<?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 Shao Zhifei on Medium]]></title>
        <description><![CDATA[Stories by Shao Zhifei on Medium]]></description>
        <link>https://medium.com/@shaozhifei?source=rss-c928948923f4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*bihRctkTruhVd3a9</url>
            <title>Stories by Shao Zhifei on Medium</title>
            <link>https://medium.com/@shaozhifei?source=rss-c928948923f4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 07:22:47 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@shaozhifei/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[Field Parameters — Probably the Best Feature of Power BI]]></title>
            <link>https://medium.com/@shaozhifei/field-parameters-probably-the-best-feature-of-power-bi-41ca72454732?source=rss-c928948923f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/41ca72454732</guid>
            <category><![CDATA[power-bi]]></category>
            <category><![CDATA[field-parameters]]></category>
            <dc:creator><![CDATA[Shao Zhifei]]></dc:creator>
            <pubDate>Fri, 27 Dec 2024 04:51:17 GMT</pubDate>
            <atom:updated>2024-12-27T04:51:17.253Z</atom:updated>
            <content:encoded><![CDATA[<h3>Field Parameters — Probably the Best Feature of Power BI</h3><p>In the world of data visualization, flexibility is key to uncovering deeper insights. Microsoft Power BI’s Field Parameters feature is a game-changer that brings a new level of interactivity to your reports. It allows users to dynamically switch between different dimensions and measures on the fly, making data exploration more intuitive and reports more streamlined.</p><p>Personally, I believe that Field Parameters might just be the best feature Power BI has to offer. Having extensively used the parameter feature in Tableau — which I thought was impressive — I was amazed to discover how much more powerful Power BI’s Field Parameters are. They not only replicate the functionality I was familiar with but elevate it to another level, offering greater flexibility and ease of use. In this post, I’ll delve into some of the best use cases of this feature that I’ve incorporated into nearly every dashboard in my work.</p><h3>Creating a Breakdown of Attributes/Measures</h3><p>One of the most common ways I use Field Parameters is to create a comprehensive list of attributes (dimensions) for my metrics. This feature allows users to flexibly select whichever attributes they want to display, essentially replicating a simple GROUP BY command in SQL. It empowers users to dynamically group and analyze data based on their specific needs without cluttering the report with multiple visuals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z933zDKbiyj4Y69hLDx24Q.png" /></figure><p>To create this dynamic breakdown:</p><ul><li>Navigate to the Modeling tab in Power BI.</li><li>Click on New Parameter and select Fields.</li><li>Add the required fields, attributes, or dimensions from your raw data to the parameter.</li><li>Rename the parameter to something meaningful — like “Breakdown” — to keep your report organized.</li></ul><p>This process allows users to select any attribute they want to display, giving them the flexibility to group and analyze data according to their specific needs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BZ6EQZcMvm-Mi6TPWSPCfw.png" /></figure><p>Additionally, I usually include an option called “All” in the Field Parameter. Selecting “All” means no breakdown is applied, allowing me to directly view the overall numbers without any grouping. This provides a quick way to see total metrics alongside the detailed breakdowns, enhancing the flexibility and usability of the report.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FCRH63rdedD7DwKA_aIHDQ.png" /></figure><h3>Scale Up Your Dashboard with More Attributes</h3><p>One of the great advantages of this implementation is the ease of scaling up by adding more attributes. It’s hard to imagine a static dashboard with a fixed number of attributes when businesses constantly demand more ways to slice and dice the data.</p><p>To add more attributes, you’ll need to directly modify the DAX formula (since Power BI doesn’t have a user interface for this yet). Here’s how you can do it:</p><ol><li>Click on the Field Parameter (e.g., the Breakdown you created earlier).</li><li>In the formula bar, you’ll see the DAX formula that defines the parameter.</li><li>Add more attributes by following the existing format and adding more rows.</li><li>Ensure that the ordering at the end of the formula is correct.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ROOCxEmTJfUSKgbUwlnR_g.png" /></figure><p>You can add almost anything to Field Parameters — not only attributes but also metrics directly. While I usually prefer to keep attributes and metrics separate, you can certainly mix them if it suits your reporting needs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B5D7sc3t8-gXkhZz1zqgvQ.png" /></figure><p>I really appreciate the flexibility and the ability to scale up by adding more attributes as time goes on. This approach accommodates the ever-changing aspects of analytics needs.</p><h3>Changing the Content of a Measure Based on User Input</h3><p>In Tableau, you can create a list of user inputs and change a calculation based on what the user selects. However, Power BI doesn’t directly offer a similar functionality. This used to trouble me because of my dashboard habits from Tableau. Then I discovered that it’s possible in Power BI, albeit in a somewhat indirect way.</p><p>For example, suppose we want to create a field called Count, which can either be Resale Count or Resale $ Value. The content of this count will change based on what the user selects. The user inputs we use are all measures or columns we’ve created — the actual content of the measures doesn’t really matter. What matters is the measure name and knowing which one the user has selected.</p><p>First, we create a Field Parameter called Count Metrics. Here, you can select any columns or measures you want — the goal is to provide options for the user to choose from.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OzuT-M7nouY7QtagDZpYIw.png" /></figure><p>Then, we create a measure using the DAX command:</p><pre>Count = SWITCH(<br>    MAX(&#39;Count Metrics&#39;[Count Metrics]),<br>    &quot;Resale Count&quot;, [Resale Count Measure],<br>    &quot;Resale $ Value&quot;, [Resale $ Value Measure]<br>)</pre><p>In this formula, MAX(&#39;Count Metrics&#39;[Count Metrics]) retrieves the user&#39;s selection from the Count Metrics field parameter. The SWITCH function then matches this selection to the corresponding measure. This effectively changes the content of Count based on the user&#39;s input, allowing for dynamic updates in your visuals depending on whether the user selects Resale Count or Resale $ Value.</p><p>By implementing this method, we can mimic the functionality of changing calculations based on user input — similar to what is available in Tableau — even though Power BI doesn’t offer this feature directly through the UI. This approach provides a flexible way to adjust measures dynamically, enhancing the interactivity of your dashboard.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/980/1*FFF8Ou24lGpxPbkr5adM3Q.png" /></figure><p>You might be wondering why we need such a feature when we already have the highly flexible Field Parameters. The reason is that this Count measure is used in subsequent calculations. By dynamically changing the content of Count based on user input, we can use the same formula for different measures, avoiding the need to create multiple similar measures.</p><p>For example, I used this approach to create Previous Count and Current Count in the RCA (Root Cause Analysis) formula mentioned in previous posts. This method streamlines the calculation process and keeps the measure definitions concise and maintainable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IMID2Zce8nqd8GBea6CQ-g.png" /></figure><h3>Summary</h3><p>In this blog post, I delve into the powerful capabilities of Power BI’s Field Parameters and how they have transformed the way I create dashboards and reports. Having extensively used Tableau’s parameter feature — which I found impressive — I was amazed to discover that Power BI’s Field Parameters offer even greater flexibility and ease of use. They not only replicate the functionality I’m familiar with but elevate it, allowing for more dynamic and interactive data exploration.</p><p>One of the key use cases I discuss is creating a dynamic breakdown of attributes or measures. By leveraging Field Parameters, users can flexibly select any attribute they wish to display, essentially mimicking a simple GROUP BY command in SQL. This approach simplifies the process of scaling up dashboards with additional attributes, accommodating the ever-changing analytics needs of a business without cluttering the report with numerous visuals.</p><p>I also explore how to change the content of a measure based on user input. While Power BI doesn’t directly offer this functionality as Tableau does, I demonstrate an indirect method using Field Parameters and DAX formulas like SWITCH and MAX. This technique allows us to create measures that adapt based on user selections, enabling the use of the same formula for different measures and reducing the need to create multiple similar measures. For instance, I show how this method can be applied to create previous and current counts in Root Cause Analysis (RCA) formulas.</p><p>Overall, this post highlights how Field Parameters can enhance your Power BI reports by making them more interactive, flexible, and scalable. By incorporating these techniques into your dashboards, you can empower users to explore data in more meaningful ways and keep up with evolving business demands.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=41ca72454732" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Root Cause Analysis (RCA) with Count Metrics (Smaller Proportion, Bigger Contribution, Better…]]></title>
            <link>https://medium.com/@shaozhifei/root-cause-analysis-rca-with-count-metrics-smaller-proportion-bigger-contribution-better-f6e21cd2f627?source=rss-c928948923f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/f6e21cd2f627</guid>
            <category><![CDATA[rca]]></category>
            <category><![CDATA[root-cause-analysis]]></category>
            <category><![CDATA[gpt]]></category>
            <dc:creator><![CDATA[Shao Zhifei]]></dc:creator>
            <pubDate>Wed, 16 Oct 2024 06:30:11 GMT</pubDate>
            <atom:updated>2024-10-16T06:30:11.868Z</atom:updated>
            <content:encoded><![CDATA[<h3>Root Cause Analysis (RCA) with Count Metrics (Smaller Proportion, Bigger Contribution, Better Insight Continued)</h3><p>In my <a href="https://medium.com/@shaozhifei/smaller-proportion-bigger-contribution-better-insight-0f005d225896">last post</a>, I introduced the RCA formula for the ratio metrics that follows a key RCA concept:</p><blockquote>Smaller Proportion, Bigger Contribution, Better Insight</blockquote><p>Now I want to extend this idea to the count metrics (e.g., MTU, Sales, Txn Count etc).</p><h3>Motivation</h3><p>Previously I used to do Count Metrics RCA using the count delta as the ranking value. However this can run into issues when a particular dimension-value takes a large proportion. Let’s imagine the following scenario:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*ilIlasXTkUDYp2LhNU3cJg.png" /><figcaption>Example: Same Growth %</figcaption></figure><p>In the above example, the overall sales grows from $7850 to $8635 by 10%, and individual country sales growth is also 10%. If we simple use the Delta as the ranking method, ID will be ranked highest with $400 delta.</p><p>But is delta ranking meaningful? Since every country grows by the same %, it’s hard to argue that something particular interesting is happening in ID. The growth maybe related to regional or some other dimensions that drives the growth (e.g., covid restriction relaxed and there are increasing travellers, i.e., <strong>user_type</strong> <strong>=</strong> <strong>traveller</strong> should be the focus). It’s not that ID here is not important (it’s important because of its size). But analyst should look into other dimensions other than country to find more useful insights, and the rank value should be able to highlight more interesting dimension — value pairs for RCA.</p><h3>RCA formula for Count Metrics</h3><p>Let me introduce my RCA formula for count metrics. Here is the notation definition:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HGBkn_lj_Drwn9PUBcCCfQ.png" /></figure><p>This is the <strong>Count RCA formula</strong> to calculate the value for ranking purpose:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YV695rSRIYuYeTXLSPSD8w.png" /><figcaption>Main Count RCA formula</figcaption></figure><p>It’s quite simple formula, but let’s look closer on what the formula is trying to achieve.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZDvtOSL1DwDRWOlpuOjmGQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-Uqcwe_qeN3-Y3mgOYlY0w.png" /></figure><p>Essentially, the rank value is the difference between actual delta and the delta if the dimension-value x grows in the same scale as overall. The difference will tell us how much the dimension-value x is different from the overall growth.</p><p>This formula follows the key RCA concept:</p><blockquote>Smaller Proportion, Bigger Contribution, Better Insight</blockquote><p>The second part maybe easier to understand: the bigger the difference (i.e., Contribution), the higher the rank. But how does this formula penalise the big proportion (i.e., give more weight to smaller proportion values)? This is done by comparing with the overall growth. If a dimension-value x takes a very large proportion, such as ID, its growth is inevitably closer to the overall growth. The bigger the proportion, the smaller of the impact of it’s growth will be, since it’s very close to the overall growth rate, therefore indirectly penalizes the large proportion dimension-values.</p><p><strong>This formula is a simple way to use one number to take a compromise between the proportion and contribution.</strong> It may not be the perfect answer, but it’s a simple and rational way to highlight important dimension values.</p><h3>How to evaluate with proportion Vs. contribution in a simpler way?</h3><p>There is a relatively simple way to evaluate the proportion vs. contribution part. I create a new column called Average Proportion, which is the average of previous and current proportion of dimension-value x. And another column called % Delta Explained:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yhpwX4WgexvaFrq-LNMQRQ.png" /></figure><p>It’s the metric delta for dimension-value x and when summed up it equals to overall delta. This is a way to evaluate the contribution.</p><p>Now if we apply the formula to the original table in the beginning (where the growth from individual country is the same as the overall), here is the result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ENZoFLDnBrSNQKFUvSWIOA.png" /></figure><p>You can see with the new formula, the rank value are all 0, which means they are not doing anything particular interesting to the overall and they should be ranked the same.</p><p>Another interesting observation is that the Average Proportion and % Delta Explained is the same. This is another sign that no particular observation is insightful because small proportion leads to small contribution and big proportion leads to big contribution, align with the key RCA concept where good insight should be small proportion but also big impact!</p><h3>Another simple example with google sheet</h3><p>Let’s see a more reasonable example with some simple data inputs, where each country has different growth rate from previous to current period. Overall regional sales grows from $7850 to $12540 by $4690, a 59.75% growth rate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fuB0l5gXg4Xbg2PjlkvGKQ.png" /></figure><p>If we only rank by Delta, ID will stand out with $2000 growth. However, if we rank with the rank value from the RCA formula, SG will rank on top with $903. It’s delta is $1500, smaller than ID, but this is from $1000 to $2500 and a 150% growth, much higher than the regional 59.75% growth rate. I believe stakeholders will agree that in this scenario, SG is a more important focus than ID. It also has a big contrast between Avg Proportion (16.34%) and % Delta Explained (31.98%). This contrast also signifies the importance of SG. The RCA formula gives a more reasonable and actionable ranking than the simple delta.</p><h3>Real world example with HDB Resale Data</h3><p>Let’s use real world <a href="https://data.gov.sg/collections/189/view">Singapore HDB resale data</a> and do some RCA! Below is recent months of HDB Resale $ Volume (Sale Price * Resale Unit Count).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IVW5pUsWQsUWRfnlY89A3A.png" /></figure><p>Let’s select 2 months for comparison: Oct 2023 Vs. Jan 2024 sales data. The overall sales volume grows from $1,260,267,044 to $1,551,587,654 by $291,320,610 with 23.12% growth rate. So what is driving this big growth?</p><p>There are some attributes available, but let’s first check flat type, which is a common way of separating different types of HDB apartments. The table is first ranked using simple Delta.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VakDbvL-HxyhR6Yng4I0Uw.png" /></figure><p>With simple delta ranking, 5 ROOM and 4 ROOM are ranked on top. Let’s pay more attention to 4 ROOM. Its sales volume grows from $596,016,196 to $693,190,199 by $97,174,003 (16.30%). It’s a sizeable growth, but weaker comparing to overall 23.12%. It’s hard to argue that resale growth is driven by 4 ROOM because of its weak performance. If we compare the proportion and % Delta Explained. It’s 45.98% vs. 33.36%, so it’s big proportion but small contribution and driving down the overall growth. That’s why it’s rank value is negative.</p><p>If we rank using the rank value from the RCA formula, here is the result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eHFJojWBaeMQmBzQtPs3xA.png" /></figure><p>5 ROOM is still ranked on top, but EXECUTIVE is also ranked high, grows from $100,116,768 to $141,959,764 by $41,842,996 (41.79%). It’s much higher than overall growth. Proportion Vs. % Delta Explained is 8.55% Vs. 14.36%, also a big contrast. It’s not ranked as high as 5 ROOM, because 5 ROOM has a bigger proportion, and also much bigger impact. So it’s still reasonable to rank 5 ROOM on top, even with smaller growth rate at 34.28%.</p><h3>Auto RCA with multiple dimensions</h3><p>In previous example, user may have to select multiple dimensions and compare the rank value across them, which is not an ideal user experience. This can be easily done through code and automatically looped through multiple attributes and ranked them in one go!</p><p>Here is an example output and driven using GPT api to interpret user intent and automatically compile results:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pvUxnqJKB6Edgb2y-TYDJg.png" /></figure><p>The user asked on question on why the increase happen across two periods, the table automatically ranked the most important dimension-value pair on top, and in this case is storey_range = 04 TO 06, which sales increased from $245,558,520 to $352,686,410 by $107,127,890 (43.6%) much higher than the overall 23.12%. It’s also ranked higher than the 5 ROOM, which most people should agree it’s reasonable. Their delta is similar but storey_range = 04 TO 06 starts from a much lower base ($245,558,520 vs. $321,921,024)</p><p>User can even ask follow up question on why storey_range = 04 TO 06 increased so much:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dV_3DjyKWfnjW2nG7Bf7KA.png" /></figure><p>Looking into the result, it’s reasonable to be EXECUTIVE since those flat are more on the lower level end. And town = YISHUN is also highlighted, and the 2nd rank from the overall is also YISHUN, which means there are some correlation between YISHUN and storey_range = 04 TO 06. And user can play around to see the actual increase:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x7PDDPPp0oMCQDzi_gsE0A.png" /></figure><p>With the help of GPT, more interesting use case can be developed such as automatic insight report generation.</p><h3>Conclusion</h3><p>In this article, we’ve shown the Count RCA formula, why it can rank the right dimension-value pairs on top, and some fake and real examples. The formula is consistent with the key RCA concept of “Small Proportion, Big Contribution, Better Insight”. The formula should work for almost all use cases, because this concept is true, for most of the use cases.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f6e21cd2f627" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Smaller Proportion, Bigger Contribution, Better Insight]]></title>
            <link>https://medium.com/@shaozhifei/smaller-proportion-bigger-contribution-better-insight-0f005d225896?source=rss-c928948923f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/0f005d225896</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[root-cause-analysis]]></category>
            <category><![CDATA[rca]]></category>
            <dc:creator><![CDATA[Shao Zhifei]]></dc:creator>
            <pubDate>Mon, 20 May 2024 08:25:18 GMT</pubDate>
            <atom:updated>2024-05-21T06:20:14.120Z</atom:updated>
            <content:encoded><![CDATA[<h3>Motivation</h3><p>This article builds on my original piece, “Metric Decomposition Formula to Understand Metric Trend.” Interestingly, the formula I introduced is the same with the one mentioned by Max Halford (<a href="https://maxhalford.github.io/blog/kpi-evolution-decomposition/#decomposing-a-ratio">https://maxhalford.github.io/blog/kpi-evolution-decomposition/#decomposing-a-ratio</a>), and I really appreciate the recognition in his article.</p><p><a href="https://medium.com/@shaozhifei/metric-decomposition-formula-to-understand-metric-trend-e693b7a4c8cf">Metric Decomposition Formula to Understand Metric Trend</a></p><p>However, with the increased usage of this formula, particularly after implementing a process called “Auto RCA” — which involves repeatedly running the formula across all dimensions and combining the results — I discovered certain limitations. The original formula sometimes highlighted dimensions with a large proportion, leading to a bigger contribution, but the insights were not always meaningful. This prompted me to revise the formula to bring more impactful dimensions to the top. The key was to identify dimensions with a significant impact while also penalizing those with a larger proportion, and the revised formula is able to achieve that.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E6aSulEk4s5FSUOzi8bfHA.png" /></figure><p>Revising the formula led me to reflect more deeply on why it’s essential to identify high-impact dimensions. Ultimately, the reasoning aligns with the article’s title: “Smaller Proportion, Bigger Contribution, Better Insight.” <strong>I believe this idea is universally true for all ratio metrics (X/Y) evaluation, and this is why the methodoloy will work for all ratio metrics.</strong></p><p>How do we determine that one insight is better than another? Often, we instinctively know it, but what are the specific criteria? I believe we seek insights with a significant impact (measured by contribution to the total delta) while maintaining a smaller proportion. For example, if the overall delta is -5pp, and we find a dimension with only 10% proportion that accounts for the entire -5pp decrease, while the remaining 90% performs the same as the last period, this dimension is certainly a valuable insight. If another dimension with only 1% proportion achieves the same impact, it’s an even better insight.</p><p>Subconsciously, we evaluate insights this way. The revised metric contribution formula effectively captures this approach.</p><h3>Revision of the Metrics Decomposition Formula</h3><p>First, let’s quickly recap the previous formula. For more details, please refer to the <a href="https://medium.com/@shaozhifei/metric-decomposition-formula-to-understand-metric-trend-e693b7a4c8cf">original article</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zAJ7A6eiXjp-NMetIFxkjg.png" /></figure><p>In the metric contribution component, dimensions with larger proportions tend to lead to higher contributions. Why can this be problematic? I’ll explain this with a chart. I couldn’t find a good public dataset to demonstrate this and didn’t want to search extensively, but this issue frequently arises in practice.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N9StUmrzADKJkXoXwg7cEw.png" /><figcaption>Demo of Metric Contribution (grey box) Vs. Metric Delta Diff Contribution (red box)</figcaption></figure><p>The demo chart aims to rank the impact of the total metric delta change of -5pp for Singapore. The ranking is based on the original metrics contribution formula, which is the metric delta multiplied by the proportion. The X-axis represents the metric delta, with the total metric delta of -5pp highlighted. The Y-axis represents the current proportion. The area (grey box) represents the metric contribution from the original formula.</p><p>The top-ranked dimension in this case is country = SG, with 100% proportion. In practice, we typically exclude country insights in Root Cause Analysis (RCA) when there is only one country involved. However, my point is that dimensions with high proportions can yield strange results. For example, insights such as country_code = SG and Dimension 1 = A may not be useful because their proportions are too high, causing confusion when these dimensions are ranked at the top. While this might be acceptable for an analyst to review, it can become troublesome when presented to stakeholders. It is especially problematic when we aim to create automatic insights with LLM models.</p><p>The solution is to penalize the proportion. This is achieved by incorporating the overall metric delta into the original formula, which becomes what I call the Metric Delta Diff Contribution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YgwqNcHkafi14xgCAQJNrA.png" /></figure><p>It’s not the best name, but it conveys the idea. Why does this penalize the proportion? Because if a dimension’s proportion becomes too large, it inevitably aligns more closely with the overall delta, which is the result when the proportion reaches 100%. This is clearly visible in the chart (grey box vs. red box).</p><p>To be frank, I only realized the proportion-penalizing aspect after changing the formula. Over time, I increasingly feel this makes sense and is a smart way to highlight dimensions with small proportions but significant impacts!</p><p>In summary, the revised Metric Decomposition Formula can be presented in the following way:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*duwABajnjg_eCiH91qu-Wg.png" /></figure><p>The Metric Delta Diff Contribution is now used to rank the metric contribution, and the proportion contribution ranking part stays the same. In the end the ranking of the two needs to be combined. This is not as straightforward as before since these two contributions are not in the same scale. I don’t have a good way to do this, but employed a naive way of ranking:</p><ol><li>Maintain the ranking from the Metric Delta Diff Contribution, but filter out the ones if metric_delta_diff_contribution &lt; total_contribution. This is because there is big proportion contribution with reverese direction, therefore not useful as the final RCA</li><li>Insert proportion contribution ranking elements if proportion contribution is higher than the metric contribution.</li></ol><pre># pick top 5 high metric and proportion contribution dimensions, if reverse direction, pick the negative ones<br>if save[&#39;total_metric_delta&#39;].max() &lt; 0:<br>    top_metric_contribution = save[save[&#39;metric_delta_diff_contribution&#39;] &lt; save[&#39;total_contribution&#39;]]<br>else:<br>    top_metric_contribution = save[save[&#39;metric_delta_diff_contribution&#39;] &gt; save[&#39;total_contribution&#39;]]<br><br>top_metric_contribution = top_metric_contribution.sort_values(by = &#39;metric_delta_diff_contribution&#39;, ascending = save[&#39;total_metric_delta&#39;].max() &gt; 0).head(5)<br>top_proportion_contribution = save.sort_values(by = &#39;proportion_contribution&#39;, ascending = save[&#39;total_metric_delta&#39;].max() &gt; 0).head(5)<br>top_metric_contribution[&#39;rank&#39;] = range(0, len(top_metric_contribution))<br><br># combine and rank the metric and proportion contribution. Maintain the order of metric contribution since that is ordered on metric_delta_diff_contribution, but for proportion contribution, it is ordered by proportion_contribution and compared to metric_contribution, insert the proportions back to the metric contribution ranks<br>for idx in tqdm(top_proportion_contribution.index):<br>    row = top_proportion_contribution.loc[idx]<br>    # Check each row of top_metric_contribution<br>    for idx2, row2 in top_metric_contribution.iloc[::-1].iterrows():<br>        # Compare proportion_contribution and metric_contribution<br>        if save[&#39;total_metric_delta&#39;].max() &lt; 0:<br>            if row[&#39;proportion_contribution&#39;] &gt; row2[&#39;metric_contribution&#39;]:<br>                top_proportion_contribution.loc[idx, &#39;rank&#39;] = row2[&#39;rank&#39;] - 0.5<br>        else:<br>            if row[&#39;proportion_contribution&#39;] &lt; row2[&#39;metric_contribution&#39;]:<br>                top_proportion_contribution.loc[idx, &#39;rank&#39;] = row2[&#39;rank&#39;] - 0.5<br><br>top = pd.concat([top_metric_contribution, top_proportion_contribution]).dropna(subset=[&#39;rank&#39;])<br>top = top.sort_values(by = &#39;rank&#39;).drop([&#39;rank&#39;], axis=1)</pre><p>When presenting the final result, I recommend to quote only the total contribution and proportion contribution part. Metric Delta Diff Contribution is good for ranking, but maybe hard to interpret by stakeholders.</p><h3>Demo with the HDB Resale data (from data.gov.sg)</h3><p>Below is an example RCA view for HDB average resale price comparison for April 2023 vs. April 2024. The price increased from $573,981 to $591,492, a rise of $17,511. This view aims to explain why the price increased.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GAWOAIzphSeKo7mCbRx3Sw.png" /></figure><p>The Singapore HDB resale data is an excellent dataset but not ideally suited to demonstrate the need for Metric Delta Diff Contribution, as none of the dimensions have an overwhelming proportion (e.g., &gt;80%). Nonetheless, let’s interpret the result.</p><p>The top RCA dimension identified is floor_area_sqm = 90 bucket. Its price increased by $41,001, significantly higher than the total delta. The total contribution is $9,113, explaining 52.04% of the total $17,511 increase. The current proportion is only 24.13%, yet it accounts for 52.04% of the delta, making this a reasonable RCA result. Combining the top two RCA dimensions explains 70.27% of the total delta increase, with a current proportion of 43.92%. Given the continuous rise in Singapore property prices, it may be challenging to find a single dimension that overwhelmingly explains the increase.</p><p>If we use the original Metric Contribution to rank, flat_type = 4 ROOM will be ranked on top. It’s total contribution is $12,152 (69.40% of $17,511) with current proportion of 43.53%. Hope you agree this is not an as good RCA as the floor_area_sqm = 90 (explained 52.04% of total delta with 24.13% of proportion).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Sf1rQ3b_QQIAC-n8zPf1rQ.png" /></figure><p>The HDB data may not be an ideal example, but in practise I’ve seen RCA that is much better explained by the new formula.</p><h3>Expectation of the Metric Decomposition Formula</h3><p>The revised Metrics Decomposition Formula improves RCA performance, though it does not guarantee finding the root cause. However, it ensures that root causes are surfaced if the relevant attributes are present in the data.</p><p>The quality of the insights depends heavily on the quality of the attributes. We should continuously seek better attributes to enhance our RCA capability, which requires strong analytical skills and experience. If the RCA results are not ideal, it may be due to a lack of good attributes. There should be an iterative process to incorporate more attributes to increase the power of RCA.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CwUI9syZ0XuyGfhz_PyNoA.png" /></figure><h3>Future with LLM</h3><p>I foresee this formula may integrate well with the LLM, since the output of the RCA is so specific and easier to be fed into the LLM models to generate concise insights, and even generating insights summary from several key metrics. This is an exciting field, and I’m thrilled by its potential.</p><h3><strong>Conclusion</strong></h3><p>In this article, we’ve explored the enhancements to the Metric Decomposition Formula, aiming to improve our ability to uncover impactful insights by penalizing larger proportions. By revising the formula, we address the limitations of the original approach, which often highlighted dimensions with high proportions but less meaningful insights.</p><p>Through examples and explanations, we’ve demonstrated how the improved formula surfaces dimensions with significant contributions and smaller proportions, leading to more valuable insights.</p><p>Additionally, we emphasized the importance of high-quality attributes in RCA and the iterative process required to refine these attributes continually. The revised formula ensures that root causes are surfaced if relevant attributes are present, enhancing the overall RCA performance.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0f005d225896" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Contribution Value to Understand Overall Metric]]></title>
            <link>https://medium.com/@shaozhifei/contribution-value-to-understand-overall-metric-f2bb38bed145?source=rss-c928948923f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/f2bb38bed145</guid>
            <category><![CDATA[root-cause-analysis]]></category>
            <category><![CDATA[data-analysis]]></category>
            <dc:creator><![CDATA[Shao Zhifei]]></dc:creator>
            <pubDate>Fri, 15 Dec 2023 07:19:04 GMT</pubDate>
            <atom:updated>2023-12-15T07:19:04.983Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Introduction</strong></h3><p>In my previous article, I delved into the details of identifying the factors that driving metrics trends, a method for scenarios such as regular metric review meetings. Now, let’s go to a different aspect of Root Cause Analysis (RCA): unraveling the underlying factors that influence the metrics as a whole. This kind of analysis is essential when we’re tasked with figuring out the reasons why our metric is underperforming or when seeking straightforward improvements (“low hanging fruit”) that can be quickly implemented to elevate our metric performance.</p><h3>Problem Statement</h3><p>Much like the factors influencing metrics trends, the performance and relative size of each subgroup are pivotal in determining the overall metric. These elements are interdependent and must be considered as a whole. A subgroup with high performance but a minuscule share in the overall proportion will barely move the needle on the aggregate metric. Hence, it is essential that our analytic approach incorporates a formula that accommodates the combined effects of performance and proportion to accurately gauge their impact.</p><h3>Developing the Solution</h3><p>First let’s familiarize with some key definitions:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wNInASJBc31XoSWrfXazBA.png" /></figure><p>One thing to note, the proportion always sums up to 100%:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2X5M-nOlUAujFyc5AtNxzw.jpeg" /></figure><p>And the overall metric can be expressed as performance times proportion:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QukOpD0xC4uMIAutuQLAuw.jpeg" /></figure><p>The influence exerted by subgroup x, which we refer to as the ‘Contribution Value’ (though may need to think of a better term), can be quantified by calculating the difference between the overall metric and the recalculated overall metric after excluding subgroup x from the population:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ESFwhnbcBJaagWJCWVdvDQ.png" /></figure><p>The concept is straightforward. For instance, if we aim to ascertain the impact of 5 ROOM flat type on the overall resale price of HDB (Housing and Development Board) flats in Singapore, we simply compare the overall average price to the average price recalculated with 5 ROOM units excluded.</p><h3>Demo with HDB Resale data</h3><p>We will still use the dataset of Singapore HDB Resale Flat Prices (which you can find <a href="https://beta.data.gov.sg/datasets/189/view">here</a>) to demo how the formula works.</p><p>First, let’s have a look of the monthly price trend from 1990 Jan to 2023 July.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NHSLrjM1IDPBigeNoXPF3Q.png" /></figure><p>Of course the price increased a lot since covid, and we discussed how to use the metrics decomposition formula to understand the trend in previous post. Now let’s focus on 2023, and answer the question of “What makes the HDB resale price so high?”</p><p>Here is the average resale HDB price since from 2023 Jan to 2023 July ($567,476):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YWpURSDjvq6Z4MR7V1DnIw.png" /></figure><p>We can put in different subgroups and see how they are impacting the resale price:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Irigimudr3W88Hxo63Zcmg.png" /><figcaption>Contribution Value by Town</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cl_VjfbTcWcfy3chXE3P_w.png" /><figcaption>Contribution Value by Flat Model</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VexGLwhghpnM4CyoX1gz5g.png" /><figcaption>Contribution by Flat Type</figcaption></figure><p>The overall effect on pricing is determined by analyzing a combination of metrics and their respective proportions. Specifically, flats with a higher price point require a significant proportion to register an impactful change. To illustrate, consider the 5 ROOM flat category. Contribution Value of $33,857 indicates that such flats increase the average resale price by this amount. Without the inclusion of 5 ROOM flats, one would expect the average resale price to decrease from $567,476 to $533,619 ($567,476 — $33,857). This figure aligns with the calculated average resale price when 5 ROOM flat types are excluded from the analysis.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bmhSzboK3GdVcrfkx-w4sg.png" /><figcaption>Contribution Value by Flat Type without 5 ROOM</figcaption></figure><p>The Contribution Value serves as a user-friendly method for understanding how subgroups influence the overarching metrics. By integrating performance data with proportional contributions, it simplifies the analysis for stakeholders, making it straightforward to interpret the combined impact of these factors.</p><p>One may also find clicking on each attribute and find out Contribution Value one by one a tedious task. It’s possible to automatically loop through all possible combinations and rank the impacts in one go:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tKAryg1silDFTu05ObzTMA.png" /></figure><p>We will go through the technique in future posts.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f2bb38bed145" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Metric Decomposition Formula to Understand Metric Trend]]></title>
            <link>https://medium.com/@shaozhifei/metric-decomposition-formula-to-understand-metric-trend-e693b7a4c8cf?source=rss-c928948923f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/e693b7a4c8cf</guid>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[trend-analysis]]></category>
            <dc:creator><![CDATA[Shao Zhifei]]></dc:creator>
            <pubDate>Sun, 17 Sep 2023 09:11:51 GMT</pubDate>
            <atom:updated>2023-09-18T03:03:12.288Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>In the realm of analytics, explaining why metrics rise and fall is a high-priority task. It’s a question that’s always on stakeholders’ minds when they look at key metrics. The challenge is not only about figuring out what’s causing these changes, but also about the time and effort it takes to unravel the reasons.</p><p>That’s where the Metric Decomposition Formula comes into play. It’s designed to ease the process and quickly identify and rank features by their contribution to the metric’s performance, with the ability to differentiate between the contribution of the metric change and the proportion change — two factors that often pose difficulties in interpretation. With this formula in hand, we can simplify our analytic tasks and present the findings more clearly and effectively. And focus the rest of efforts on generating more useful features, and increasing the Root Cause Analysis (RCA) capability.</p><h3>Problem Statement</h3><p>On the surface, tracking changes in overall metrics seems straightforward. However, a closer dive reveals subtleties that might escape the untrained eye. Shifts in metrics are not always solely due to performance changes within individual subgroups, proportion change often comes into play, affecting the overall outcome.</p><p>Consider an analysis of housing price trends in two distinct regions: Region A, a premium locality, and Region B, a suburban area with relatively lower average resale prices. Let’s imagine that the average housing prices in both regions individually increase. However, due to a surge in new sales triggered by a housing launch, Region B sees a larger volume of transactions. Consequently, the combined average housing price across both regions dips — given Region B’s increased contribution with its lower prices — a counter-intuitive result! This phenomenon, known as Simpson’s Paradox, often perplexes people with its confusing outcomes.</p><p>The underlying reason behind Simpson’s Paradox lies in the fact that the change in metrics doesn’t solely depend on the performance of individual groups. Subgroup proportion shifts also play a crucial role in affecting the overall metrics. When a high-performing subgroup increases its proportion, the overall metrics improve. Conversely, when a weak-performing subgroup gains more weight, the general metrics take a hit.</p><p>And that’s how the Metric Decomposition Formula can help. It separates these two factors and simplifies the RCA process, making it easier to understand. With this formula, we can confidently tackle metric analysis in a manner that’s efficient, effective, and offers clarity.</p><h3>Developing the Solution</h3><p>Before we delve into the intricacies of the Metric Decomposition Formula, let’s familiarize ourselves with some key definitions:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bpdTWXWAkwigUu6UlVpaAQ.png" /></figure><p>One thing to note, the proportion always sums up to 100%:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c9TxlqWxl833KngPhOOVsQ.png" /></figure><p>The formula also introduces an important concept of ‘Metric Margin’. Essentially, it’s a way to measure how a subgroup is doing compared to the overall performance. And it forms the basis of Proportion Contribution.</p><p>Now, with these concepts in hand, let’s unveil the Metric Decomposition Formula:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IWgtOJ-_wRlk6F5BZFeEFg.png" /></figure><h3>The Formula Explained</h3><p><strong>Metric Contribution Properties:</strong></p><ul><li>Defined as metric shifts scaled by current proportion</li><li>An increase in metric shifts corresponds to a positive increment in the overall metric, whereas a decrease leads to a negative impact in the overall metric</li><li>Importantly, a subgroup with a larger current proportion will inherently have a greater impact on the overall metric due to its relative size</li></ul><p><strong>Porportion Contribution Properties:</strong></p><ul><li>Defined as proportion shifts scaled by metric margin</li><li>The impact of proportion shifts depends on the metric margin (good or bad subgroup). Proportion increase on a positive metric margin, will give positive impact to the overall metric, whereas a negative metric margin leads to negative impact to overall metric</li><li>Both proportion shift size and metric margin size will have impact to overall metric change</li></ul><p><strong>Rational Explanation on the Formula:</strong></p><ul><li>The metric contribution is primarily associated with the current proportion. That’s because its impact is significant only when the subgroup has a presence in the current period. If the current proportion is 0, changes in metrics should not induce an influence.</li><li>On the other hand, the proportion contribution solely relies on the metric margin, determined using only the previous period’s metrics. The rationale is that we distinguish ‘good’ or ‘bad’ subgroups based on historical data.</li><li>While the formulas can be adjusted to include both current and previous metrics and proportions, it would inevitably complicate the formula, making it more time-consuming to comprehend and explain. The primary objective here is to elucidate the underlying reasons behind changes in metrics, and the current formula achieves this while maintaining understandability. Its simplicity ensures accessibility, making it a practical tool in everyday analysis.</li></ul><p><strong>Numbers Always Add Up:</strong></p><ul><li>A chief characteristic of the Metric Decompression Formula is its inherent stability: the numbers consistently tally.</li><li>Metric contribution from each subgroups sums up to total metric contribution</li><li>Proportion contribution each subgroups sums up to total proportion contribution</li><li>The overall increase or decrease in the metric (metric delta) equates to the sum of the metric contribution and proportion contribution. This equation holds true for both subgroups and the total.</li></ul><p><strong>Proof:</strong></p><p>The overall metric can be expressed as performance times proportion:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2xdkDJxAwU-m-kfqMX1m1g.png" /></figure><p>Therefore metric delta can be expressed as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8Yqy7a0pQdkryWQAVDf3XA.png" /></figure><p>In order to create the metric delta of individual subgroup, we add an extra component of Mxpre * Px:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UsExJjOUM13gPlhGVTBwIQ.png" /></figure><p>And the formula can be rearranged as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Uy6htnOfO5qt-YVv6X5MMw.png" /></figure><p>We can see the metrics delta part works, but for the proportion change, it’s always positive. To reflect the impact of good and bad subgroup, we can add another component Mpre (constant) to the end:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wY_OdM2ljzhHrWSSAWAjqQ.png" /></figure><p>This holds true because we can add any constant to the proportion difference, since after adds up all the subgroup it always sums up to 100%:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0FuuzCa3JILKJALkfUKt3w.png" /></figure><h3>Demo with HDB Resale Data</h3><p>In order to illustrate the practical application of the Metric Decomposition Formula, we’ll be using the dataset of Singapore HDB Resale Flat Prices (which you can find <a href="https://beta.data.gov.sg/datasets/189/view">here</a>). Our goal: explain the trend of the average resale unit price using our formula.</p><p>First, let’s have a look of the monthly price trend from 1990 Jan to 2023 July</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-qNyHZjkY2Hm8mo_RQ-CyQ.png" /></figure><p>We can see a big unit price increase since 2020 (covid starts). With the formula, you can compare the trend between any time range. Let’s pick before covid time 2015 Jan –2019 Dec, where price is relatively stable, vs 2023 July, where the price reached all time high. Here is a quick comparison:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mgVQ_7HFRb_f_-AA-RGoAA.png" /></figure><p>The average unit price increased from $438,190 to $577,057, a total of $138,867 increase. What caused the change? We can start to unpack using additional attributes and the Metric Decomposition Formula.</p><p>Let’s first check town contribution, and we will focus on 3 towns to explain:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aYywGLSkQ3TouVzJ0cuOSg.png" /></figure><p>Sengkang emerges as the primary contributor to the overall price increase, with its total contribution at $12,209, 8.79% share of the $138,867. This significant increase is attributed to a steep hike in prices within Sengkang itself, leaping from $436,817 to $582,832. Additionally, Sengkang’ substantial share in overall sales, approximately 8.37% in July 2023, intensifies its influence. The proportion contribution is slightly negative at -$12, this is because $436,817 &lt; $438,190, making it a marginal weak subgroup, and it’s proportion increased from 7.48% to 8.37%, resulting a small negative contribution.</p><p>Interestingly, despite Bukit Timah experiencing a greater price surge of $302,914, but contributed mere $93 to the total price increase due to its considerably smaller share — 0.15% of the total Singapore resale flats at July 2023. This scenario underscores the considerable role of proportion in swaying the overall metric change and the important insights offered by the Metric Decomposition Formula.</p><p>Remarkably, there is only one town — Central Area — that delivered a negative impact (-$628) to the overall price. How is that possible given Central Area should be quite hot? To understand why this occurred, let’s delve into the specifics of Central Area:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qwIjIo_bG5oUtaqNQR7V4w.png" /></figure><p>Upon investigating multiple attributes, it appears that ‘flat type’ serves as a key explanatory factor. Interestingly, there were no sales recorded for 5-room flats in July 2023, resulting in a -$45,577 impact to the overall price. Since there is no sale in July 2023, 5-room metric contribution is recorded as 0. But it was considered as a very good subgroup during the previous period since it’s much higher price than overall, its drop in proportion contributed a big negative impact.</p><p>Given that July 2023 is a mere snapshot of a single month and Central Area constitutes less than 1% of total sales, it’s understandable to observe periodic variations, such as high-valued 5-room flats reporting zero sales. Such fluctuations significantly influence trends in the Central Area. However, with the assistance of the Metric Decomposition Formula, these variations become easier to identify and understand.</p><h3>Conclusion</h3><p>In this dynamic world of data analysis, quantifying and attributing changes in key metrics can be a complex task, and some draining analysts bandwidth to focus on deeper insights.</p><p>The Metrics Decomposition Formula offers easy way to RCA metric trend impact against different attributes. The numbers shown in this article are derived from a Power BI dashboard (or Tableau) with the formulas built in. Users can easily click and compar on different attributes and different periods.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jtthTKFmVFctlMvJ7UxXEw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e693b7a4c8cf" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>