<?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 Fahril Ahsan on Medium]]></title>
        <description><![CDATA[Stories by Fahril Ahsan on Medium]]></description>
        <link>https://medium.com/@f-ahsan?source=rss-fcbf640af245------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Tc9fvN2vx-jXdwRF8djKJg@2x.jpeg</url>
            <title>Stories by Fahril Ahsan on Medium</title>
            <link>https://medium.com/@f-ahsan?source=rss-fcbf640af245------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 16:36:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@f-ahsan/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[Three Questions to Ask Before Trusting Any Feature Impact Analysis]]></title>
            <link>https://f-ahsan.medium.com/three-questions-to-ask-before-trusting-any-feature-impact-analysis-1ff03d9ad458?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ff03d9ad458</guid>
            <category><![CDATA[causal-inference]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[product-analytics]]></category>
            <category><![CDATA[experimentation]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Sun, 09 Aug 2026 03:29:55 GMT</pubDate>
            <atom:updated>2026-08-09T03:30:25.616Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XwunNlJyY-aPnZGALDUHcA.png" /><figcaption>The same feature impact claim, before and after being run through three questions. The confident version has no comparison group, no timeframe, and no way to be proven wrong. The qualified version names all three</figcaption></figure><p>As usual, once I finished the analysis, I dropped the result into our product Slack channel. “Huge win, the new feature drove 30% higher retention among users who saw it.” A few rocket and fire emojis showed up within minutes. I almost added one myself too 🔥.</p><p>Then the familiar voice showed up. It never shows up before I hit send, only after.</p><blockquote>“Are these numbers actually true, or did I just find my most engaged users and call it a win?”</blockquote><p>I work in product analytics, and a lot of my job is sitting between excitement and reality, usually with myself first. Most feature impact analyses fail one of three sanity checks. Not because the analyst is sloppy. I’ve been plenty sharp about numbers that still turned out wrong. It’s just that the questions aren’t culturally normalized. Nobody asks them out loud, so nobody answers them in the analysis, including me, half the time.</p><p>This article is the three questions I now ask myself before I let that Slack message go out.</p><h4>Why this matters</h4><p>Most teams don’t have a culture of stress-testing impact claims. Wins get celebrated, kills get debated, but the analysis behind them rarely gets challenged. The PM trusts the analyst. The analyst trusts the dashboard. The dashboard trusts the data. Nobody questions the chain.</p><p>That’s how you end up with a portfolio of “successful” features that didn’t actually move the business.</p><p>The three questions below don’t require a stats PhD. They require a habit. Apply them consistently and you’ll separate the wins that hold up from the wins that fall apart at the next review cycle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yXzjn4FHE69KIz9fiBL2IQ.png" /><figcaption>A claim has to survive three checkpoints before it’s trustworthy. Most claims that enter don’t make it to the other side</figcaption></figure><h3>Question 1: Compared to what?</h3><p>This is the most fundamental question, and it’s the one people skip most.</p><p>I keep seeing this pattern over and over. Compare users who used the feature vs users who didn’t, and whoever comes out ahead is ‘the proof.’ Case closed.</p><p>The problem is those two groups were never going to look the same to begin with.</p><p>Let’s say a streaming platform ships something called Mood Picker. You tell it what mood you’re in, it recommends shows to match. A month in, the numbers look great:</p><ul><li>Users who tried Mood Picker: <strong>80% retained</strong></li><li>Users who never tried it: <strong>55% retained</strong></li></ul><p>A 25 point gap</p><p>Slack message gets sent. Fire emojis were lit.</p><p>But who tried Mood Picker? The most engaged users. The ones who explore the app, tap on new things, watch more shows. They were always going to stay subscribed at higher rates. The feature didn’t make them loyal. It found them because they were already loyal.</p><p>Fix it by defining your comparison group carefully. Here are a few options, from least to most rigorous:</p><p><strong>Regression with pre-feature behavior as controls.</strong> Cheap. Usually good enough. You include things like prior weekly active days, prior watch hours, tenure on the platform as controls. The coefficient on “used the feature” is your adjusted effect. If it shrinks to nothing, the feature was a passenger.</p><p><strong>Propensity matching.</strong> A step up. For each feature user, find a non-user who looked identical before launch. Compare retention within those matched pairs. Cleaner intuition for stakeholders. More setup work.</p><p><strong>Randomized exposure.</strong> The gold standard. Show the feature (or a promotion of the feature) to a random half of eligible users. Compare outcomes regardless of who actually used it. Rarely done because teams don’t want to “hold back” the feature from anyone.</p><p>If you can’t describe what your counterfactual is, you don’t have an analysis. You have a comparison. Those are not the same thing.</p><h3>Question 2: Is the outcome decision-relevant AND decision-timely?</h3><p>This is the timing question, and it’s the one that creates the most cultural tension.</p><p>(Side note: I wrote about this in more depth in a separate piece. I’ll keep it short here.)</p><p>The lagging outcomes that actually matter like retention, lifetime value, or churn take months to mature. The fast outcomes like clicks, adoption, or session counts are easy to game and easy to misread.</p><p>Most analyses fall into one of two pits:</p><ul><li><strong>Measuring something that matters but won’t mature in time.</strong> Honest but useless. The team will make the decision without you.</li><li><strong>Measuring something fast but irrelevant.</strong> Quick but misleading. The team will make the wrong decision because of you.</li></ul><p>The way out is leading proxies. These are behaviors that predict the lagging outcome but show up fast enough to catch inside your decision window. For a streaming platform, that might be weekly active days, content completion rate, or session frequency in the first 30 days. You check these against historical data first, to confirm they actually predict retention. Then you measure the feature’s impact on the proxy, not on retention itself.</p><p><strong>The rule</strong>: the right outcome is the one that’s both decision-relevant and decision-timely. If your outcome is one without the other, you’re either too slow or too superficial.</p><h3>Question 3: What would convince you that you’re wrong?</h3><p>This is the question that could break your own result, and it’s the one I see asked the least.</p><p>Every analysis has a hypothesis (“the feature drove retention”). Most analyses skip the part where you commit, in advance, to what would falsify it. That’s how you end up with motivated reasoning. You see the result, you decide what counts as “real,” and the answer is always whatever you wanted.</p><p>Three tools to fix this:</p><p><strong>Negative controls.</strong> Pick an outcome the feature couldn’t possibly have caused, and check if it looks “lifted” too. The classic move: look at users who first tried Mood Picker only in month 3 of their subscription, after they had already shown high weekly activity in month 1 and month 2. The feature couldn’t have caused activity that happened before the user ever touched it. If those “future Mood Picker users” already looked different from non-users back in month 1, then your “lift” is mostly selection bias, not the feature. This is one of the most underused tricks in product analytics.</p><p><strong>Placebo periods.</strong> Run the exact same analysis on a time window before the feature existed. If you “find” the same effect in a period when the feature literally couldn’t have done anything, your method is finding ghosts. Useful for validating that your matching, your controls, your model isn’t manufacturing signal out of noise.</p><p><strong>Pre-registration.</strong> Decide what counts as a real effect before you look at the data. Write it down. “We will consider this feature a win if the proxy lift is at least X points, in at least Y subgroups, surviving Z robustness checks.” If you let yourself decide what counts as a win after seeing the numbers, you will always find a win. Humans are remarkably good at that.</p><p>A good analysis tells you in advance what would make you change your mind. If your analysis can’t lose, it can’t really win either.</p><h3>What this looks like in practice</h3><p>Let’s put the three questions on the Mood Picker example.</p><p><strong>Compared to what?</strong> Not “users who didn’t try it.” Instead, users who didn’t try it but looked similar before launch on weekly active days, tenure, and watch hours. Or even better, a randomized rollout where we promote Mood Picker to half the eligible users and compare outcomes regardless of usage.</p><p><strong>On what outcome?</strong> Not “30-day retention” alone. That’s both noisy at one month and biased by who self-selected. Instead, weekly active days as a leading proxy, validated against historical cohorts, measured in the first 2 to 4 weeks.</p><p><strong>What would convince me I’m wrong?</strong> Pre-register: “I’ll call Mood Picker a real win if the proxy lifts by 1+ days per week, holds for low-engagement users in stratification, and survives a negative control check.” If the negative control shows that future Mood Picker users were already different from non-users before they ever used the feature, the whole story falls apart and I have to rewrite it.</p><p>That’s three questions. Maybe two hours of extra work for a meaningful feature launch. The team gets a real answer instead of a celebration that quietly evaporates by Q3.</p><h3>The ugly truth</h3><p>Most analyses I’ve shipped failed at least one of these questions.</p><p>The checklist isn’t about gatekeeping or making anyone feel bad. It’s about making the failure visible, so the team can decide where it’s worth digging deeper.</p><p>If you have one hour, just ask Question 1. Compare apples to apples, even crudely.</p><p>If you have a day, add Question 2. Pick a proxy that matures fast enough to matter.</p><p>If the stakes are real (you’re killing a feature, you’re investing more, you’re showing the deck to a VP), add Question 3. Pick something that could falsify your story, and check it.</p><p>You don’t need all three every time. You just need to be honest about which ones you skipped.</p><h3>Closing</h3><p>I’ve made this mistake myself. Early in my career, I linked a feature to subscriber retention, ran a matching approach to make the comparison feel more credible, and wrote up the gap as a win. What I hadn’t asked was who chooses to try a new feature in the first place. Usually, it’s the people who were already going to stick around. My “feature effect” was mostly a photograph of who was already engaged, dressed up as a causal story.</p><p>Nobody caught it in the room. The number looked clean, the method sounded solid, and it told the team what they wanted to hear. A slower pass later showed most of the lift had nothing to do with the feature.</p><p>The teams that ask these questions out loud, in review meetings, build better products. The teams that don’t end up with a graveyard of “successful” features that didn’t move the business, and analysts, myself included, who trusted a number longer than they should have.</p><blockquote>An impact number you can’t stress-test isn’t a result. It’s a hope.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ff03d9ad458" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Feature Adoption is a Popularity Contest]]></title>
            <link>https://f-ahsan.medium.com/your-feature-adoption-is-a-popularity-contest-fa3b8e364710?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/fa3b8e364710</guid>
            <category><![CDATA[impact-measurement]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[product-analytics]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Mon, 15 Jun 2026 04:06:00 GMT</pubDate>
            <atom:updated>2026-08-09T01:24:08.666Z</atom:updated>
            <content:encoded><![CDATA[<p>There’s a moment that every product analyst knows.</p><p>Your team ships a feature. A month later, the dashboard looks great. People who used it stayed subscribed at much higher rates than people who didn’t. Adoption is healthy. You start drafting the win memo.</p><p>Then a quiet voice in your head:</p><blockquote>Are these people sticking around <strong>because of the feature</strong>, or because they were already the kind of people who stick around?</blockquote><p>I’ve been in that moment many times. It’s not paranoia. It’s the haunting question of product analytics. And the worst part isn’t that the question is hard. The worst part is that you can’t ignore it, and you also can’t take six months to answer it. Your product team has OKR review in a few weeks.</p><p>This is the story of how I stopped trusting feature adoption dashboards. Not because they were lying to me on purpose, but because the whole setup was designed to mislead.</p><h3>The shape of the problem</h3><p>Most feature impact analyses start with the wrong outcome.</p><p>For a streaming platform, the outcomes that actually matter are monthly retention, churn rate, lifetime value. These are the numbers leadership cares about. They’re also the numbers that take months to settle. If you launched a feature in March, you might not have a clean retention picture until September.</p><p>So what does the team default to? <strong><em>Adoption</em></strong>. Daily active users on the feature. Click rates. The number of times someone opened the new Watchlist tab. The only thing that matures fast.</p><p>And here’s the thing about adoption dashboards: they are basically a popularity contest among your already-engaged users. The people who try a new “Smart Watchlist” feature in week one are the same people who try every other feature in week one. They were always going to stay subscribed. They didn’t need the feature to be loyal.</p><p>That’s the trap. Slow correct outcome plus fast wrong outcome equals decisions made on the wrong signal. The analyst gives the team an adoption number. The team treats it like an impact number. Nobody is intentionally cheating. <em>Everybody is fooling themselves</em>.</p><h3>Why adoption dashboards lie to you</h3><p>Let me make this concrete.</p><p>Imagine a streaming platform ships a feature called Smart Watchlist. The platform suggests shows you might want to save, you tap a heart, and it lands in your queue. After 30 days, the data looks beautiful:</p><ul><li>Users who used Smart Watchlist at least once: 85% retained in the next month</li><li>Users who never touched it: 60% retained in the next month</li></ul><p>A clean 25 point gap</p><p>The Product Manager is delighted. The slide goes into the deck. The decision is made to invest more in the feature.</p><p>Now look at it differently. Who actually used Smart Watchlist in the first month? Almost certainly the most engaged users on the platform. The people who open the app five times a week. The people who finish three shows a month. They were always going to come back next month. They’ve been coming back for years.</p><p>The feature didn’t lift them. It just labeled them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s8xYM6MjwnnzAU48gnfxdw.png" /><figcaption>Feature usage - retention correlation example</figcaption></figure><p>The dose-response curve makes it worse. You bucket users by usage (1 time, 2 to 5 times, 6+ times) and you see a beautiful upward staircase. More usage, more retention. Looks like proof. But that staircase might just be tracking how engaged the user already was. Heavy app users use everything more. They retain more. The feature is a passenger in that car, not the driver.</p><h3>The real fix isn’t a better method</h3><blockquote>Most analysts who hit this wall reach for fancier statistics. Propensity matching. Difference-in-differences. Regression with controls. Survival analysis.</blockquote><p>All of those help. None of them fix the timing problem.</p><p>You can do the cleanest causal inference work in the world and still be stuck waiting six months for retention data to mature. By the time you have a clean answer, the product team has already moved on, killed two other features, and shipped three more.</p><p>The fix isn’t statistical. It’s structural.</p><p>Build a two-layer measurement system:</p><ul><li><strong>Leading proxies</strong> for decisions. They mature in weeks. They predict the outcome you actually care about.</li><li><strong>Lagging outcome</strong> for confirmation. It matures in months. It settles the debate after the decision has already been made.</li></ul><p>Stop trying to answer “<strong>did this feature lift retention in a month?</strong>” That question is unanswerable in a month.</p><p>Start answering “<strong>did this feature lift the behaviors that historically predict retention, in a month?</strong>” That question is answerable.</p><h3>How to build the proxy layer</h3><p>This is the part most analysts skip. Picking a proxy isn’t guessing. It’s a small modeling exercise that pays off forever.</p><p><strong>Step 1: Find behaviors that historically predict your lagging outcome.</strong></p><p>On historical data, build a simple model where the outcome is “did this user stay subscribed to month 2” and the inputs are short-window behaviors:</p><ul><li>weekly active days</li><li>total watch hours in the last 14 days</li><li>content diversity</li><li>session frequency</li><li>completion rate</li></ul><p>A logistic regression is fine. You don’t need anything fancy. Look at which features have the strongest signal.</p><p><strong>Step 2: Validate the proxy before you trust it.</strong></p><p>Take a cohort that has already matured. Compare its proxy values against its actual retention. If the proxy is solid, the relationship should hold across multiple cohorts. If it only worked once, it’s noise. If it works consistently across four or five cohorts, you have a real predictor.</p><p><strong>Step 3: Now measure the feature’s impact on the proxy.</strong></p><p>Not on retention. On the proxy. Did adopters lift their weekly active days relative to matched non-users within 2 to 4 weeks of using the feature? That’s actionable inside your OKR window. That’s the number you bring to the team.</p><h3>The nuance most people miss</h3><p><strong>Even with proxies, you still have the selection problem</strong>. The Smart Watchlist users are still the engaged users. The proxy is shorter, but the bias is the same.</p><p>The difference is that the bias is more tractable on a short outcome.</p><p>Stratify by pre-feature behavior. Split users into low, medium, and high engagement based on their behavior before the launch. Look at the proxy lift within each stratum separately.</p><p>If the proxy lifts for low-engagement users, the feature is probably doing something real. It’s pulling people up who wouldn’t have lifted on their own.</p><p>If the proxy only lifts for high-engagement users, the feature is just labeling. It’s identifying who was already going to be fine.</p><p>One more diagnostic worth doing: look at users who used the feature for the first time after they had already shown the behavior you care about. The feature couldn’t have caused that behavior, the behavior came first. If those users still look “lifted” relative to non-users, you have unmeasured selection bias and your story is shakier than you think. I’ll write about this trick in more detail in a follow-up piece.</p><h3>What to actually say in the meeting</h3><p>Here’s the payoff.</p><p>The way you talk to leadership changes. You stop choosing between two bad options.</p><p><strong>The bad option: </strong>“Smart Watchlist lifted retention by 25 points.” Premature. Almost certainly wrong. Will embarrass you later when retention data matures and the number isn’t there.</p><p><strong>The useless option:</strong> “We don’t know yet. Ask me in 6 months.” Honest, but unhelpful. The team can’t act on that. They will go make the decision without you, probably using the bad option.</p><p><strong>The honest option: </strong>“30-day proxy data shows a 4 point lift in weekly active days, concentrated in users who weren’t already heavy app users. The lagging retention confirmation comes in month 4. Based on this, I’d keep the feature, kill the low-impact variant, and watch the third one. If the proxy lift doesn’t hold by month 2, we revisit.”</p><p>That’s a position the product team can actually act on. It’s not certain. It doesn’t pretend to be. But it’s grounded in something stronger than an adoption dashboard, and it shows up inside their decision window.</p><h3>What I learned (in a hard way)</h3><p>I used to think the gap between “good analysis” and “fast analysis” was a tradeoff you had to make. Either I gave the team a number they could act on now, or I gave them a number that was actually true.</p><p>It turns out that was a false choice. The trade was real because I was anchored to the wrong outcome metric. Once I let go of “we have to measure retention directly within a month,” the rest got easier.</p><p>Better analysis isn’t always about better statistics. Sometimes it’s about picking a metric that matures inside your decision window.</p><p>The best causal method in the world won’t save you if you’re measuring the wrong thing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fa3b8e364710" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[One Query, Many Charts: The Dashboard Practice That Changed How I Work]]></title>
            <link>https://f-ahsan.medium.com/one-query-many-charts-the-dashboard-practice-that-changed-how-i-work-38ee2e9cff10?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/38ee2e9cff10</guid>
            <category><![CDATA[semantic-layer]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[dashboard]]></category>
            <category><![CDATA[sql]]></category>
            <category><![CDATA[analytics-engineering]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Thu, 04 Jun 2026 04:01:02 GMT</pubDate>
            <atom:updated>2026-08-09T01:25:03.414Z</atom:updated>
            <content:encoded><![CDATA[<p>I used to build dashboards the obvious way. One chart, one query. Retention rate overall? Write a query. Monthly trend? Another query. Breakdown by store? One more. By gender? Sure, add that too.</p><p>It felt clean. Each query had one job. Everything was neat and separate.</p><p>Until the business changed something.</p><p>A column got renamed. A new store got added. The definition of “active customer” shifted from 30 days to 45 days. And suddenly <strong>I had to open every single query and make the same update, again and again</strong>, hoping I didn’t miss one. Sometimes I did miss one. And then two charts on the same dashboard showed different numbers for the same metric, and someone in a meeting noticed.</p><p>That’s the quiet frustration nobody really talks about when they discuss BI work.</p><h3>The Fix: One Query as the Source of Truth</h3><p>The practice I landed on is simple in concept. Instead of writing a separate query for each chart, you write one query that computes everything the dashboard needs, across all the dimensions you care about. Every visualization just becomes a filtered view of that single result.</p><p>Update the logic once, and the entire dashboard reflects it. No hunting. No inconsistencies. No awkward moment in meetings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RD8KxFnE6RasBGxJKZgTRA.png" /></figure><p>The hard part isn’t writing this kind of query. The hard part is designing it so it actually works cleanly for multiple visualizations at once. That’s where the structure matters.</p><h3>The Metric-Value Pattern</h3><p>The technique that makes this work in practice is something I call the metric-value pattern. Instead of returning wide output with separate columns for every breakdown, the query returns a long-format result with two key columns.</p><p>The first column, metric, tells you which dimension or breakdown this row represents. Something like overall, store, gender, or age_group. The second column, value, holds the specific member of that dimension. Store A, Male, 26-35, and so on. Then a third column holds the actual number you&#39;re tracking, retention_rate or whatever the measure is.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jGk0xRucXGsdxwtoR2R_lQ.png" /><figcaption>output sample</figcaption></figure><p>In your dashboard tool, each chart is just a filter on top of this result. The overall KPI card does WHERE metric = &#39;overall&#39;. The store breakdown does WHERE metric = &#39;store&#39;. The gender chart does WHERE metric = &#39;gender&#39;. All from the same query, all by definition using the same business logic.</p><p>Adding a new breakdown is also just adding a new UNION ALL block in the query. The dashboard can immediately start using the new metric value. No new query to write, no new logic to maintain in parallel.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-d7fxyScg6UoXbTKPmeOdQ.png" /><figcaption>query structure</figcaption></figure><h3>When to Run Live vs. When to Materialize</h3><p>One thing that makes this pattern actually usable in production is knowing when not to run the query live on every dashboard load.</p><p>The decision is pretty straightforward. If the query is lightweight, run it live. The dashboard always shows fresh data and there’s nothing extra to maintain. If the query is expensive, scanning large tables or joining across multiple sources, you materialize it. Run it on a schedule and write the result to a dedicated table. Your dashboard queries the materialized table instead of the raw sources.</p><p>This isn’t a workaround or a compromise. It’s the right engineering call depending on scale. The one-query pattern works either way, you just choose how to deploy it based on cost and freshness requirements.</p><h3>The Real Benefits</h3><p>The main win is <strong>maintainability</strong>. When business logic changes, you touch one place and the whole dashboard updates. That sounds obvious until you’ve spent an afternoon hunting down six queries that all needed the same fix.</p><p>Consistency follows automatically. When all charts come from the same query, they can’t disagree with each other. The numbers on the KPI card and the numbers on the breakdown chart are always computed the same way because they literally come from the same rows.</p><p>There’s also a scalability benefit that shows up over time. As the business adds new dimensions you want to track, like a new customer segment or a new product category, each one is just a block. The dashboard structure doesn’t need to change.</p><h3>What the Industry Actually Calls This</h3><p>Here’s where it gets interesting, because this isn’t just a personal trick. What I’ve been describing is actually a well-established concept in the data industry. It goes by a few names, but the most common ones are <strong>Single Source of Truth (SSOT)</strong> and <strong>Semantic Layer</strong>.</p><p>The core idea behind <strong>SSOT</strong> is that every metric, every business definition, should be mastered in exactly one place. Not defined in ten different SQL queries across ten different dashboards. One place, and everything else reads from that place. When it changes, it changes everywhere at once.</p><p>A <strong>Semantic Layer</strong> takes this a step further. It’s an abstraction layer that sits between your raw data and all the tools that consume it, dashboards, reports, spreadsheets, notebooks, whatever. You define your metrics once inside the semantic layer, and every downstream tool just requests them by name. The layer handles the SQL.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oFrLJ3rlj-eZHkvMMpBXYg.png" /></figure><p>The reason this matters is consistency at scale. In a typical company without a semantic layer, the analyst building the dashboard defines retention rate one way. The analyst building the executive report defines it slightly differently. The data scientist computing it for their model uses a third definition. Nobody is wrong exactly, they just each made slightly different judgment calls. And now the company has three versions of the same metric, and when someone asks “what’s our retention rate?”, the answer depends on who you ask.</p><p>A semantic layer eliminates that. There is one definition. Everybody uses it. If the definition changes, it changes everywhere, immediately.</p><p>Tools like dbt’s Semantic Layer and LookML inside Looker are built entirely around this principle. You write the metric definition once in YAML or LookML, and it flows through to Tableau, Looker, Google Sheets, whatever BI tool you’re using. The SQL is generated automatically from the definition. Analysts stop copy-pasting logic and start referencing shared, governed definitions instead.</p><p>What I’ve been describing in this article is a SQL-native version of the same idea, applied within a single dashboard. You don’t need a semantic layer tool to benefit from SSOT thinking. The discipline of centralizing your metric logic, making it the single place where retention rate is defined for this dashboard, is the same principle, just implemented with UNION ALL instead of YAML.</p><p>The practical difference is scope. Enterprise semantic layer tools work across your entire data stack, across every tool, every team, every consumer. The single-query approach works within one dashboard. But if you’re a BI developer who doesn’t have a semantic layer in your stack yet, applying this pattern at the dashboard level is genuinely the right instinct. You’re doing what the industry eventually builds formal tooling for.</p><p>One useful way to think about it: a semantic layer is what happens when a company decides that the single-query discipline matters so much, it should be infrastructure, not just practice. The practice comes first. The infrastructure formalizes it.</p><h3>When This Doesn’t Make Sense</h3><p>To be fair, there are cases where this pattern doesn’t fit.</p><p>If two charts on your dashboard have genuinely different grain levels, one showing order-level detail and another showing monthly aggregates, combining them into one query gets awkward fast. The join logic gets complicated and the result is harder to reason about. Keep them separate.</p><p>If parts of your dashboard are owned by completely different teams with different update cycles and different business logic owners, shared queries introduce coordination overhead that might cost more than it saves. Independent queries can be the right call.</p><p>And if two metrics are just unrelated, no shared tables, no shared filters, no shared logic, there’s no benefit to forcing them together. The pattern is about sharing logic that’s genuinely common, not about combining arbitrarily.</p><h3>How to Start</h3><p>If you want to try this on an existing dashboard, the approach is pretty practical.</p><p>Start by mapping all your charts before writing any SQL. List every visualization and identify which dimensions they share. If they share the same base logic, they’re good candidates for a single query.</p><p>Design the output schema before writing the query. Decide what metric values you need and what value represents in each case. The schema drives the structure, and it&#39;s much easier to reason about upfront than midway through writing UNION ALL blocks.</p><p>Use CTEs to keep the base logic clean and separate from the breakdown blocks. A well-structured WITH clause for joins and filters makes each UNION ALL block easy to read and modify on its own.</p><p>Document your metric values somewhere. Even a comment in the query itself works. Future you, and your teammates, will be grateful.</p><p>And if you know the tables are large, plan for materialization from the start. It’s easier to design a scheduled query upfront than to add it as an afterthought when the dashboard starts getting slow.</p><h3>The Bigger Point</h3><p>The best dashboards aren’t just beautiful to look at. They stay accurate as the business changes, without needing someone to go in and update ten queries every time a definition shifts.</p><p>The single-query pattern is one of the cheapest ways to build that kind of durability. It asks you to think about the complete shape of a metric before you start building, rather than treating each chart as a separate problem. That shift is small. The payoff compounds over time.</p><p>And if you ever end up at a company that has a proper semantic layer, you’ll already understand why it exists. Because you’ve been doing the same thing manually, one dashboard at a time.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=38ee2e9cff10" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Was Tired of Doing Admin for My Own Villa. So I Built a Bot.]]></title>
            <link>https://f-ahsan.medium.com/i-was-tired-of-doing-admin-for-my-own-villa-so-i-built-a-bot-0b6ef9400e0c?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/0b6ef9400e0c</guid>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[telegram-bot]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[automation]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Wed, 03 Jun 2026 03:53:13 GMT</pubDate>
            <atom:updated>2026-06-04T02:09:41.768Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kof6r16xaV4ip6lJv3IL2Q.jpeg" /></figure><p>I’ve been a data analyst for 5 years. Mostly analysis, dashboards, decks that live in someone’s Google Drive and get opened twice. You know the drill.</p><p>But outside of my 9-to-5, I run a villa. End-to-end. Marketing, operations, guest communications, pricing strategy, admin, finances. All of it. I’m basically a one-person company where the product happens to have a pool.</p><p>And like any one-person company, I have a list of things I hate doing. At the very top of that list: repetitive admin work.</p><h4>The Problem (aka I’m Just Lazy)</h4><p>Every booking follows the same pattern. A guest sends a booking template via WhatsApp. I open it, manually copy the details somewhere, then open another tool to generate an invoice, format it, export it as PDF, and send it back.</p><p>To make it worse, here’s what the actual flow looked like every time:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HIct0qC9ygiFBFSQ00TScg.png" /><figcaption>before</figcaption></figure><p>1. Guest sends booking template on WhatsApp<br>2. I read it and manually type the details into Google Sheets<br>3. Open the invoice app<br>4. Re-enter the exact same data into the invoice app<br>5. Export as PDF<br>6. Send it back to the guest on WhatsApp</p><p>That’s the same data, entered twice, by hand, every single booking. Not because I had to. Just because I never sat down to fix it.</p><p>I looked at invoice apps. There’s no shortage of them. But they’re either too generic, too expensive for what I actually need, or they solve the wrong problem. None of them do what I specifically needed: take a booking template, extract the data automatically, log it, and spit out an invoice. So I kept doing it manually. For a long time. Like a person who complains about the dishes every night but never buys a dishwasher.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/236/1*1wnxhlY4eCznV2fsOZqYSw.png" /></figure><h4>My First Idea Was Boring (and Wrong)</h4><p>My first instinct was to build a web form. A nice little data input page where I could fill in the guest details and hit submit. Clean, standard, very “proper software engineer” of me.</p><p>But then I thought about it for more than five minutes and realised… I already live in Telegram. My whole operational communication runs through it. I use it to store important messages, booking records, notes to myself. Opening a separate browser tab just to do data entry felt like solving a problem by creating another one.</p><p>So I asked myself a different question. What if the interface was just a chat?</p><p>The workflow I had in mind: when a booking comes in on WhatsApp, I already forward the template to my Telegram for record keeping. Now instead of forwarding it to a personal chat, I could forward it directly to a bot. The bot reads it, extracts what it needs, logs it to a spreadsheet, and generates an invoice. No forms. No tabs. No copy-pasting.</p><p>That felt right.</p><h4>The Random Evening That Started Everything</h4><p>I wasn’t planning to build anything. It was just a regular weekday evening and I was messing around with Python like I usually do these days. I’ve been practicing OOP recently as part of trying to move toward a more engineering-focused role, so I often just open a file and start playing with something to keep the momentum going.</p><p>That night I started thinking about the bot idea again. I opened a chat with Claude, typed something like “let’s build a Telegram bot that reads booking messages”, and just kept going.</p><p>It’s funny how vibe coding works. You don’t plan it, you don’t block out time for it, you’re not trying to build a portfolio piece. You’re just curious. And then four hours later you look up and there’s something real in front of you.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tn69efPVcfHNgcXZogsNug.png" /></figure><h4>What I Actually Built</h4><p>Okay here’s where it gets a little <em>fancy</em> and I’m not gonna pretend otherwise. It’s called <strong>Chita</strong> (Chat-based Invoice &amp; Transaction Input Automation).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*izUGA8qUdjIPv6nbJRDLPg.png" /><figcaption>after</figcaption></figure><p>The full pipeline looks like this:</p><p>A booking comes in on <strong>WhatsApp Business</strong>. I forward the message to Telegram, and instead of saving it to a personal chat like I used to, I forward it straight to the bot.</p><p><a href="https://platform.claude.com/"><strong>Claude API</strong></a> handles the intelligence part. When the booking message arrives, Claude reads it and extracts the structured data: guest name, check-in and check-out dates, room type, price, whatever I need. No regex, no brittle parsing rules. Just natural language understanding that actually works on messy, inconsistent booking templates.</p><p><strong>Google Sheets</strong> acts as the database. Every extracted booking gets logged as a new row. Simple, visible, shareable.</p><p><strong>PDF Generator</strong> takes the extracted data and produces a proper invoice. Formatted, branded, ready to send. I save it for my own records and forward it manually to the guest on WhatsApp.</p><p><a href="https://railway.com/"><strong>Railway</strong></a> is where it all lives. The bot runs 24/7 on a cloud server. My laptop doesn’t need to be open. It just works.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/236/1*o331zFrJ9NY10SNyY2Om_g.jpeg" /><figcaption>chatbot in action</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*moe4LZX6T9NH0klkCVA9mQ.png" /><figcaption>invoice PDF result</figcaption></figure><h4>The Walls I Hit</h4><p>It wasn’t all clean and smooth, because nothing ever is.</p><p>The main one: Railway. First deployment failed. Then the second. Turned out I was missing two tiny files — a `Procfile` and a `runtime.txt` that tell Railway what kind of app it’s dealing with and how to start it. Neither of these show up in most tutorials. You just hit a wall, stare at the logs, and slowly piece together what’s missing.</p><p>Two files. Maybe ten lines total. Forty minutes to figure out they needed to exist.</p><p>These are the moments that don’t make it into the tutorials but are honestly 60% of the actual experience of building something real.</p><h4>It Works. And Honestly That Felt Weird, LOL.</h4><p>The first time the bot responded to a real booking message, extracted the right data, logged it to the sheet, and generated the invoice without me doing anything, I just stared at it for a second.</p><p>Not because it was magical. Just because something I built was running on a server somewhere, doing a job, and my laptop was closed.</p><p>That’s a new feeling for me. Most of the work I do in data lives in notebooks or dashboards. It’s useful but it’s static. This thing is alive in a way that feels different. I can forward a booking at midnight and the bot handles it instantly. No waiting, no manual work, no context switching.</p><p>For someone who has spent five years producing outputs rather than products, that’s a bigger deal than it probably sounds.</p><h4>What I Accidentally Learned</h4><p>A few things clicked during this project that hadn’t fully clicked before.</p><p>OOP made more sense with a real codebase. Reading about classes and objects is fine. Using them to actually organise a bot with different responsibilities — extraction, sheet writing, invoice generation, Telegram handling — is when it sticks.</p><p><strong>Deployment is mostly config, not code</strong>. The code was mostly done in one evening. Getting it to run reliably on Railway took longer because of environment variables, credential management, and all the invisible plumbing that tutorials skip over.</p><p>Building for yourself is how you actually finish things. I’ve started side projects before and abandoned them because the motivation ran out. This one I finished because I genuinely needed it. The laziness was load-bearing.</p><h4>So, Yeah</h4><p>I built this because I didn’t want to copy-paste booking details into a spreadsheet anymore. That was the whole motivation. No grand plan, no portfolio strategy, just an annoying task I wanted to make disappear.</p><p>It turned out to be the most complete thing I’ve built in Python. It uses real APIs, has an actual user interface, generates documents, connects to a database, and runs in production. I’m also practicing OOP right now and this gave me a real codebase to apply it to, which is worth more than any exercise.</p><p>If you’re sitting on some manual workflow you’ve been complaining about for months, maybe just open your laptop one evening and see how far you get. You might surprise yourself.</p><blockquote>The dishes won’t wash themselves. But apparently the invoices will.</blockquote><p>*Built with: Python, Claude API, python-telegram-bot, gspread, reportlab, Railway*</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0b6ef9400e0c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Progress Was Easy to Measure. Impact Was Not.]]></title>
            <link>https://f-ahsan.medium.com/progress-was-easy-to-measure-impact-was-not-1948e4bdd23b?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/1948e4bdd23b</guid>
            <category><![CDATA[impact]]></category>
            <category><![CDATA[self-reflection]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Fri, 26 Dec 2025 10:13:02 GMT</pubDate>
            <atom:updated>2025-12-26T10:13:02.705Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ehux6WClQI_YQRHO.jpg" /><figcaption><em>“An Experiment on a Bird in the Air Pump” </em>by Joseph Wright of Derby (1768)</figcaption></figure><p>Paintings from the early scientific revolution often showed moments of discovery, but they also captured unease. In <em>An Experiment on a Bird in the Air Pump</em>, progress is clearly happening. Knowledge is created in front of the audience. Yet the people in the room react differently. Some watch with curiosity, while others look away, disturbed by what it takes to learn. Even when science promised to change the world, not everyone understood its meaning in the same way. Progress moved ahead, but understanding did not always follow.</p><p>For a long time, progress felt obvious to me too.</p><p>As a kid, it began with PC games. I did not just play them. I wondered how they worked. That curiosity pulled me through graphic design, web coding, small systems, hardware experiments, and eventually into software and data. Each step rewarded effort with visible improvement. I learned something new, built something better, and moved on. Like the observers in that room, I focused on whether it worked and rarely on what it demanded.</p><p>Back then, I never questioned whether this trajectory meant anything beyond growth. Learning itself felt like enough.</p><p>As I moved from creative work to engineering, from graphic design and video editing to building systems, the pattern repeated. I kept chasing leverage, the idea that a small technical decision could scale to many users/audience.</p><ul><li>IoT projects taught me how software could touch the physical world</li><li>Software engineering taught me reliability and structure</li><li>Data analytics taught me how to see behavior at scale</li></ul><p>Each phase expanded my toolkit, and my confidence grew alongside it.</p><p>In hindsight, I was accumulating capability faster than I was accumulating clarity.</p><p>There were values guiding me, even then, but they were quiet. I believed deeply in contributing to progress, in how engineering and science, from the first industrial revolution to today, have reshaped how civilization works.</p><blockquote>I wanted to be part of it, building things that could matter beyond where I was.</blockquote><p>At the same time, my faith taught me that beneficial knowledge carries lasting value. <strong>Knowledge, I learned, should serve others, not just its creator.</strong></p><p>Those beliefs were real. But they were abstract. They didn’t yet demand anything from my daily work.</p><p>That changed when I moved fully into data.</p><p>On paper, it was the perfect role for someone who cared about impact. Data promised objectivity. You could measure outcomes, identify problems, and improve systems efficiently. I worked on user behavior, internal processes, manual work automation, profitability, and business sustainability. I solved problems that made organizations run better. Things improved. Metrics moved.</p><blockquote>And yet, something felt incomplete.</blockquote><p>I realized that measurement had quietly replaced meaning. Impact started to look like dashboards, percentages, and KPIs trending in the right direction. The work was useful, but often in an instrumental way. We optimized flows, reduced friction, increased efficiency. But we rarely paused to ask whether users were actually succeeding in the ways they hoped.</p><p>I wasn’t failing. I was succeeding too cleanly.</p><p>That was the friction I could not ignore. My values, especially the idea of beneficial knowledge, started to act less like inspiration and more like a constraint. They no longer motivated me only to build more, but to question what I was building and why.</p><p>If knowledge is meant to benefit others, then improvement without understanding is not enough. Solving problems efficiently does not guarantee that I am solving the right problems. I learned that optimization can improve numbers while skipping the harder work of understanding how people actually feel and what they truly need.</p><p>This didn’t make me reject data or technology. It made me more demanding of them.</p><p>I began reframing what “impact” meant to me. It wasn’t just about scale or efficiency, but about alignment between:</p><ul><li>what we build</li><li>why we build it</li><li>and how people actually experience it</li></ul><p><strong>Progress that only exists in metrics is fragile. Progress that helps people succeed, even imperfectly, lasts longer.</strong></p><p>Today, I still work in data. I still care about systems, performance, and sustainability. But I’m more suspicious of easy wins. I’m more willing to sit with uncertainty, to ask why before asking how much. I no longer see my career as a straight line of advancement, but as an ongoing negotiation between capability and responsibility.</p><p>I don’t think I’ve figured it out.</p><p>What I do know is this. Progress is easy to measure, usefulness is not. If I am going to contribute to the future I believe in, one shaped by technology that genuinely helps people, I want my work to be judged by the latter, even when it is harder to prove.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1948e4bdd23b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Most Digital Products Aren’t Sustainable]]></title>
            <link>https://f-ahsan.medium.com/why-most-digital-products-arent-sustainable-88ba9e821396?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/88ba9e821396</guid>
            <category><![CDATA[user-behavior]]></category>
            <category><![CDATA[digital-product-strategy]]></category>
            <category><![CDATA[subscription]]></category>
            <category><![CDATA[product-analytics]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Wed, 03 Dec 2025 17:59:42 GMT</pubDate>
            <atom:updated>2025-12-26T09:04:24.050Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/793/1*25Gdv7XT8VwyRJM9x2dYTw.png" /><figcaption>Image from sketchappsources.com</figcaption></figure><p>Every few months, a new product goes viral. Sometimes it’s a snack everyone rushes to try, other times it’s an app that explodes because of “genius” marketing or a compelling discount offer. People show up out of curiosity, take a bite, tap-scroll-swipe around, and then move on. Virality might pull crowds through the door, but it rarely keeps them inside.</p><blockquote>Sustainable businesses run in a different way. <strong>They don’t rely on trends or promotions , they rely on delivering something people return for, not because it’s new, but because it’s genuinely valuable.</strong></blockquote><p>A good example is the difference between a language-learning app someone downloads on impulse and one they stay committed to for months or even years. Most people don’t subscribe because of the ridiculous amounts of promos. They subscribe because they want to see themselves change: speaking more, improving steadily, and gaining the confidence to continue. But from what I’ve seen,<strong> many products stop delivering value the moment the subscription is paid</strong>. They provide content, not direction; features, not outcomes. And this is where the gap between viral interest and sustainable value becomes painfully obvious:</p><blockquote><strong>Curiosity brings users once, but only real progress brings them back.</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IdXxbk-xKXlnWHZK37HXBw.png" /><figcaption>Online Course Landing Page</figcaption></figure><p>It’s the same pattern we see in many other products. What about online courses? people enroll because they picture themselves gaining a skill that could change their career. The landing page promises transformation, the syllabus looks polished, and the testimonials sound convincing. But after the initial excitement fades, many learners find themselves watching videos without truly learning, or completing modules without feeling any more capable than before. The course technically delivers content, yet the learner doesn’t experience progress. And without that progress, commitment drops. When the promise doesn’t match what users feel, their excitement quickly fades.</p><p>When we pay for a digital product, we aren’t buying the interface, the menu options, or the content inside it. We’re buying the outcome we hope the product will help us reach. In language-learning app, nobody subscribes because they think a library of lessons is inherently valuable. They subscribe because they want to be able to confidently order food in Tokyo, talk to a Korean friend, or unlock a new job opportunity. The same dynamic exists in online courses, fitness tools, budgeting apps, or productivity platforms, we are willing to pay because we believe the product will help us get somewhere we can’t reach alone.</p><p>That’s the real nature of value exchange. users give money, and in return they expect movement toward the outcome that made them sign up in the first place. If users don’t feel clearer, better, or more capable over time, the exchange weakens, even if the transaction technically succeeded. That’s why sustainable products focus on outcomes above everything else. The core question shifts from “<strong>What else can we build?</strong>” to “<strong>Are we helping people succeed in the way they hoped?</strong>”</p><h3><strong>The “Bait and Fail” Problem</strong></h3><p>Even though users pay for outcomes, many products still deliver only content. This is where the gap between expectation and experience becomes painfully visible. In a language-learning app, the marketing paints a picture of confidence and fluency. But once inside, users often find themselves completing lessons without feeling any real improvement. The exercises are there, the curriculum is there, but the sense of “I’m actually getting better” is missing. And for most people, that feeling (not the content) is what keeps motivation alive.</p><p>You see the same pattern in countless other digital products. Online courses give access to hours of video but leave learners unsure whether they’re progressing. Productivity apps offer templates and tools but don’t help people build habits that actually stick. In each case, the product technically delivers what was advertised, yet the user doesn’t experience the momentum they were promised. That’s the heart of the bait-and-fail problem. The product sells a transformation, but delivers a checklist. And once people notice that disconnect, no amount of reminders or new features can convince them to stay.</p><blockquote>What makes a business sustainable isn’t the purchase, it’s the progress users experience.</blockquote><p>Acquisition and retention often get talked about as if they belong in the same conversation, but they operate on completely different rules. <strong>Acquisition is about persuasion</strong> — convincing someone that your product might help them reach the outcome they care about. That’s why marketing, pricing, and positioning work so well here. They speak to the dream of fluency, confidence, mastery. But once a user signs up, the dream stops being hypothetical. The question shifts from “Does this look promising?” to “Is this actually helping me?” The moment that shift happens, persuasion loses power and the product itself has to take over.</p><p><strong>Retention</strong>, on the other hand, is brutally honest. It <strong>measures whether the product is delivering enough real progress for people to keep showing up</strong>. No amount of ads or onboarding gimmicks can compensate for a user who isn’t feeling improvement. And this is where many businesses struggle:</p><blockquote><strong>They invest heavily in getting people in, but far less in making sure those people succeed after they arrive.</strong></blockquote><p>Retention isn’t driven by hype or visibility, it’s driven by outcomes. When users feel themselves getting better, even a little, they stay. When they don’t, no discount or push notification will save the relationship.</p><h3>A Real Product Mindset Starts with Pain Points, Not Features</h3><p>A real product mindset begins with understanding what actually holds people back from reaching the outcome they care about. For example learning apps that promise to teach anything from design to coding to public speaking. People start with excitement, hoping to gain a new skill that feels meaningful. But the obstacles they face are often small and familiar:</p><ol><li>lessons that feel too long,</li><li>progress that’s hard to measure,</li><li>routines that are difficult to maintain,</li><li>or moments where they’re unsure whether they’re actually improving.</li></ol><p>These little frictions may seem minor from the outside, but they’re exactly where users decide whether the product is genuinely helping them grow or simply becoming another tool they abandon over time.</p><p>This is why good product teams don’t chase feature lists, <strong>they chase the pains that matter</strong>. Not every frustration deserves attention, and not every request leads to real progress. The job is to figure out which obstacles, if removed, would actually change the user’s experience. That’s where the work becomes strategic by prioritizing the pains that unlock momentum, make learning feel lighter, or help users see their progress more clearly. When a product consistently removes the right barriers, users naturally stay longer. Not because the product traps them, but because it genuinely helps them move forward.</p><h4>How Analytics Helps Prioritize What Actually Matters</h4><p>Once you understand the pains that shape a user’s experience, the next challenge is deciding which ones are worth solving first. This is where analytics becomes more than dashboards or event counts. Good analytics helps us see patterns that individual users can’t describe:</p><ul><li>where motivation drops?</li><li>when people stop coming back?</li><li>which behaviors lead to long-term success?</li><li>and which frustrations quietly push people away?</li></ul><p>Users might say they want more content, but the data might show that most of them stop after lesson three because they don’t feel themselves improving. That insight changes everything, it tells us which problem is truly blocking progress.</p><p>And when analytics and product thinking work together, prioritization becomes clearer. You’re no longer guessing which ideas sound nice or which features are easiest to build. You’re choosing the problems that, once solved, create real momentum for the user. Maybe that means surfacing progress more clearly. Maybe it means tailoring lessons to match the user’s pace. Maybe it means strengthening review cycles so people don’t feel like they’re forgetting everything. The point isn’t the feature, it’s the outcome. Analytics simply shines a light on the places where solving one pain creates the biggest lift in user success.</p><p>If you look at products that survive beyond their initial hype, they almost always share one thing in common:</p><p><strong><em>They invest heavily in helping users succeed after the first day.</em></strong></p><p>You can see it in:</p><ul><li>how language apps use streaks and bite-sized lessons to keep learners coming back</li><li>how productivity tools guide users toward quick wins so they actually feel the benefit of staying organized</li><li>streaming platforms do it with personalized playlists that reduce the friction of finding something you like</li><li>education platforms build spaced repetition systems so learning sticks instead of evaporating overnight.</li></ul><p>None of these examples are just “features”. They’re deliberate systems designed to support user progress to make the outcome <strong>feel more achievable, more visible, and more rewarding</strong>. And the companies behind them understand that:</p><blockquote>Long-term growth doesn’t come from acquiring more users, but from helping the users you already have get what they came for.</blockquote><p>When people feel momentum, they return. When they don’t, the relationship quietly fades. Sustainable businesses know this, which is why they design not for the moment of purchase, but for everything that happens after.</p><p>At the end of the day, sustainability has less to do with how loudly a product launches and far more to do with what happens once the excitement fades. Any business can attract users with a good campaign or a clever promotion. But keeping those users by earning their trust, their time, and eventually their loyalty requires something deeper. <strong>It requires helping them make real progress toward the outcome they cared about from the start.</strong></p><p>That’s why the work of product and analytics teams goes beyond building features or watching metrics move. The real work is understanding where users struggle, choosing which pains matter most, and designing a product that steadily removes those barriers. When a product does that well, users don’t need to be convinced to stay. They stay because they feel themselves getting somewhere. And in a world full of hype-driven launches and short-lived trends, that simple, steady sense of progress is what separates businesses that disappear from those that endure.</p><p>So, which products in your life deliver outcomes, not just features?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=88ba9e821396" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Future-Proofing Your Analytics Dashboard — Why Ugly Query Is the Enemy You Should Avoid]]></title>
            <link>https://f-ahsan.medium.com/future-proofing-your-analytics-dashboard-why-ugly-query-is-the-enemy-you-should-avoid-1b687a064795?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b687a064795</guid>
            <category><![CDATA[query]]></category>
            <category><![CDATA[sql]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[clean-code]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Sat, 06 Apr 2024 17:30:53 GMT</pubDate>
            <atom:updated>2024-04-06T17:30:53.059Z</atom:updated>
            <content:encoded><![CDATA[<h3>Future-Proofing Your Analytics Dashboard — Why Ugly Query Is the Enemy You Should Avoid</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PmrBzr2tfBuji3an6Syb5Q.jpeg" /><figcaption>Image generated by DALL·E</figcaption></figure><p>Have you ever been stuck fixing something in a query for your dashboard, only to run into a mess of code that seems impossible to understand? It’s like trying to solve a puzzle with missing pieces and no picture to guide you. The longer you stare at it, the more confusing it gets, and the harder it becomes to fix anything. Dealing with what we call “ugly code” is frustrating and time-consuming. But it’s not just the code that can be messy — the queries that fetch data for your dashboard can be just as tangled and bewildering.</p><p>Imagine this:</p><ul><li>legacy queries that seem to have been written in another era, with convoluted syntax and outdated methods,</li><li>non-standardized syntax across queries, leaving you scratching your head as you try to make sense of inconsistent structures,</li><li>multi-nested subqueries that resemble a Russian nesting doll, where each layer adds another level of complexity and confusion,</li><li>and unstandardized aliases that leave you guessing about the meaning of each column.</li></ul><p>It’s enough to make any dashboard creator want to pull their hair out. But why does it have to be this way? And why should we care about keeping our code and queries clean and tidy? Let’s dive in and find out why messy code and queries, including these examples, are the last thing any dashboard creator wants to deal with.</p><h3><strong>Why analysts/engineers keep writing ugly codes/queries?</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kLJlYATdHULeLhhvAl0EXg.jpeg" /></figure><p>In many cases, encountering ugly codes and queries stems from a variety of factors, chief among them being lack of experience and absence of standardized practices. When engineers or analysts are relatively new to dashboard creation or lack exposure to best coding practices, they may inadvertently produce messy, convoluted code that is difficult to maintain and understand. Additionally, without established guidelines or standardized syntax within a team or organization, individuals may adopt their own coding styles, leading to inconsistencies across projects and making it challenging for others to interpret their work.</p><p>But why is this lack of cleanliness such a big deal? The consequences may not be immediately apparent, especially to those new to the field or unfamiliar with the intricacies of code maintenance. However, the repercussions can be severe and far-reaching.</p><blockquote>As a dashboard evolves and grows in complexity over time, the presence of messy, unstandardized code becomes a ticking time bomb waiting to detonate.</blockquote><p>What may seem like a minor inconvenience during the initial development phase can snowball into a nightmare of debugging and troubleshooting down the road.</p><h3>The Benefits of Clean Query Practices</h3><p>Clean code/queries are not merely aesthetic preferences; they are essential pillars of effective dashboard development. By prioritizing cleanliness, analysts/engineers can overcome the myriad challenges posed by ugly code and queries, thereby enhancing readability, maintainability, reusability, performance, scalability, and collaboration.</p><ul><li>First and foremost, <strong>clean code/queries are inherently more readable</strong>. When code is well-organized, properly commented, and follows consistent naming conventions, it becomes significantly easier for others (and even your future self) to understand its logic and functionality. This enhanced readability facilitates faster debugging and troubleshooting, minimizing the time spent deciphering obscure lines of code or queries.</li><li>Moreover, <strong>clean code/queries are more maintainable in the long run</strong>. By adhering to standardized practices and avoiding complex, convoluted structures, developers can streamline the process of making changes or updates to the dashboard. This not only reduces the likelihood of introducing new bugs but also ensures that the dashboard remains adaptable and responsive to evolving business needs.</li><li>Additionally, <strong>clean code/queries promote reusability</strong>, allowing developers to leverage existing components and modules across multiple projects. By modularizing code and adhering to principles such as DRY (Don’t Repeat Yourself), developers can avoid duplicating efforts and foster a more efficient development workflow.</li><li>In terms of performance and scalability, clean code and queries can make a significant difference. <strong>Well-optimized code and queries are inherently more efficient, resulting in faster execution times and improved overall performance of the dashboard</strong>. Furthermore, clean code and queries are better equipped to handle increasing volumes of data and user traffic, ensuring that the dashboard remains responsive and scalable as the business grows.</li><li>Last but not least, <strong>clean code/queries facilitate collaboration among team members</strong>. By establishing clear coding standards and conventions, developers can foster a cohesive development environment where team members can easily understand and contribute to each other’s work. This collaborative ethos not only enhances productivity but also fosters a sense of ownership and accountability within the team.</li></ul><h3>How to write good queries?</h3><p>Now, we’ll delve into the essential principles and techniques for writing good queries based on the principles above.</p><h4>Indentation, and Spacing</h4><p>Query indentation and spacing play a crucial role in enhancing the readability and maintainability of SQL queries. When properly implemented, these formatting practices can significantly improve the clarity and comprehensibility of code, making it easier for developers and analysts to understand, debug, and modify queries as needed.</p><p><strong>Indentation</strong> helps to visually organize the structure of the query, making it easier to discern the hierarchical relationships between different elements such as SELECT statements, JOIN clauses, WHERE conditions, and GROUP BY clauses. It helps developers can quickly identify the logical flow of the query and understand how each component contributes to the overall functionality.</p><p><strong>Spacing</strong> between keywords, operators, and identifiers can further enhance the readability of SQL queries. Ample spacing around keywords such as SELECT, FROM, WHERE, and GROUP BY helps to delineate their significance within the query and makes them stand out more prominently. Similarly, spacing around arithmetic operators, comparison operators, and logical operators helps to improve the visual separation between different components of the query, reducing the likelihood of errors and misunderstandings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ohbHL58JOHwdhLMvQUMjEw.png" /></figure><h4>Common Table Expressions vs Subquery</h4><p><strong>Subquery is highly not recommended for use</strong>. Utilizing Common Table Expressions (CTEs) in BigQuery offers several compelling advantages over using nested subqueries. CTEs enhance the readability and maintainability of SQL queries by providing a clear and structured way to define and reference intermediate result sets. This not only simplifies the query code but also makes it more intuitive for developers and analysts to understand and troubleshoot.</p><p>Additionally, CTEs promote code reusability, as they allow complex subqueries to be defined once and then referenced multiple times within the main query, reducing redundancy and the risk of errors. Moreover, CTEs often result in more efficient query execution plans, as the query optimizer can better optimize and cache the intermediate results, potentially leading to faster query performance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XybwBh9M390wM6AytXrJ7A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kCPSMo0eWm0u_V9hg9mjRg.png" /><figcaption>Subquery vs Common Table Expression (CTE)</figcaption></figure><h4>Naming Standardization for Variables, Table Expressions, and Table Aliases</h4><p>Using good naming in SQL is essential because :</p><ol><li><strong>descriptive and meaningful names make the SQL code more readable and understandable</strong>. When someone reads the code, they can quickly grasp the purpose and content of the CTE, table, or variable without needing to decipher cryptic or generic names.</li><li><strong>Well-chosen names convey the intent and purpose of the CTE, table, or variable</strong>. This clarity helps both the original developer and other team members understand the logic and functionality of the SQL code.</li><li><strong>Reduced Ambiguity</strong>: Using distinct and specific names reduces the potential for naming conflicts and ambiguities within the SQL code. It minimizes the chances of accidentally referencing the wrong object or introducing errors due to naming collisions.</li></ol><p>Here are some good naming practices:</p><ol><li><strong>CTE (Common Table Expression) Names</strong>:</li></ol><ul><li>Use descriptive names that indicate the purpose of the CTE.</li><li>Use underscores to make the name readable.</li><li>Avoid overly long names, but prioritize clarity.</li><li>Start with a letter and use alphanumeric characters.</li></ul><p>Example :</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*76uHP0-m8-1FnCTzcjShUg.png" /></figure><p><strong>2. Table Aliases:</strong></p><ul><li>Use short and meaningful aliases to represent table names in your queries.</li><li>Avoid single-letter aliases that can be confusing.</li><li>Be consistent with your alias naming throughout your query.</li></ul><p>Example :</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kclMu_Q2UqWp6-Zopn-b5A.png" /></figure><p><strong>3. Variable Names:</strong></p><ul><li>Use meaningful and descriptive names for variables.</li><li>Follow the same conventions as CTEs.</li></ul><p>Example :</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dbGY_tfcmxQ9GZ_iWazUoA.png" /></figure><h4>Comment Utilization for Explaining Logic</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aWNs04wI4uCpt-OXc0SKgQ.png" /><figcaption>Example of using comments in SQL</figcaption></figure><p>Comments serve as a form of documentation within the code. They provide context and explanations about how certain parts of the code work, why specific decisions were made, and any potential caveats or considerations. This documentation is invaluable for both the original developer and anyone else who may need to read or work with the code, including team members and future maintainers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*smmdX6idF0XhWXq_IESfbg.png" /><figcaption>Comment utilization to explain specific query</figcaption></figure><h3>Conclusion</h3><p>While we’ve covered several aspects of clean code and query practices in this article, it’s essential to recognize that there’s much more to explore and learn in query management. I hope that the insights and examples shared in this article have provided you with valuable guidance and inspiration for improving your SQL coding skills.</p><blockquote><strong>Prioritizing cleanliness and clarity in your code not only enhances your own productivity and efficiency but also fosters a culture of collaboration and excellence within your development team.</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/657/1*rnJaNibdPRDzdoZ19EO_AQ.gif" /></figure><p>As you continue on your coding journey, keep exploring, experimenting, and refining your skills. Your commitment to clean code and query practices will undoubtedly pay dividends in the long run.</p><p>Happy query-ing!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b687a064795" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[User Retention Analysis: A Comprehensive Guide with Google BigQuery]]></title>
            <link>https://f-ahsan.medium.com/unlocking-user-retention-analysis-a-comprehensive-guide-with-google-bigquery-548dc4b59491?source=rss-fcbf640af245------2</link>
            <guid isPermaLink="false">https://medium.com/p/548dc4b59491</guid>
            <category><![CDATA[bigquery]]></category>
            <category><![CDATA[sql]]></category>
            <category><![CDATA[retention-rate]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[retention]]></category>
            <dc:creator><![CDATA[Fahril Ahsan]]></dc:creator>
            <pubDate>Sat, 28 Oct 2023 19:01:39 GMT</pubDate>
            <atom:updated>2024-04-29T17:04:47.455Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>“<em>There is only one boss. The customer. And he can fire everybody in the company from the chairman on down, simply by spending his money somewhere else</em>”</blockquote><blockquote>— Sam Walton, the founder of Walmart</blockquote><p>Unlocking the secrets of user retention analysis is like discovering the hidden gems in a vast treasure chest of data. Picture this: users who eagerly renew their subscription to Netflix, Spotify, or similar services month after month, creating a thriving, loyal community. It’s a phenomenon that often takes a back seat to the quest for acquiring new users but holds immeasurable significance for businesses.</p><p>In this comprehensive guide, we’ll delve deep into why user retention is a metric worth understanding and how you can harness the power of Google BigQuery to dissect this critical aspect of your business. So, let’s embark on this journey to unveil the insights that will not only retain users but also empower your growth strategy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yQaKpvDL3TBdCeNhW2xIQw.png" /></figure><h4>Why Retaining Customer?</h4><p>Customer retention isn’t typically the go-to yardstick for gauging the initial quality of users brought on board. Instead, it operates like a magnifying glass, honing in on the intricate aspects of user engagement and the effectiveness of your user acquisition strategies. It’s like examining the pages of a well-loved book to understand the story’s depth and intricacies. You see, <strong>retention reveals how well your users are connected to your product or service</strong>, offering insights that go beyond just counting heads. Are they sticking around because they find value or because they’re locked into contracts? It’s like peering into the reasons behind their continued presence in your digital ecosystem.</p><p>Retaining users is all about the long game, making it a well-worn tool for assessing how effectively a product or service maintains its user base over time. Think of it as nurturing a garden, where you tend to your plants and ensure they flourish rather than planting new ones every season. This is where you get the real scoop on whether your offering is meeting the needs and expectations of your audience. By diving into retention metrics, you can spot trends, identify pain points, and refine your strategies to keep users engaged and loyal. In essence, retention analysis is your trusty compass in the journey to keep customers not just onboard but actively involved and enthusiastic about what you have to offer.</p><h4>Measuring Customer Retention</h4><p>Measuring retention can be as varied as the businesses themselves, each with its own unique flavor and rhythm. The crucial aspect is aligning the metrics with your specific goals and industry dynamics. For some businesses, retaining a customer might mean they’re actively using the product or making a transaction within the next day, while others may consider retention when engagement stretches into weeks or months.</p><blockquote><strong>Picture it like a diverse menu at a restaurant</strong></blockquote><p>Where each dish is meticulously crafted to cater to distinct tastes. To measure retention effectively, first, <strong>you need to understand your business’s life cycle and user behavior</strong>. Once you’ve got a handle on that, you can select the right metrics that resonate with your goals. Whether it’s daily, weekly, or monthly, the key is tracking user actions over time, and these insights can be a powerful compass guiding your strategic decisions.</p><p>Let’s adopt the most common calculation, a monthly retention model, where we consider a customer retained if they choose to resubscribe or repurchase to our product in the subsequent month.</p><p>The first step is to identify the :</p><ul><li>acquisition month (first transaction)</li><li>completed month (month of the completed transaction)</li></ul><p>of each of the customers using the following query:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IAM4TgV6ieBk3Cy8VWxZ3w.png" /><figcaption>pro tip: don’t forget to filter transaction status, we don’t want to calculate the failed or canceled transaction right?</figcaption></figure><p>Now let’s check one user_id as an example. In the following table, user 35421 completed his first transaction in May 2023. Then completed another transactions in August and September 2023.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*pypRwtfDaf12MScQgkVwew.png" /></figure><p>We don’t actually track the number of transactions completed for each month. If a customer makes a transaction in a particular month, a new ‘completed_month’ will be recorded.</p><p>The next step is calculating the month difference from the ‘completed_month’ and the ‘acquisition_month’ to find the transaction month from each of the users:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ScdvT73SjY_oCSkx-uOv0Q.png" /><figcaption>Let’s only use users who acquired in 2023</figcaption></figure><p>We’ve make an addition to our data with the inclusion of a new column, named ‘m_trx’. This column serves as a helpful indicator of the transacting month number since the acquisition month. To illustrate, in the following table, let’s consider a user acquired in May 2023, with both the ‘acquisition_month’ and ‘completed_month’ being May 2023. In this scenario, ‘m_trx’ would be assigned a value of 0, effectively marking it as an acquisition month identifier.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*Pnen1ZkUhYQ5aFHI5x5Z0Q.png" /></figure><p>Now, fast forward to a case where the ‘completed_month’ is August 2023, the ‘m_trx’ takes on the value 3, or where the ‘completed_month’ is September 2023, the ‘m_trx’ takes on the value 4. This tells us that the user is not only done a single transaction in his acquisition month, but has continued to transact for three to four months following their initial acquisition back in May 2023.</p><p>While acquisition months for each user and the subsequent months in which they make transactions is in the hand, we’re ready to dive into some insightful calculations. We can promptly determine the number of users acquired and those who’ve stayed on board for each subsequent month within our chosen time frame. Typically, we focus on the next 12 months, but the beauty of this approach is its flexibility. You have the freedom to assess retention over a shorter 6-month period, or if you’re after the long game, extend your analysis to cover the span of 24 months. This adaptability lets you tailor your insights to the specific goals and time horizons of your business.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*viQgGI__2psRaWgQuU1u1Q.png" /></figure><p>Using the prior user_id, 35421 previously acquired in May 2023 resulting his ‘m_trx’ = 0 also falls in May 2023 and eventually the ‘new_users’ will be added by 1. While he transacted in August 2023 and September 2023, the m3 and m4 retention will be added by 1 respectively.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dQQfoBUAq9AasLCMm-p1Jg.png" /><figcaption>retention of user_id 35421</figcaption></figure><p>Now here are the results if we remove the user_id filter and show the entire data. For example, in July 2023, we acquired 1291 users, and there were 30 users who made another transaction in the following month.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TJOwecoN5xQrlRL5VF8avw.png" /><figcaption>retention of users who acquired in 2023</figcaption></figure><p>Following this, we’ll delve into the real retention percentages. One of our key business goals is to enhance our retention rates over time, meaning we aim for an upward trajectory as we move along the timeline. In other words, if we can keep a larger percentage of new users who signed up in the next month compared to those in the previous month, we’re making progress in the right direction.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iR1Hh9zKLMnf2a1nfCToEw.png" /></figure><p>As depicted in the table provided below, here’s what unfolded: In July 2023, <strong>2.32%</strong> of new users came back in the subsequent month, while <strong>4.01%</strong> of newcomers from August 2023 returned in the following month. This indicates a positive trend! These are huge swings in the retention rate. If there is something to explain it, like we’re on a ramadan period or a school holiday, fine. But we need to have an explanation. You may ask the marketing or the product team if there’s a certain campaign running in that period.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6RcBizeRjZJGCJjDVZcitA.png" /></figure><p>Next, we’ll turn our attention to another metric — the pace at which customers disengage. <strong>Naturally, over time, we anticipate a decline in retention</strong>, which is why we consistently seek to bring in new users. However, if we can minimize the rate at which customers depart, resulting in a gradual decrease between each period, we can allocate more resources towards catering to our customer’s needs rather than solely focusing on acquiring new ones. In doing so, we foster stronger retention as our users appreciate the increased attention we dedicate to their satisfaction.</p><h4>Final Thought</h4><p>By incorporating retention analysis into your data-driven decision-making process, you unlock the ability to make informed, strategic choices that impact your bottom line. It’s not just about numbers and percentages; it’s about understanding the journey your users take with your product or service. The insights derived from retention analysis provide a window into the minds and behaviors of your audience, enabling you to make precise adjustments, build stronger relationships, and optimize the user experience.</p><p>Furthermore, user retention analysis acts as a litmus test for the effectiveness of your user acquisition strategies. It’s one thing to bring users in the door, but it’s another to make them feel at home and keep them coming back. Retention analysis allows you to scrutinize the health of your user base — identifying whether your acquisition efforts are aligned with the long-term goals of your business.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=548dc4b59491" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>