>And track performance across all steps, in real time</li></ul><blockquote>No new headcount. No more phone call. Just smarter data, better timing, and a system that works on its own.</blockquote><p>In short: Turn a reactive workflow into a proactive engine.</p><p>“But the moment we tried to run it at scale, things started to break — and the issues weren’t in the automation logic. They were in the data itself.” 🥶</p><h3>😎 What You Can’t See in the Dashboard, Can Break You</h3><p>On paper, everything looked fine. Dashboards showed renewal rates, agent close-rate, task volumes. But something wasn’t adding up and we couldn’t see it from the top.</p><p>It began with a small discovery: a batch of customers who never received their SMS reminders. No one noticed because no one traced it. <strong>That single gap triggered a chain reaction</strong>: missed conversions, wrong priced in follow-ups, and conflicting task records across system. The dashboard told one story. <strong>But the data behind it, the part we weren’t watching — was silently break our flow.</strong></p><blockquote>“A dashboard tells you what happened. But if you don’t understand the data underneath, you’ll never know why.” When we looked deeper, the data wasn’t telling the same story as the UI.</blockquote><h3>📱It started with a missing phone number.</h3><p>We discovered that a segment of customers never received their SMS. At first, we thought it was a deliver issue. But when traced the records upstream, we found something worse:</p><ul><li><strong>Some customer profiles weren’t properly linked to contact information.</strong></li><li>No phone number = No SMS = No chance to renew</li></ul><p>This seemingly small error meant:</p><ul><li>Lost conversion opportunities</li><li>Increased manual follow-up</li><li>Lower self-renewal rate, the very thing automation was supposed to improve.</li></ul><p>And most importantly — <strong>no one saw it coming.</strong> Because the dashboard only showed “SMS sent” count — <strong>not who got missed.</strong></p><h3>💸 Then came the wrong prices.</h3><p>After fixing the linkage, we expected smooth outcomes. But suddenly, our support team began getting calls from confused customers.</p><blockquote>Why is the renewal price in the SMS so weird (too high, too low) from what I was quoted last year?</blockquote><p>We found the issue in the logic. The system generating SMS pulled pricing from a field that wasn’t consistency mapped across all plans.</p><p>This wasn’t just a data mistake. It was a <strong>break in trust </strong>which is much harder to rebuild than code.</p><h3>📊 And then the reports stopped matching.</h3><p>We later discovered that even when a customer <strong>did</strong> renew, and an agent did close the deal in Salesforce, our <strong>back-office system didn’t always reflect it.</strong></p><p>That led to:</p><ul><li>Duplicated or missed tasks</li><li>Miscalculated commissions</li><li>Conflicting performance reports</li><li>And even potential fraud that had gone unnoticed</li></ul><p>Why? Because <strong>no one was tracking the flow end-to-end. </strong>Each system had a “dashboard” but none of them were speaking the same language.</p><h4>⛓️‍💥The damage wasn’t technical. It was systemic.</h4><p>🐛There weren’t just bugs. They were a sign that we’d build automation <strong>on top of flawed assumptions about data visibility.</strong> And <strong>what didn’t show up on the dashboard, </strong>the missing joins, unvalidated mappings, silent sync failures , <strong>had the power to break the very system we were trying to scale.</strong></p><h3>We Didn’t Add More Code, We Made the System Smarter with Data. 👩🏻‍💻</h3><p>When we saw the layers of failure missing SMS links, broken quote mapping, inconsistent reporting , the obvious response might have been to add more checks, more code, or more people.</p><p>But we paused and asked a different question:</p><blockquote>What if the system was wrong because the data model was incomplete?</blockquote><p><strong>We did’t fix the dashboard. </strong>Most people think of data as something you renew things happen. You wait for the report. You look at a chart. You analyze a trend. <strong>But we use data differently.</strong></p><ul><li><strong>We used it to prevent things from breaking in the first place.</strong></li><li>We didn’t throw more code at the problem.</li><li>We didn’t add headcount.</li><li>We changed how the system behaved by changing what the system knew.</li><li>We made sure that each process step had just enough intelligence built into self-correct or raise a flag.</li></ul><p>Each of those was powered by data, hot hardcoded conditions.</p><h4>No new tools. No new teams. Just better logic, better design. ✅</h4><p>Each change was deceptively small. But it was made with one goal: to turn scattered logic into an integrated, self-aware system. And then changed everything</p><blockquote>The errors stopped.<br>The quotes matched.<br>The SMSs made sense.<br>And the business stopped firefighting and started trusting the system.</blockquote><h3>😊 Coming Next</h3><p>We though fixing the SMS system would be simple. Until we found that <strong>~two out of three customers never received their renewal reminders </strong>and root cause wasn’t the tools. <strong>It was missing data.</strong></p><p>We followed was more than just a fix.</p><blockquote><strong>It took just 3 lined of schema redesign and that change helped unlock over a million baht in retained revenue every month.</strong></blockquote><p>In the <a href="https://medium.com/@gympornpimol/designing-for-data-integrity-when-good-automation-fails-bad-data-a3a6263aec60">next post</a>, I’ll breakdown:</p><ul><li>The silent data field that caused it all.</li><li>How we rebuilt our mapping logic to prevent future breaks.</li><li>And the three data validation layers that made the system smarter, learner, and scalable.</li></ul><p>🔗 <a href="https://medium.com/@gympornpimol/designing-for-data-integrity-when-good-automation-fails-bad-data-a3a6263aec60">Link for the next post</a></p><p><strong>😉<em>“Good data doesn’t just describe the past. It designs the futures.”</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AZNHH8DCq3VN-nDX8yYr7w.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6797ba4c0a57" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>