<?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 Sridhar Manthripragada on Medium]]></title>
        <description><![CDATA[Stories by Sridhar Manthripragada on Medium]]></description>
        <link>https://medium.com/@srmp1382?source=rss-9d15721aedf4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*CMTEC2PQtOd5OMP4</url>
            <title>Stories by Sridhar Manthripragada on Medium</title>
            <link>https://medium.com/@srmp1382?source=rss-9d15721aedf4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 20:35:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@srmp1382/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[Creating a fare calculator for PMPML buses: A very enriching learning]]></title>
            <link>https://medium.com/@srmp1382/creating-a-fare-calculator-for-pmpml-buses-a-very-enriching-learning-c86c13f81a8c?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/c86c13f81a8c</guid>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[transportation]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[pmpml]]></category>
            <category><![CDATA[pune]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Sun, 16 Aug 2026 19:00:38 GMT</pubDate>
            <atom:updated>2026-08-16T19:00:38.611Z</atom:updated>
            <content:encoded><![CDATA[<p>I had written earlier that I was running out of quality data to analyze for my learning, and now I found one dataset that was worth the time and effort.</p><p>I started working on the PMPML dataset recently with one singular idea: How can we create a google maps style dashboard that can tell us the distance, number of stops, the bus route, and fare between 2 points that we select?</p><p>Now I have worked and posted on this data some years ago where I had shown the longest and busiest routes.</p><p>This time the goal was different and it resulted in a very good learning, everything I did was around fixed LODs.</p><p>Those who want the data can find it on the <a href="https://opendata.pmc.gov.in/datasets?id=100">PMC website</a>, and believe me, it is worth trying out.</p><p>I started by cleaning the data by relating the different tables together as shown and used chatgpt to ensure that all the lat-long coordinates were properly updated and missing and wrong values were corrected.</p><figure><img alt="Data cleaning of the pmpml dataset after cleaning" src="https://cdn-images-1.medium.com/max/1024/1*E1tD-kkz0FKZXg1Y1681GA.png" /><figcaption>Structure of the pmpml data tables after cleaning</figcaption></figure><p>This data is an exhaustive list of BRTS and non BRTS stops and routes.</p><p>Then I tried a few things like finding the longest routes by the distance (which is given) and my biggest learning here is that the context filter fails if there are multiple items selected in the filter that is not in the context, hence we need to use a rank and filter it manually.</p><p>I would like to thank the tableau community trailblazers Hari Ankem and Himanshu Shekhar for helping me and guiding me on this dataset.</p><figure><img alt="Longest routes by km in both the BRTS and non-BRTS segments" src="https://cdn-images-1.medium.com/max/1024/1*ZHqzMEcrkvFsvNMHds53QQ.png" /><figcaption>Longest routes in the BRTS and non-BRTS categories by distance (in km)</figcaption></figure><p>The next thing I did was look at the order in which things happen.</p><ol><li>It all starts with a route ID, and there are stops under a route.</li><li>The stops have a sequence number, and several stops are divided into stages.</li><li>The routes are suffixed as D for down, U for up, and R for round trip.</li></ol><p>Therefore by using an LOD calculation fixed on route ID, I was able to get the start and end points, the min stop sequence pertains to the start of the route and the max stop sequence is the last stop or end of the route.</p><figure><img alt="Start and end points of each route using the stop sequences: I used an LOD fixed on the route ID (Start point shown here, I did the end point in a similar manner)." src="https://cdn-images-1.medium.com/max/1024/1*95PZ2CbbwD3aVhUA7na9MQ.png" /><figcaption>Start and end points of each route (Showing example of the fixed LOD used for the route start, I did the same for the route end as well, only the aggregation changed to MAX instead of MIN.</figcaption></figure><p>To vary the origin and destination, I created 2 parameters-origin stop and destination stop.</p><p>Again I used a fixed LOD and got a stop sequence within each route id for the origin stop and destination stop.</p><p>Then simply subtracting them gave me the number of stops to travel between them.</p><figure><img alt="Origin sequence shown here, similarly I found destination sequence as well." src="https://cdn-images-1.medium.com/max/893/1*IOlrQnnXZ59gq6qPVQL_qw.png" /><figcaption>origin stop sequence-similarly I calculated the destination sequence as well.</figcaption></figure><p>To get the distance between these points proved more challenging.</p><p>Because I joined tables, and Tableau performs row level calculations, it needs both the origin and destination points in the same row before it can perform spatial calculations (otherwise it displays the points and does not perform the calculations because either origin or destination is null in each row).</p><p>Therefore I took help from the tableau community, and again the solution is using a fixed LOD.</p><p>By fixing the data on the route ID and using an LOD, Tableau ignores the fact that the values belong to different tables and brings them together to do the calculation.</p><p>The alternative would have meant a lot of cleanup and duplicate fields in excel or via tableau prep.</p><figure><img alt="Using a fixed LOD to get the latitude and longitude coordinates for the  origin and destination stops (origin latitude example shown here)." src="https://cdn-images-1.medium.com/max/1024/1*Cxo0q8F8IN4nquPy4HOvUg.png" /><figcaption>Shown here is an example of the fixed LOD for calculating the lat of the origin-I used similar calculations for origin long, destination lat, and destination long as well.</figcaption></figure><p>One more precaution was needed and I found the bug while testing.. basically the destination stop sequence must be greater than the origin stop sequence, else it will show the number of stops as negative.</p><p>Which means, for a the return journey, the whole stop sequence is shown in reverse.</p><p>I tried using other methods, by filtering for direction up and down, and using sum(stops travelled)≥0, and they worked sometimes and did not in other cases.</p><figure><img alt="Ensure that the destination stop sequence must be more than the origin stop sequence, else we will get stops travelled &lt;0" src="https://cdn-images-1.medium.com/max/502/1*Fl_U_MXMLPFV0_w_N8ucuA.png" /><figcaption>The destination stop sequence in the destination must be more than the origin stop sequence for the stops travelled to be positive</figcaption></figure><p>Once this is done, all we need to do is just add the distance, the stops travelled, the route ID, and the fare to the details in the marks card and display all fields in the tooltips.</p><figure><img alt="Final worksheet showing the map and you can select the origin and destination points and it will show you the route, distance, and fare." src="https://cdn-images-1.medium.com/max/1024/1*0QHtgNgJuI5W_eLvtltlCQ.png" /><figcaption>Final worksheet showing the route, distance, and fare givenn an origin and destination point</figcaption></figure><p>For now the one limitation we have here is that it only shows me if a direct bus is available else the map will go blank, I will need to work on that some other time.</p><p>Meanwhile I hope you enjoyed reading this.</p><p>This was a very good learning because I learnt something new about LODs and context filters as well as about testing my outputs.</p><p>Please do reach out with any suggestions or feedback for my learning and improvement, I will be happy to implement.</p><p>I will check out new datasets and hope to come back with a better analysis of something more challenging the next time around.</p><p>Till I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c86c13f81a8c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Analyzing the Lex Europa dataset from the FP20 challenge: Some good learnings]]></title>
            <link>https://medium.com/@srmp1382/analyzing-the-lex-europa-dataset-from-the-fp20-challenge-some-good-learnings-49e1e14abb4a?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/49e1e14abb4a</guid>
            <category><![CDATA[insights]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[europa-legal-challenge]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Fri, 07 Aug 2026 13:39:17 GMT</pubDate>
            <atom:updated>2026-08-07T13:39:17.714Z</atom:updated>
            <content:encoded><![CDATA[<p>I have been searching for a good dataset for some weeks now to analyze, and finally settled on a <a href="https://fp20analytics.com/live-challenge/">legal dataset from the FP20 challenge</a> for this month.</p><p>While I am not entirely satisfied with the quality of my analysis, and I also believe that I need to find a better and more interesting dataset, I believe that this analysis helped me think openly and try attempting what I otherwise find to be very abstract.</p><p>This is a dataset about law cases across Europe, and it tracks data even into the future (maybe it is projected data).</p><p>This was also a good learning experience because I once again ignored the standard questions in the file during my analysis (and maybe that is why I can do better), and choose to present my opinion on this.</p><p>I began by analyzing the growth of the number of cases, and I looked at their priority and complexity.</p><p>So I looked at the growth in the number of open cases wrt 1st-jan-2024 and found that the number of high, urgent, and priority cases is growing very fast in some countries-compared to the general growth in cases across the board.</p><p>And these have a seasonal nature to it-for some reason people sue more often during some specific periods of the year, as shown in the graph on the left side.</p><figure><img alt="The number of critical complexity cases has grown 600% over the past 2 years and it has a seasonality." src="https://cdn-images-1.medium.com/max/1024/1*w4Daxzh4rS6J56M5RnIfAQ.png" /><figcaption>The critical complexity cases have grown 700% in the last 2 years and has a seasonality, peaking every august</figcaption></figure><p>Then I drilled country wise and looked at open cases, and I found that there are 3 countries where the number of open cases is very high.</p><figure><img alt="UK, France, and Germany have the highest number of open cases across Europe." src="https://cdn-images-1.medium.com/max/1024/1*IuAnuWozY-sPiVP5uFmmwg.png" /><figcaption>UK, France, and Germany have the highest number of open cases till date.</figcaption></figure><p>I then went back and looked at the one indicator that people associate with court cases-how long does it take?</p><p>So I plotted the average time the case stays open (for open/ongoing cases) vs the number of such cases, and showed it for the different priority cases by country.</p><p>What I see is that in some countries, highly critical cases stay open for more than 700 days on average, and I highlight that below, there is an opportunity to help the people in the Czech republic solve their cases faster.</p><p>For critical cases, we have fewer cases and they stay open longer, while non critical cases are more and they are closed faster.</p><figure><img alt="Critical cases are fewer and they stay open longer, one example of the Czech republic is shown in the figure." src="https://cdn-images-1.medium.com/max/1024/1*hmxsdIA2AjBq1Xs69ZySyA.png" /><figcaption>High and urgent priority cases stay open longer and they are fewer in number.</figcaption></figure><p>Then I shifted my attention to the other big question: how much does it cost, and is the billing tied to achieving the outcome.</p><p>The chart below shows my findings, which are that the pricing depends on the complexity and risk, and has very little if anything to do with the actual winning the case.</p><p>I also found that it does not appear to depend on the industry sector, or the intensity of utilization of the lawyers.</p><figure><img alt="The money charged in terms of billing are not based on the outcome, and depend only on what the lawyers define as complexity." src="https://cdn-images-1.medium.com/max/1024/1*9HhPcyxkFPW6dvNrimhkxA.png" /><figcaption>The amount charged is not influenced by the outcome of winning or losing</figcaption></figure><p>This basically means there is an opportunity for firms doing outcome based pricing to enter the market here and differentiate themselves.</p><p>Then I asked how much money lawyers make on average.</p><p>The data here shows that the profit margin hovers around 50% irrespective of the complexity or the outcome or any other factor.</p><p>What particularly interested me was to see if there are sectors or countries where lawyers lose money even when they win cases.</p><p>So I plotted the revenue and cost for cases that were won and lost, and I found that there are some countries and industry sectors where the costs are consistently higher than revenues, even on winning or closing early.</p><figure><img alt="In belgium and denmark, lawyers are losing money on average even when they win cases." src="https://cdn-images-1.medium.com/max/1024/1*QbCy6_DmvOpUBq1UiHjGNQ.png" /><figcaption>Lawyers are losing money on average for cases won in belgium and denmark</figcaption></figure><p>Similarly, I found some industry sectors where the cost of winning is higher than the revenue that comes in.</p><figure><img alt="The lawyers are losing money on average even while winning cases in the automotive, telecom, finance, and media sectors." src="https://cdn-images-1.medium.com/max/1024/1*DMwwtxUhcUoawxmmGTM2FA.png" /><figcaption>Industry sectors where lawyers lose money on average even when cases are won</figcaption></figure><p>So my understanding is that these sectors and countries will see an increase in the cost of doing business eventually.</p><p>We see a similar trend when we plot lawyer utilization-it is more for cases lost than won, and this is specifically for complex cases.</p><figure><img alt="For complex high priority cases, lawyer utilization is more on cases lost than won, in many sectors." src="https://cdn-images-1.medium.com/max/1024/1*v82heZyoz9OI-r5PYHyfqQ.png" /><figcaption>Lawyer utilization is higher on cases lost than won in many sectors for highly complex cases</figcaption></figure><p>So this I believe is an opportunity for people to use more technology and statistical methods to predict the possible outcome and then adjusting the fees or effort accordingly.</p><p>I also checked what the invoice value depends on, and one more parameter I found was the number of documents submitted-the more the number of documents submitted, the higher the invoice value, as shown clearly below.</p><figure><img alt="The invoice value is based on the complexity and the number of documents submitted, it is not based on the outcome or lawyer effort." src="https://cdn-images-1.medium.com/max/1024/1*cDz9yGeD8F3yWwCI-n6Ttw.png" /><figcaption>The invoice value is directly related to the number of documents shared, it has little to do with how long the case is open or the actual result of the case</figcaption></figure><p>I tried to check if the pricing has any correlation to winning or losing, and I found that at a very high level, the correlation appears very weak.</p><p>The moment I added any other dimension, the correlation disappeared.</p><figure><img alt="The invocing for lawyers appears to be based more on the number of documents submitted, their calculations of risk and complexity than on their effort, time, or outcome." src="https://cdn-images-1.medium.com/max/1024/1*jC_ytX3EIXtFQdl_c_L9tw.png" /><figcaption>There is a very vague correlation between the outcome and the invoice at a high level, this is not significant.</figcaption></figure><p>So finally, my learnings and recommendations from analyzing the data are listed below.</p><ol><li>Specific industry sectors such as automotive, telecom, and financial services, particularly in Denmark and Belgium are likely to see an increase in the cost of doing legal business as we showed above.</li><li>My understanding is that the pricing should ideally be outcome based and should depend on fields like the days the case is open for, the % of lawyer utilization, and the outcome, along with the complexity of the business.</li></ol><p>Hence I believe there will be an opportunity for those who build such pricing models to enter and differentiate themselves to win.</p><p>3. I believe that there is a potential to predict outcomes at a more granular level, with the advent of tools which can analyze massive amounts of data.</p><p>Hence there is an opportunity to use these to put more reliable models in place and adjust the pricing accordingly.</p><p>4. There is an opportunity for firms to look at why there are so many critical complexity cases which are open for more than 700+ days, specifically in the Czech republic, and reduce the number of days outstanding.</p><p>I hope you enjoyed reading this analysis.</p><p>What I learned from this was:</p><ol><li>It is not necessary to pick up every data point, sometimes it may be better to leave some points there than try everything.</li><li>I revised my understanding on creating sets and side by side views and table calculations.</li><li>I found this to be a more abstract topic, and that was part of the challenge.</li></ol><p>Finding good data is important, and I am trying to look at some other newer data around attrition which I saw recently, so I hope to do better and come up with good insights the next time around.</p><p>Please do reach out with any feedback or suggestions for my learning and improvement, I will look to implement.</p><p>Until I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=49e1e14abb4a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Fun and learning with set actions to do multilevel drill down and drill up in tableau: Using the…]]></title>
            <link>https://medium.com/@srmp1382/fun-and-learning-with-set-actions-to-do-multilevel-drill-down-and-drill-up-in-tableau-using-the-43066857e69d?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/43066857e69d</guid>
            <category><![CDATA[multiple-datasources]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[pune]]></category>
            <category><![CDATA[drilldown]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Sat, 18 Jul 2026 18:14:51 GMT</pubDate>
            <atom:updated>2026-07-19T03:35:10.308Z</atom:updated>
            <content:encoded><![CDATA[<h3>Fun and learning with set actions to do multilevel drill down and drill up in tableau: Using the Pune facilities dataset</h3><p>It has been a while since I tried and learnt something new on the tableau side of things, so I decided to try this out as I prepare to teach the subject again from the next week.</p><p>This time, I started by collating data about the types of facilities and buildings in Pune and combining them.</p><p>For a start, I took some data of government and private hospitals, schools, gardens, libraries, and combined them with data on other facilities ward wise in Pune.</p><p>I have tried collecting data from the <a href="https://opendata.pmc.gov.in/datasets">PMC website</a> as well as the <a href="https://data.opencity.in/dataset/pune-healthcare-hospitals-and-clinics">Open City portal</a> for this so far, I will check out other sources as well.</p><p>I believe that we can unlock a lot of insights by combining various datasets and then ask the right questions.</p><p>For a start, I picked up government health facilities (PMC hospitals + clinics + dispensaries) vs private facilities (again hospitals + clinics) for analysis, and I show what I found below.</p><figure><img alt="Some areas have very few government healthcare facilities and many more private faciltiies, while others have a good mix of both." src="https://cdn-images-1.medium.com/max/1024/1*uTyPHMtrRw8343NKRJ-Z_w.png" /><figcaption>Some areas as shown here have very few government facilities and far more private ones, while others have a good mix.</figcaption></figure><p>We see areas such as Kasba-Vishrambagwada, which has 90 private hospitals and just 30 government facilities, while other places such as Bhawani Peth have 21 government facilities vs 9 private ones.</p><p>So maybe planners should go drill deeper into their data to see if there is an opportunity to put more government facilities in the Kasba-Vishrambagwada area?</p><p>What I also started to do is to use set actions to drill down and display the number of facilities of each type in each ward, and then drill further and display the location of each facility.</p><p>This is something nice that I learned today, that I can now execute multi level drill downs (or even drill down by multiple dimensions in the same sheet at the same level at the same time) using set actions.</p><p>I found the post at this <a href="https://help.salesforce.com/s/articleView?id=001470582&amp;type=1">link</a> particularly useful, and I tried to use it to good effect.</p><p>I put the steps below.</p><ol><li>First, I want to drill down and show the types of facilities and their count under each ward.</li></ol><p>For this, I used a regular set action (I created a set on ward name) and calculated field as shown below.</p><figure><img alt="Calculated field for a single level drill down using a set action" src="https://cdn-images-1.medium.com/max/1024/1*xfKgtsGjB8nw7q52uobQYg.png" /><figcaption>Calculated field for the drill down using a set action</figcaption></figure><figure><img alt="Set action for the calculated field shown in the figure above this." src="https://cdn-images-1.medium.com/max/1024/1*sw_JB-w8wAl5ljLWWy4nSw.png" /><figcaption>Use of the first set action.</figcaption></figure><p>Now for the second level drill down, I used the calculated field we created in the first step to make the new set-this is the key to drill down at the second level.</p><figure><img alt="Set for the second level drill down." src="https://cdn-images-1.medium.com/max/1024/1*Gd9cq_snQPSRZbMpiSl3xA.png" /><figcaption>Using the output of the first calculated field as a set for the next drill down</figcaption></figure><p>Then in the next calculated field we use the first <em>and</em> the second set both.</p><p>In the second set action, we use the set we created from the first calculated field above.</p><figure><img alt="The second set action and corresponding calculated field." src="https://cdn-images-1.medium.com/max/1024/1*uHtqhD-CXJBH26Vt3axMxA.png" /><figcaption>In the second calculated field we need to use the “and” clause to combine both the fields from the first and second sets.</figcaption></figure><p>Now we just drag these calculated fields into the marks card, and we are ready.</p><p>It works as shown in the steps below.</p><figure><img alt="Drill down at the first level: Notice the items in the sub level have appeared in colours other than blue." src="https://cdn-images-1.medium.com/max/1024/1*xW_UzsktU4iK9g6eWenJsg.png" /><figcaption>Drill down at the first level: The colours other thsn the blue are the items at the sub level</figcaption></figure><figure><img alt="Drill down at the second level: notice the number of items that have appeared in the one type of facility under the area." src="https://cdn-images-1.medium.com/max/1024/1*tgBKSOIUSiL1x6AkGiE5nw.png" /><figcaption>Drill down at the second level when we click on private clinics (The tooltip for one of them is shown below).</figcaption></figure><p>So this was my learning for today which I wanted to share.</p><p>What I hope to do is combine the data for more facilities (such as metro stations, bus stops, or public toilets, etc.,) together and display them on the map like this.</p><p>Then we can query and ask nice questions like which schools, hospitals, or libraries lack access to say a metro or bus stop, for example, and then use that to recommend where to build the next facilities to serve the public better.</p><p>I will work on it and analyze it and post again soon, meanwhile I hope you enjoyed reading this one.</p><p>This post is more about sharing my learning on how to visualize than about the analysis per se, I will maybe do that next time and share insights and recommendations.</p><p>Please do share any feedback for my learning and improvement, I will be happy to implement.</p><p>Till I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=43066857e69d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ideas from the analysis of the electric cars dataset from Kaggle: The best electric cars to…]]></title>
            <link>https://medium.com/@srmp1382/ideas-from-the-analysis-of-the-electric-cars-dataset-from-kaggle-a-short-and-simple-story-66104d3aeabf?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/66104d3aeabf</guid>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[storytelling]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[electric-vehicles-india]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Sat, 11 Jul 2026 08:24:01 GMT</pubDate>
            <atom:updated>2026-07-12T15:23:27.961Z</atom:updated>
            <content:encoded><![CDATA[<h3>Ideas from the analysis of the electric cars dataset from Kaggle: The best electric cars to consider</h3><p>It has been a while since I found a good dataset, so I felt I must do this when I saw it.</p><p>It was Clive Humby in 2006, who famously said that data is the new oil, and in more recent times, many people have written that it is like water.</p><p>What I see most people missing out whether they use the oil or the water analogy, is the scarcity.</p><p>We have all seen the impact of the war on Iran, where the price of oil has skyrocketed causing inflation in general.</p><p>Also we see the way the countries which control the reserves of oil also control its price, and hence the growth of the world’s economy.</p><p>Similarly, we have been seeing that with water as well, we see a shortage, and those who control the ground water resources and supply of water are those who will control our lives and the economy.</p><p>Use the same analogy on data, and it becomes clear that we are facing a shortage of usable data, and those who can generate usable datasets are the ones who will eventually control the data economy, because they will market it at a price they want.</p><p>So let’s face it, we are already seeing inflation due to data shortage since the past year, so maybe it is time to get into building our own datasets to remain self reliant, just like the <a href="https://www.itcpspd.com/Page/plantations">ITC company</a> in India plants its own forests which it then uses for its paperboards.</p><p>Anyway, I found a new dataset and I wanted to share some of my findings in this post.</p><p>This dataset on the electric vehicles in India from kaggle has data of 2025, and it has fields where models have been scored on their value for money apart from granular data on fields influencing value.</p><p>So the first thing I did is to put a graph on the number of models and variants each brand has, and it appears that this is dominated by the budget brands of Tata, MG, and Mahindra, which have the highest assortment (and it may also be the reason we see so many cars of these brands on the roads nowadays).</p><figure><img alt="Tata, MG, and Mahindra have the highest assortment of electric vehicles." src="https://cdn-images-1.medium.com/max/1024/1*uru7N3HPvDhEgEVxjOhEJg.png" /><figcaption>Tata, MG, and Mahindra have the largest number of models and variants in electric vehicles.</figcaption></figure><p>I also checked if the value for money score is related to the price (as it should be), and which models provide good value, and I am showing that below.</p><figure><img alt="Models which are priced too high have very low value for money scores." src="https://cdn-images-1.medium.com/max/1024/1*QNndIySR0TqcIc2g6Znffg.png" /><figcaption>Models which are high priced have very low value for money scores, and vice versa</figcaption></figure><p>Then I checked on the mileage (km for every battery charge) and claimed vs actual and value for money.</p><p>What I found was a very significant difference between the claimed and the actual reported km range, and this varies across brands.</p><p>The pattern that I found here, is that on an average, the “premium brands” like BMW, Mercedes, and Porsche, are more believable in their claims, the difference between the claimed and actual was lesser for them than for brands like Tata and Mahindra which are aimed at the mass market at affordable prices.</p><p>To illustrate this, I used a dual axis chart and plotted the discrepancy in the claim on one axis and the price on the other, with a shared x axis of the brand.</p><figure><img alt="On an average, budget carmakers such as Tata, Vayve, Maruti, and Mahindra make taller claims than the premium models such as Porsche, BMW, Mercedes, Volvo, and Tesla." src="https://cdn-images-1.medium.com/max/1024/1*U5QgwFQCVi6OitPDvLLV7g.png" /><figcaption>Premium brands are more consistent in their mileage claims than budget carmakers</figcaption></figure><p>The other thing I did was check if the “value for money” score provided is related to the range per charge-does a vehicle which gives longer range have a higher value score?</p><p>And the answer was surprising as shown below.</p><p>I plotted the two axes and I found that this holds well for vehicles priced below the 50L price, for higher prices I see several outliers.</p><figure><img alt="A higher range means a lower value score, for some reason, and this correlation holds for vehicles priced below 55L." src="https://cdn-images-1.medium.com/max/1024/1*tF4wEFju_AR37zX10j1z9g.png" /><figcaption>Higher range of km does not translate to a higher value score.</figcaption></figure><p>So based on this, I tried to extract which are the best electric cars we can consider, and which ones to avoid altogether (for the average buyer).</p><p>From the graph above, I tried to extract the list of cars below the 55L price, which have a high km range and high value for money score.</p><p>And I believe that the Honda City V and Vx variants are the ones where both these scores are high.</p><p>On the other hand, there are models like the Porsche Cayenne E Hybrid, the Volvo Xc90 T8, the BMW X5 Xdrive 50E, the Mercedes Benz Gle 350DE 4matic, etc., which have a low range and a low score on value, and are also very costly.</p><p>Then I tried to add the see which models charge fastest yet give the highest range per charge, and in that I tried to look at the ones with the highest battery capacity.</p><p>I show my results below, these I believe are the vehicles we should look at, which are worth buying.</p><figure><img alt="Vehicles categorized by the speed of charging (right side of the axis is faster), having higher range per charge, and having a higher battery capacity." src="https://cdn-images-1.medium.com/max/1024/1*M1wCH9eVqbEAyN5S6RvhBQ.png" /><figcaption>Look at the top right quadrants for the vehicles which are ideal, they charge faster and have better mileage and higher battery capacities</figcaption></figure><p>And the list that I see here, the best ones to consider are:</p><p>The Mahindra BE6 EV, the Tata Harrier EV, the Byd Seal, Kia EV4, etc., are the cars in the category below 55L.</p><p>In the higher priced segment, the Hyundai Iconiq 5, Kia EV6, Mercedes Benz Cla electric and eqe models, Tesla model Y, MG Cyberster, etc., will satisfy our criteria.</p><p>I list the models and variants below that match our filtered criteria, which I believe add more value than the others.</p><figure><img alt="Cars with good mileage above 400 km and charging in less than 20 min with prices below 55L and above 55L." src="https://cdn-images-1.medium.com/max/1024/1*3-DUdD4ISROGw90fDwVFaQ.png" /><figcaption>List of cars with good mileage and faster charging, which are worth considering</figcaption></figure><p>I hope you enjoyed reading this analysis.</p><p>As I explained earlier, good and new challenging datasets for storytelling are difficult to come by nowadays, and this was a good dataset, even if it was a short post.</p><p>I do see a lot of interesting datasets around electric vehicles, so maybe I will pick up something around charging stations infrastructure next time as well.</p><p>What I learnt from this was to understand which benefits or features are important for an average prospect looking at electric vehicle data.</p><p>I hope to come back soon with a better and more challenging dataset and a newer analysis soon.</p><p>Please do share any feedback for my learning and improvement, I will be happy to implement.</p><p>Till I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=66104d3aeabf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Analysis of the recent Tamilnadu election results in 2026 from the Codebasics resume challenge]]></title>
            <link>https://medium.com/@srmp1382/analysis-of-the-recent-tamilnadu-election-results-in-2026-from-the-codebasics-resume-challenge-126f70b98442?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/126f70b98442</guid>
            <category><![CDATA[tamil-nadu]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[election-data]]></category>
            <category><![CDATA[tableau]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Wed, 20 May 2026 19:28:35 GMT</pubDate>
            <atom:updated>2026-05-22T14:49:31.422Z</atom:updated>
            <content:encoded><![CDATA[<p>I have been searching for nice and challenging datasets to learn and analyze, and I came across this dataset in the <a href="https://codebasics.io/challenges/codebasics-resume-project-challenge/26">Codebasics resume challenge</a>.</p><p>This dataset made me go back to the fundamentals of how we report numbers and look at how to ensure accuracy of results and conclusions from data.</p><p>I also enjoyed analyzing this because it gave me good learnings and insights which were very surprising.</p><p>For a start, while we see elections as a fight between a few political parties, I found that there were 105 parties which contested the state elections in Tamilnadu this year, and 109 in the previous election in 2021 across 233 constituencies.</p><figure><img alt="Out of the 4200+ parties, only some 20 parties are signficant and capture 99.7% of the votes polled." src="https://cdn-images-1.medium.com/max/1024/1*WVqsr2Q3f5mY22rodRj9ag.png" /><figcaption>The top 20 parties got 99.7% of the votes out of 4257 parties in 2026.</figcaption></figure><p>Now everyone following an election result knows that there are two main ways to look at things: the share of popular vote and the number of constituencies won by a party.</p><p>To find the share of votes in each party, I aggregated it at the district and regional level (just so it becomes easier to visualize).</p><p>We can see that in 2026, TVK ate into roughly half the share of DMK and AIADMK compared to earlier.</p><figure><img alt="We see that in 2026, the TVK party has garnered nearly 30%+ of the votes polled in each region, and eaten into the vote base of the 2 incumbent parties." src="https://cdn-images-1.medium.com/max/1024/1*FT6oJivYkttILtn6G5irZQ.png" /><figcaption>The TVK has captured 30%+ of the popular vote across each region in 2026, and both the DMK and AIADMK have lost roughly 50% of their earlier share</figcaption></figure><p>At the district level, I have shown the leaders below by ranking based on the votes polled, and here it clearly shows the winners by popular vote share at each district-as you see it is not dependent on the number of votes polled at the district.</p><figure><img alt="Leading parties in each election based on votes polled at a district level." src="https://cdn-images-1.medium.com/max/1024/1*cWs6Y-KixkXXYPg0HhMhWg.png" /><figcaption>Leading parties in election based on votes polled at the district level</figcaption></figure><p>And now I will come to the constituencies won and where we see the pattern.</p><p>Here I mapped the winners of in each constituency in each election and then filtered for the DMK, and we see that in 2026, the DMK lost most of the constituencies in the northern part of the state.</p><figure><img alt="The DMK in 2026 lost much of the northern constituencies to the TVK in 2026." src="https://cdn-images-1.medium.com/max/1024/1*pqD_9KykoNNCKg2bsYucHg.png" /><figcaption>The DMK lost much of its hold on the northern constituencies in 2026</figcaption></figure><p>The pattern for the AIADMK is the other way around- as shown below they lost most of the constituencies in the southern part of the state while holding on to the north.</p><figure><img alt="The AIADMK has lost much of its southern constituencies in 2026." src="https://cdn-images-1.medium.com/max/1024/1*fcDjW01M88j6CN7WN-AWwQ.png" /><figcaption>The AIADMK has lost much of its southern constituencies in 2026 while holding and winning in the north</figcaption></figure><p>In contrast, the TVK has won some patches in the north and some in the south as well.</p><figure><img alt="The TVK has won some constituencies in the north and some in the south as well." src="https://cdn-images-1.medium.com/max/1024/1*5GJQMwLIc6LbnxrcbwbwqA.png" /><figcaption>TVK has won in patches across the state in 2026</figcaption></figure><p>Now that we have spoken about winning constituencies, let us see which constituencies actually changed winner.</p><p>For a start, I looked at how many seats the parties actually contested out of 233, and I found that none of the main parties in 2021 or contested for all the seats, they only fielded for about 170–180.</p><p>On the other hand, we see from the figure that the TVK has contested all the 233 seats and won about 45% of them.</p><figure><img alt="TVK has contested on all the 233 seats with a better win rate unlike other parties which contested on around 180 seats." src="https://cdn-images-1.medium.com/max/1024/1*UkWWR3zG9a4RChC22xrelA.png" /><figcaption>TVK chose to contest all seats unlike other parties who fielded for only about 180 out of 233 seats</figcaption></figure><p>Now I talk about those seats where winning parties have changed since the last election.</p><p>This was actually difficult to do since a swing could be present and yet it might not impact the party in question, especially since I am taking only the top 3 parties for comparison here.</p><p>For example, if we have 3 parties A, B, and C, and say a constituency moves from party B to C, then the party A will not be impacted even though the winner in that constituency has changed.</p><p>Similarly, if a party C holds on to the constituency, then it will not impact the parties A and B because they were not involved.</p><p>So after filtering such cases, I verified and ended up with the figure shown below.</p><figure><img alt="Swing for and against parties and no swing-seats which have changed winners" src="https://cdn-images-1.medium.com/max/1024/1*T9KU9Vr69mDoBhLRgfkeGA.png" /><figcaption>Swing for and against parties-which seats have changed winners</figcaption></figure><p>So to show how many seats parties have won, lost or held on to wrt the last election in 2021, I filtered for the top 10 parties by the seats won, and found the % of swing for and against.</p><p>And this is what I ended up with.</p><figure><img alt="% swing for and against the top 10 parties." src="https://cdn-images-1.medium.com/max/1024/1*-CqFhM9q1hPqX5y5uFsZFA.png" /><figcaption>% swing for and against for the top 10 parties</figcaption></figure><p>Here we see that the incumbent DMK has lost 61% of its earlier seats, and only 10% of its seats in 2026 are from new wins.</p><p>The AIADMK performed better by losing only about 50% of its earlier seats and 25% seats in 2026 come from fresh wins, while new parties such as TVK and IUML have gained the most from these seats.</p><p>To verify this, I show the seats where there was a swing from each of the major parties earlier, and the results below are clear.</p><figure><img alt="We see that 70% of the seats which swung away from the dmk in 2026 went to the TVK." src="https://cdn-images-1.medium.com/max/1024/1*E-i9dg6mQ_w1Huz0Uo3C6A.png" /><figcaption>70% of the seats that swung away from the DMK in 2026 went to the TVK</figcaption></figure><figure><img alt="60% of the seats that swung away from the AIADMK in 2026 went to the TVK." src="https://cdn-images-1.medium.com/max/1024/1*6OMx1djR5kkWyuZYxu9Eyg.png" /><figcaption>60% of the seats that swung away from the AIADMK in 2026 went to the TVK</figcaption></figure><figure><img alt="We see that nearly 84% of the seats won by the TVK in 2026 comes from the two major players from 2021." src="https://cdn-images-1.medium.com/max/1024/1*EByr0W0ubxxXeAbOanVPow.png" /><figcaption>Nearly 84% of the seats won by the TVK in 2026 have come from the top 2 parties in 2021</figcaption></figure><p>Just to show where the swing happened, I also visualize the constituencies as below.</p><p>As we see below, apart from Chennai, TVK has won heavily in southern constituencies from the DMK (and a few from the INC), while those in the north and west were won mostly from the AIADMK.</p><figure><img alt="Constituencies that swung away from the DMK in 2026" src="https://cdn-images-1.medium.com/max/1024/1*3a38f7zke4QohP5KVVFIAg.png" /><figcaption>DMK has lost more southern constituencies to the TVK in 2026</figcaption></figure><figure><img alt="AIADMK lost more seats in the north and west to the TVK in 2026." src="https://cdn-images-1.medium.com/max/1024/1*hVO0vQ96hPxqOGTtu_hBRg.png" /><figcaption>AIADMK has lost most of the western and northern constituencies to the TVK in 2026.</figcaption></figure><p>So I hope you enjoyed this analysis.</p><p>This was a good and challenging dataset, because it helped me ask questions and define business terms-for example “What is a swing and how do we calculate it”? or “how do we visualize and accurately depict the gain by a party wrt others”?</p><p>I always tell my students that the most dangerous errors are those that do not show up as errors, and need to be verified-one more reason why they are so dangerous is because of the butterfly effect, aggregation will cause the errors to compound and cause huge differences just a few steps down the line.</p><p>This analysis also helped me understand how we can and should use LODs and also their limitations and workarounds, so it was good learning all the way.</p><p>Please do share any feedback for my learning and improvement, and I hope to come back soon with a new dataset with better analysis.</p><p>Till I post again then.. keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=126f70b98442" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Insights and ideas from the analysis of Netflix movies and TV shows data from kaggle]]></title>
            <link>https://medium.com/@srmp1382/i-got-the-idea-for-this-analysis-when-i-saw-a-student-of-mine-showing-a-power-bi-report-based-on-a-eae39fc8d77b?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/eae39fc8d77b</guid>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[insights]]></category>
            <category><![CDATA[movies]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[netflix]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Fri, 15 May 2026 11:17:45 GMT</pubDate>
            <atom:updated>2026-05-15T11:56:30.185Z</atom:updated>
            <content:encoded><![CDATA[<p>I got the idea for this analysis when I saw a student of mine showing a power bi report based on a Netflix dataset from Kaggle.</p><p>There is a pattern to most of the reports that I see on Linkedin where people say they built a “dashboard”, what they mean is they have a set if KPIs that they have shown on historical data.</p><p>While I do take a detailed class on why the term “dashboard” is very often misused, especially at beginners who analyze historical data (and usually with a report on Power BI desktop), my agenda today is to tell a story based on analyzing the movies vs tv shows dataset from Kaggle, combining data on Amazon Prime, Netflix, Disney, and Hulu till 2021.</p><p>This dataset is easily available <a href="https://www.kaggle.com/datasets/shivamb/netflix-shows">here</a> on Kaggle, and has been overused multiple times for learning exploratory data analysis.</p><p>The first thing I noticed is that this data needed to be cleaned, and we need to create multiple pivots from parts of the existing columns-this is not directly possible in Tableau desktop, and it requires a stronger ETL tool.</p><p>Hence I have used Tableau prep builder to do the data cleansing and created a union of the files as shown below.</p><figure><img alt="Data cleansing of the OTT platforms using Tableau Prep, because creating more than one pivot and pivoting based on split fields is a limitation in Tableau desktop." src="https://cdn-images-1.medium.com/max/1024/1*LzHnVtuhN8cygtamd_D_CQ.png" /><figcaption>Data cleansing using Tableau prep builder (this enables multiple pivots based on calculated or split fields not possible directly in Tableau desktop)</figcaption></figure><p>Then I further filtered the data to include just those movies and shows released after 2000, removed blank values, and took the top 3 countries based on the number of releases.</p><figure><img alt="Top 3 countries and top 3 genres in each of these countries based on number of releases year on year." src="https://cdn-images-1.medium.com/max/1024/1*_nKqpdtNcuvEfwonTyTA5w.png" /><figcaption>Top 3 countries and top 3 genres in each country by number of releases each year.</figcaption></figure><p>What we see here is that the US, UK, and India are the 3 markets where maximum releases have happened in the past 20 years.</p><p>We also see that for movies, most of the movies released in the US are comedy and drama, while movies released in India are more international-maybe that indicates that despite all the localization, nearly 50% of the movies released in India annually are international.</p><p>Hence, I believe there is an opportunity for OTT companies to further tailor their content and make it more local for India.</p><p>For the rest of the world beyond these countries, we see that Canada, France, and Japan are where the maximum movies are being released.</p><p>I also checked and found that all these movies were from Netfix, Amazon, Hulu, and Disney had no presence here.</p><figure><img alt="Canada, France, and Japan are where maximum movies are released beyond the US, India, and the UK." src="https://cdn-images-1.medium.com/max/1024/1*axjMDvyDoHUsCmtakC7EeA.png" /><figcaption>Majority of the movies released in France is international, while Canada has more comedy movies.</figcaption></figure><p>What this also shows me is that the more emerging markets, Russia, Brazil, China, do not figure on the list of countries where these platforms operate.</p><p>So again, there is a huge opportunity for OTT companies such as Netflix and Amazon to venture into other countries beyond Europe and India.</p><p>While the top genres across TV shows for the US and UK are similar, I found something very nice and surprising about Japan that I want to put here-The shows in Japan are more about animes than the traditional content.</p><figure><img alt="Japanese TV shows are more about animes and action shows compared to just international shows." src="https://cdn-images-1.medium.com/max/1024/1*6OWxYiCxVIum7u3ToQtHkA.png" /><figcaption>Japanese TV releases are more in the animes genre compared to other countries where more international content is aired.</figcaption></figure><p>Finally, I looked at the average length of the movies and TV shows across the top 3 countries and the rest of the world, and this is what I found.</p><p>The average length of a movie has clearly reduced from around 110 min in 2000–01 to around 92 min by 2020–21, indicating that companies are making shorter movies based on users preferences and shortening attention spans.</p><figure><img alt="The average length of movies has reduced by 8% between 2000 and 2020 clearly indicating a shift in the users preferences and lower attention spans." src="https://cdn-images-1.medium.com/max/1024/1*_t2Mrku7z2cpLW6Ao29Msw.png" /><figcaption>The average length of a movie has reduced by about 8–9% between 2000–01 and 2020 from about 110 min to 92 min across the world-this indicates the company is making movies smaller.</figcaption></figure><p>The same trend is visible for TV shows as well-since TV shows are spread across multiple episodes in each season, I guess this shows that companies are bringing in more change and varying content regularly, by shortening the seasons for each show.</p><figure><img alt="The average content duration of a TV show has reduced from around 8–9 seasons in 2000–01 to around just one season in 2020–21." src="https://cdn-images-1.medium.com/max/1024/1*42HD3U_q8AbndpgCmumh2g.png" /><figcaption>The length of tv shows has reduced on average from about 8–9 seasons in 2000 to just one season in 2020–21</figcaption></figure><p>So finally, here are my learnings from analysing this dataset:</p><ol><li>I strengthened by skill on cleaning data using Tableau prep builder and I realized the limitations of Tableau desktop with cleaning data.</li><li>I learned the importance and use of context filters and how the interpretation of data can change based on the order of filter application.</li><li>I learned that US and UK do more comedy releases, while releases in other countries are more of “international” content, particularly in India.</li><li>I also learnt that the top markets for Netflix and Amazon are the US, India, UK, Canada, France, and Japan-they are yet to become stronger in China, Russia, African and Latin American countries-there exists an opportunity to penetrate there.</li><li>I confirmed what we have been hearing all across for a long time that people prefer shorter movies or TV shows-and I believe it is a trend that will continue.</li></ol><p>I hope you enjoyed reading this.</p><p>I will come back next time with better analysis on a different dataset, meanwhile I would appreciate any feedback and suggestions for my improvement.</p><p>Until I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eae39fc8d77b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Analysis of ATM data from Mar 2025 to Feb 2026 from the RBI data: This time from a different angle]]></title>
            <link>https://medium.com/@srmp1382/analysis-of-atm-data-from-mar-2025-to-feb-2026-from-the-rbi-data-this-time-from-a-different-angle-f8a80cc254c6?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/f8a80cc254c6</guid>
            <category><![CDATA[storytelling]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[insights]]></category>
            <category><![CDATA[banking]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Thu, 23 Apr 2026 11:17:18 GMT</pubDate>
            <atom:updated>2026-04-23T11:17:18.611Z</atom:updated>
            <content:encoded><![CDATA[<p>It has been a while since I have written here, primarily because I ran out of data to write, and also because I had started to focus more on moving to the deep learning and agents side.</p><p>However, I soon realized that storytelling with data is something that I enjoy, and I found myself teaching Tableau once again.</p><p>In this context, I went back in search of data and found that I could analyse the same data from a different perspective to give different insights.</p><p>Hence I have started writing again, and I hope you enjoy this analysis below.</p><p>First off, this data is from the reserve bank of India data for ATM and card related transactions from March 2025 till February 2026.</p><p>Here I attempt to tell the story of the growth of different types of cards and QR codes in the past year for 4 of the major banks (I had analyzed it by type of bank earlier as well, and this was published in a book chapter about 2 years ago so I will do it differently this time).</p><p>First off, I wanted to put that I am using a different approach using measure values and measure names unlike the pivot approach I had used in a lot of my analyses earlier.</p><p>The data for this analysis comes from the RBI, you can access the files <a href="https://www.rbi.org.in/Scripts/ATMView.aspx">here</a>.</p><p>The first thing I did is to check how different types of spending mechanisms are growing over since Feb 2025.</p><figure><img alt="% growth rate of various transaction mechanisms. We see that the UPI and POS methods have grown well, while Bharat QR and ATMs have not." src="https://cdn-images-1.medium.com/max/1024/1*eoHD5Ko7vAySSbACjdE3Zg.png" /><figcaption>% growth rate by type of withdrawal or transaction</figcaption></figure><p>It is very clear that the banks (especially ICICI bank) are actually reducing their ATMs, while the Bharat QR codes have not grown at all.</p><p>The other part that I had highlighted in my earlier article as well, is that most of the banks are growing their number of credit cards faster than their debit cards-this means they are consciously focusing on selling more credit cards compared to debit cards.</p><figure><img alt="Most banks are growing their credit card base more than their debit cards." src="https://cdn-images-1.medium.com/max/1024/1*IttUjMLF9x35f18S8ct2GQ.png" /><figcaption>Credit card issuance is growing faster than that of debit cards, across banks.</figcaption></figure><p>In the light of the graph above, I also found that the number of credit card transactions are growing faster than those of debit cards over the past year, whether for ATMs, POS, or for online transactions.</p><figure><img alt="Number of credit card based transactions is growing faster than debit card transactions for online transactions-SBI and ICICI are growing their online spending faster than POS spending." src="https://cdn-images-1.medium.com/max/1024/1*3CpGpSiL3Wt-FW_4g4pi6w.png" /><figcaption>SBI and ICICI are growing online spending</figcaption></figure><p>What is interesting is the approach that is visible between various banks-Here we see above that SBI is growing its online transactions faster than others while the graph below shows that Axis Bank and HDFC are growing through the use of POS transactions.</p><figure><img alt="Here we see the difference wrt the earlier graph, that HDFC and Axis bank are growing their POS spending compared to their online spending, in contrast to SBI and ICICI." src="https://cdn-images-1.medium.com/max/1024/1*19Iah_M6MPO3QcMFuuvaMQ.png" /><figcaption>We see that HDFC and Axis Banks are banking on the growth of POS spending</figcaption></figure><p>And when we see ATMs, we see the decline in the ATM withdrawals across banks.</p><figure><img alt="The volume of ATMs has declined across banks wrt Mar 2025." src="https://cdn-images-1.medium.com/max/1024/1*qaoZBq7Yb1ef1suzdWk0_A.png" /><figcaption>The volumes of ATM withdrawals have decreased significantly since 2025 March</figcaption></figure><p>When I see the growth rate of the value of the transactions, I see something different.</p><p>The growth rate of online transactions is not differing significantly wrt the type of card, and I found the same story ATM, online, or POS transactions.</p><figure><img alt="The growth rate of transaction value is not differing based on debit or credit cards." src="https://cdn-images-1.medium.com/max/1024/1*WXH3a2fNJtHEvRJAhXYUGA.png" /><figcaption>The pattern of growth rate does not significantly vary wrt the type of card</figcaption></figure><p>Therefore I decided to check if the average value per transaction was differing, whether people using debit cards have grown their shopping value more than those using credit cards.</p><p>And that is more or less what we actually see in the graph, that the average value per transaction is growing strongly in debit cards because the number of debit card transactions is declining, and the total value is not.</p><figure><img alt="The average value per transaction is growing faster in debit cards compared to credit cards because the number of transactions is declining, the total value is not." src="https://cdn-images-1.medium.com/max/1024/1*w-bfZVatnouGmDJOM_p0PQ.png" /><figcaption>The average spend per transaction is growing faster in debit cards because the number of debit cards is declining</figcaption></figure><p>Finally, which spend categories are the fastest growing?</p><p>The answer is clear that people are spending more online compared to POS, and ATM withdrawals are reducing anyway.</p><p>The figure below also shows that out of our 4 banks, SBI is growing all 4 categories well for its credit cards, and while ICICI Bank is also growing, the other banks need to catch up and increase their average value per transaction.</p><figure><img alt="Online shopping is the fastest growing category by average spending per transaction for credit cards." src="https://cdn-images-1.medium.com/max/1024/1*Ak5F5iqvGbiup7uj5Sqstw.png" /><figcaption>SBI and ICICI are growing their average value per transaction in online spending, while HDFC and Axis bank need to catch up.</figcaption></figure><p>With debit card, the fastest growing segment is the POS, and this indicates that people use different cards for different purposes.</p><figure><img alt="The POS segment is the fastest growing segment for average value per transaction. And this is clearly also correlating to the way Axis and HDFC bank are growing the number of POS transactions." src="https://cdn-images-1.medium.com/max/1024/1*u06g5bWRJ-KWGtKgLgtzTQ.png" /><figcaption>The POS transactions are growing fastest in terms of average value per transaction.</figcaption></figure><p>So finally, what did I learn from analysing the dataset?<br>1. The banks are controlling and changing the supply of cards and ATMs differently, however the total transaction value is not differing much.</p><p>This means that people tend to use some types of cards for some purposes-for example, credit cards are growing faster in online use, and debit cards for POS use.</p><p>2. The data shows that some banks like the State Bank of India and ICICI, are growing their online transaction volumes, while others like HDFC and Axis Bank choose to focus on POS transactions.</p><p>However, we see that while HDFC has grown its credit card transactions for POS, the real growth in POS comes from the average value per transaction which is growing fast in the debit card segment.</p><p>So <strong>recommendation</strong>: I believe there may be an opportunity for the bank to focus on issuing more debit cards so that they can tap this growth.</p><p>So this was my comeback post after more than an year, and I hope you enjoyed reading this.</p><p>My learnings from this work were on using the measure values and measure names effectively in tableau, I completely avoided using pivots this time, just for the fun of it and for learning.</p><p>My understanding is that using measure values and names can give more control and flexibility in terms of usage.</p><p>Another very big learning is that I managed to perform the end to end cleanup very efficiently using Tableau prep and demonstrate the whole thing in class.</p><p>So please do reach out with your comments and ideas and suggestions for my improvement and I will take them up and do better.</p><p>I hope to come back more regularly and analyze different datasets to bring out better insights and recommendations next time.</p><p>Until I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f8a80cc254c6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Analysis of primary data from my students on preferences with OTT platforms]]></title>
            <link>https://medium.com/@srmp1382/analysis-of-primary-data-from-my-students-on-preferences-with-ott-platforms-df2076e4c3b9?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/df2076e4c3b9</guid>
            <category><![CDATA[primary-data]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[insights-from-data]]></category>
            <category><![CDATA[tableau-desktop]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Fri, 21 Feb 2025 10:11:13 GMT</pubDate>
            <atom:updated>2025-02-21T10:11:13.923Z</atom:updated>
            <content:encoded><![CDATA[<p>It has been about 2 months since I wrote my last article here, and one of the reasons was that I sensed we were running out of data sources to analyse.</p><p>This led me to understand that we must mine our own data, and that meant primary research.<br>Given that I am a teacher, it means that I must pick small research hypotheses where I can collect data from my students and then use it for our improvement.</p><p>So I wanted to share my findings of working with analysing student responses on their preferences wrt OTT platforms, and maybe this will be a useful guide on how others can follow my route to get useful insights from their own data.</p><p>Before I start, a disclaimer: My data here has only about 40–50 samples, and what works for our students may not work outside unless we roll it out and replicate it.</p><p>This is an exercise that I did in class today and it took exactly 2 hours end to end right from designing a questionnaire to collecting responses to analyzing it to making recommendations.</p><p>I took OTT platforms because most of us use them, and also because it is easy to design a questionnaire.</p><p>Using the Ps of a marketing mix, I came up with some basic questions around which aspects of an OTT might just influence their usage, and I listed them out to be:</p><ol><li>Assortment: Availability of a wide variety of genres</li><li>Freshness: Availability of latest trending content</li><li>Viewing experience</li><li>Pricing and offers</li><li>Customer service</li></ol><p>I may have left out many more aspects like recommendation systems, comparison with competition, etc., and that is because I wanted to keep things simple to get responses easily.</p><p>One of my earlier students, <a href="https://www.linkedin.com/in/simhashini-sharma-5ab880212?miniProfileUrn=urn%3Ali%3Afs_miniProfile%3AACoAADXgy1MBsr6bPc2ubLFWAng8-ENaHL43aZk&amp;lipi=urn%3Ali%3Apage%3Ad_flagship3_search_srp_all%3BOahIOSciQHa1hrhIWNTg0Q%3D%3D">Simhashini Sharma</a>, did a more detailed study on this last year and her results are very instructive, I drew on those and simplified it to show it here.</p><p>I analyzed the results across gender, and which platforms people used, and I put some of the questionnaire below.</p><figure><img alt="Which is the most important aspect for you when deciding the OTT platform? (Choose one)" src="https://cdn-images-1.medium.com/max/406/1*2Fd5hb449-GX0ZmQfzY9Nw.png" /><figcaption>Most important aspect while deciding the OTT platform (choose one)</figcaption></figure><p>For the questions on customer satisfaction, I used a rating scale as shown below.</p><figure><img alt="Rating scale used for customer satisfaction on each factor (1 is lowest, 5 is highest)" src="https://cdn-images-1.medium.com/max/852/1*oapbZ4rXM5qL0Moiq0l57A.png" /></figure><figure><img alt="Rating scale used for customer satisfaction on various parameters (1=lowest, 5=highest)" src="https://cdn-images-1.medium.com/max/879/1*TdXFaOb4BvAygHL7eVEa7w.png" /><figcaption>Rating scale used for customer satisfaction</figcaption></figure><p>Once I collected the responses, google forms shows the stats as below:</p><figure><img alt="Most people use Netflix and Youtube, and people use them very frequently (more than 3 days each week)." src="https://cdn-images-1.medium.com/max/737/1*02nu87c2r7JSTRgCxZVa-A.png" /><figcaption>Netflix and YouTube are the most watched OTT platforms, and most people watch more than 2–3 times a week</figcaption></figure><p>Putting the results in Tableau showed me something more interesting that I can relate to:</p><figure><img alt="Men and women expect differently when it comes to OTT platforms-Women are price conscious, while men expect more service." src="https://cdn-images-1.medium.com/max/1024/1*HB96qBo0Vnpab8ZbNjdehQ.png" /><figcaption>Men and women differ in expectations from OTT platforms: Men look for availability of genres, while women look at prices</figcaption></figure><p>This data shows that women value low pricing, while men look for service attributes such as availability of a wide variety of genres.</p><p>(I had taken an earlier survey with a larger group of students a few months ago on their expectation from their first job after their course, and found a similar trend with 90% confidence: that even there, women were more focused on a “salary”, while men wanted a “good role”).</p><p>The next thing was to look for customer satisfaction scores and the difference between genders: What I see is that on customer service as well as availability of more genres as well as the latest content, men were significantly more dissatisfied than women, while on things like pricing and viewing experience (think high definition images or speed of streaming content, etc), there was no big difference in satisfaction levels.</p><p>This shows that the primary audience in our college students are the men who watch more and are emotionally inclined, so if a provider like say, Netflix were to want to look at what to do to target them, it must improve its service related outputs in these 3 areas shown below.</p><figure><img alt="Men vs women differ on their customer satisfaction scores on various aspects of OTT." src="https://cdn-images-1.medium.com/max/1024/1*CA-S02vAdAQ0xeFQuUlNvQ.png" /><figcaption>Customer service scores for men vs women</figcaption></figure><p>Then given that Netflix and YouTube were the most watched platforms by our students, I decided to compare their ratings, and found something very surprising: that Netflix has higher customer satisfaction on every aspect except availability of a wide variety of content, while YouTube scores there.</p><figure><img alt="Netflix is better than YouTube on every aspect except availability of a large variety of content." src="https://cdn-images-1.medium.com/max/1024/1*U3HprRhtp1hdlk6uZ1Y6wg.png" /><figcaption>Customer satisfaction for Netflix vs YouTube</figcaption></figure><p>Armed with this data, I decided to go and do some chi-square tests in Jamovi to show my students how to confirm whether our hypothesis was statistically significant.</p><p>And I found the p value came to 0.4, which I attribute to a small sample size: As I mentioned earlier, I found a similar trend with 90%+ confidence.</p><figure><img alt="The p value comes out to be 0.4 which shows that the discrepancy which I found in tableau is not significant given the small sample size." src="https://cdn-images-1.medium.com/max/564/1*hpMtQz5w78kaRthiUVTvwg.png" /><figcaption>The small p value shows that the patterns I am seeing in Tableau are not significant because my sample size is barely 50</figcaption></figure><p>I found similar p values for the other factors as well: For customer satisfaction with assortment it came to 0.51 and for customer service experience it was 0.43.</p><p>This shows that my sample size is too small, and I will need to replicate this study on a larger group to get more reliable results.</p><p>I hope you enjoyed reading this article.<br>I will try to get more data and analyze it and write regularly going further as well, and I would appreciate any feedback for my learning and improvement.</p><p>Till I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=df2076e4c3b9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Insights from analysis of the taxi dataset from the Codebasics challenge for Dec 2024]]></title>
            <link>https://medium.com/@srmp1382/insights-from-analysis-of-the-taxi-dataset-from-the-codebasics-challenge-for-dec-2024-035886d8ee30?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/035886d8ee30</guid>
            <category><![CDATA[code-basics]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[insights]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Sun, 01 Dec 2024 11:28:56 GMT</pubDate>
            <atom:updated>2024-12-01T11:28:56.284Z</atom:updated>
            <content:encoded><![CDATA[<p>Keeping current and competitive is very important, as is adding value in what we do. And in this context, I find the datasets from the Codebasics resume challenges to be very useful. They allow us to add value, and we can ensure we are competitive and learning when we analyse such datasets.</p><p>So the latest dataset is about analysis of taxi services across cities, and it yielded quite a few nice insights.</p><p>For a start, I will put down some understanding of how repeat customers behave when compared to new ones.</p><p>I teach in my classes on CRM that loyal customers are less likely to switch for just minor price reasons, and are more likely to continue doing business unless something much more competitive comes up.</p><p>I also teach that companies should aim to bring down the prices for repeat customers to retain them, encourage them to spend more, passing on at least some benefits from the economies of scale.</p><p>And with this, I will launch into my analysis here.</p><p>Now this taxi company operates in multiple cities, so we see that the revenue from the top 3 cities comprises 60% of their total revenue.</p><figure><img alt="Overall, the top 3 cities account for 60% of the total revenues of the company." src="https://cdn-images-1.medium.com/max/1024/1*Zkt2pG7NDAyPgKVTt46OVg.png" /><figcaption>The cities of Jaipur, Kochi, and Chandigarh account for 60.32% of the total revenue</figcaption></figure><p>A little deep dive shows something interesting.</p><figure><img alt="The revenue from new customers is much higher than repeat customers, suggesting that the company is overly dependent on new customers. Therefore I suggest getting repeated customers to use their services more by reaching out to them with offers." src="https://cdn-images-1.medium.com/max/1024/1*S_KjrRMuYEDJdt6VA7hF5g.png" /><figcaption>Repeat customers contribute much less than new customers to the revenues.</figcaption></figure><p>And this pattern bears it out, as I will show in the next few graphs as well.</p><p>So what do I suggest that the company do to increase its revenues from repeat customers?</p><ol><li>I believe it can reduce the prices per customer per ride to get more customers to use the services.</li><li>It can reach out to specific groups of repeat customers and tailor services based on what it learns (this can happen for new customers as well).</li></ol><p>For example, elderly people use taxis for short distances so the company can offer specific packages for them, which have nothing to do with distances or prices, maybe provide a walking stick in the taxi, maybe adjust the seat height accordingly so they can sit easily, maybe accept cash on delivery and not cancel rides for them, maybe train drivers to deal with elderly people, etc.</p><p>Next, I looked at how the city operations teams are increasing their repeat customers month on month. And this shows that the Surat and Lucknow offices derive most of its revenue from repeat customers, while Jaipur and Kochi have a higher cost due to their dependence on new customers.</p><p>The other thing that comes out is that the repeat customer’s usage falls in the month of June for most cities, and for some like Vizag, it starts falling earlier in the month of May itself.</p><figure><img alt="Surat and Lucknow have the maximum repeat customers." src="https://cdn-images-1.medium.com/max/1024/1*xtKS2OPZQp_XvbV2V0TLBQ.png" /><figcaption>Surat and Lucknow have a larger share of repeat customers</figcaption></figure><p>So we know that the total revenue for most cities comes from new customers. However, when we take the value per customer, we find a very different story-that the revenue from a repeat customer is almost triple that of the revenue from a new customer across cities.</p><figure><img alt="Average value per customer for repeat customers is already triple that of the revenue per new customer" src="https://cdn-images-1.medium.com/max/1024/1*xdTEt6nnlFhzJVdUXbnfZQ.png" /><figcaption>Value per repeat customer is already very high compared to value per new customer</figcaption></figure><p>From the graphs above, I believe that the problem we need to look at is to increase the number of repeat customers, and not necessarily to increase the value from each repeat customer.</p><p>So before we go there, I believe we should look at which markets have room to grow, which are expanding. And for that, I looked at which cities have achieved their new customer acquisition targets. What I found was this.</p><figure><img alt="Cities like Surat and Vadodra and Lucknow have achieved their targets for new customer acquisition. Only Jaipur has more room to grow." src="https://cdn-images-1.medium.com/max/1024/1*N0jYrBpOKH67LGL_A51cyw.png" /><figcaption>Only Jaipur has room to grow</figcaption></figure><p>So we see from the graph above, that cities like Surat, Coimbatore, Lucknow, and Vadodara have achieved their new passenger acquisition targets. Given that they already have a higher rate of repeat customer conversion compared to Jaipur and Kochi, I believe these cities are saturated for now.</p><p>I would say that we should be looking for cities which have a larger area, maybe lack of sufficient organised public transport at a local level, and a cosmopolitan community for taxi services to thrive. And the cities of Jaipur, Chandigarh, and Kochi fit the bill.</p><p>So <strong>recommendation to the taxi company</strong>: If you want to invest in more taxis now, Jaipur or Chandigarh are the cities to get into.<br>It may be a good idea to wait and see how the surrounding infrastructure and cosmopolitan community develops in the other cities like Surat or Indore or Lucknow or Mysore, before venturing there.</p><p>The same point is shown in the graph below.</p><figure><img alt="The high growth and value markets are Jaipur, Kochi, and Chandigarh." src="https://cdn-images-1.medium.com/max/1024/1*hyWoxv4jznHg2UCDSyGhqg.png" /><figcaption>High and low growth and opportunity markets</figcaption></figure><p>That brings me to the next question: What does the amount spent by a customer depend on? And the simple answer is the distance travelled per trip, and not on how many times a person travels. <br>I illustrate this in the graph below, where I plot the number of trips taken by the customer and the total distance travelled.</p><figure><img alt="The amount spent by the customer depends on the distance travelled and not the number of trips." src="https://cdn-images-1.medium.com/max/1024/1*X1k-yi72ThIGbfj1fuso2g.png" /><figcaption>The distance travelled, and not the number of times the customer used the taxi, decides the amount that he spends on average</figcaption></figure><p>So my <strong>recommendation to the taxi company:</strong>If you want to increase the revenue per customer in cities that are saturated, you must focus on outstation trips that are long distance-maybe by to nearby holiday destinations, to suburban areas, etc. <br>One example from my experience is that many of the private engineering colleges or newer universities are located in sprawling campuses just outside the cities, due to land availability constraints.<br>Most students live in the city and commute to these colleges usually by college buses from the cities (because public transport in the outskirts is not as good as that within the city).<br>So ferrying students, guests, etc from the city, may be a viable business model in these cities for the taxi companies, especially at odd timings.</p><p>One more question I asked is how ratings affect customer value, and do repeat customers expect less, like I teach in my classes?<br>This time, I was a little surprised to see the answers: that while higher ratings do mean higher value, repeat customers are actually being far more critical in rating when compared to new ones. <br>And it is not just for customers, it is for drivers as well-what we see is that new rides get the benefit of doubt while rating.</p><figure><img alt="Higher rating does mean higher value, and repeat customers and drivers are more strict with rating." src="https://cdn-images-1.medium.com/max/1024/1*LMaUL-Ch03pApH41AIvFbA.png" /><figcaption>Higher rating gives more value, and repeated experiences are rated more strictly</figcaption></figure><p>So which cities have the best passenger ratings on an average?<br>And do passengers and drivers rate in a similar manner?<br>The graphs below show the answers.</p><figure><img alt="The best cities with good average passenger ratings are Mysore, Jaipur, Kochi, and Vizag. The worst ones are Vadodra, Lucknow, and Surat." src="https://cdn-images-1.medium.com/max/1024/1*ldiK9Z3V3XyVFaVrIeSomQ.png" /><figcaption>Best and worst cities by passenger ratings</figcaption></figure><p>The graph above highlights that while Mysore, Jaipur, Kochi, and Vizag have great ratings, Vadodara, Surat, and Lucknow are cities where repeat customers have given very bad ratings.</p><p>Hence <strong>recommendation to the taxi company</strong>:The offices in Vadodara, Surat, and Lucknow need to improve their customer service quality to improve their ratings.</p><p>We see the same pattern with the driver ratings as well, only it is much less critical compared to passenger ratings.</p><figure><img alt="Patterns in the driver ratings are similar to passenger ratings, only they are less critical." src="https://cdn-images-1.medium.com/max/1024/1*O-ahSrm81g4GuMgbCniHkQ.png" /><figcaption>Driver rating trends are similar to passenger rating trends, only they are far less critical.</figcaption></figure><p>So I hope you enjoyed reading this dataset.</p><p>I certainly enjoyed analysing this because while the data appeared straightforward at first glance, it was challenging when I got into the details.<br>When I got into this analysis, I realised that it was not as straightforward, it took a lot of calculations to verify that the data was right.</p><p>Here, I learnt that it is important to test for whether some fields affect others, and then apply the formulae accordingly.<br>The same data presented a little differently, might yield very different numbers, prompting very different decisions, and people looking at the graphs will not even know.</p><p>This was a really good dataset to analyse, and I hope to come back next time with new data in a new domain, with different learnings and insights.</p><p>Please do reach out with any feedback for my learning and improvement.</p><p>Until I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=035886d8ee30" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Insights from analysis of internet speeds of Indian telecom providers over the past year]]></title>
            <link>https://medium.com/@srmp1382/recommendations-from-internet-speeds-of-indian-telecom-providers-over-the-past-year-5ce7dbf6b9e3?source=rss-9d15721aedf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/5ce7dbf6b9e3</guid>
            <category><![CDATA[internet-speed]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[mobile-telecom-internet]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Sridhar Manthripragada]]></dc:creator>
            <pubDate>Wed, 20 Nov 2024 18:25:51 GMT</pubDate>
            <atom:updated>2024-11-20T18:35:41.560Z</atom:updated>
            <content:encoded><![CDATA[<p>Internet is today an essential part of our lives. Much of what we do today at work and beyond uses the internet, whether it be checking our mail, holding a meeting, learning online, chatting or calling a friend, downloading articles to read, or even writing this article.</p><p>And hence it is important that we have access to a decent capacity and speed of internet services.</p><p>I had written an earlier article on the customer satisfaction ratings with the MyCall app, and now, I have analysed the government’s data on internet speed over the past year.</p><p>This data shows the download and upload speeds and signal strengths of various service providers across states, and I attempt to use it to show how various providers are faring, as well as how the market is undergoing a consolidation to form a duopoly.</p><p>For a start, I took data only for Oct 2024 and plotted the data for different network types for upload as well as downloads.</p><p>The data for some states is not showing on the map as the circles in telecom are not exactly mapped, so I will show it in a tablular form as well.</p><figure><img alt="Leading operators by network type for download/upload speeds in each state" src="https://cdn-images-1.medium.com/max/1024/1*uKAoGHbEeWfx13bcpV-oyw.png" /><figcaption>Leading service provider in each state for download and upload for different network types</figcaption></figure><p>We see from the graph above that for 3G, most states are dominated by BSNL (CellOne) while for 4G, we see that Airtel is most popular for uploads and Jio for downloads.</p><p>I show this using a map as well which makes it very clear (some circles such as UP East and UP West and NorthEast, for example do not match states boundaries).</p><figure><img alt="Leading operators for internet speed across states for Oct 2024" src="https://cdn-images-1.medium.com/max/1024/1*U06d9SFpRIbgNFpwkA2Hqw.png" /><figcaption>Leading operators in each state by internet speed in Oct 2024</figcaption></figure><p>So <strong>recommendation 1: </strong>If you are a firm that uploads massive data regularly, maybe you should prefer an Airtel connection. On the other hand, if you are a user who essentially looks at downloads, then you may better off buying a Jio connection.</p><p>Next, I asked how the market changed between 2023 Nov and 2024 Oct, and I drew the graph below for 4G below.</p><figure><img alt="How did the market change between 2023 Nov and 2024 Oct? We see a clear trend of consolidation in the 4G space." src="https://cdn-images-1.medium.com/max/1024/1*aOUofoUZbkMZdUvNBxWqEw.png" /><figcaption>4G market between Nov 2023 and Oct 2024-We see a clear trend of consolidation</figcaption></figure><p>This graph above shows a clear trend for consolidation, and in particular, it shows that Vodafone-Idea has disappeared from its leadership position from several states in the course of this year.</p><p>The same trend holds out for 3G as well. Again, as we see below, VI is being edged out and we have only BSNL leading in all states barring Karnataka.</p><p>Also unlike the 4G segment, we see there is no difference in the leaders in the upload and download categories.</p><figure><img alt="In the 3G space as well, there is consolidation." src="https://cdn-images-1.medium.com/max/1024/1*imYVL3xUiM-L2h0-bUIyEw.png" /><figcaption>3G market between 2023 Nov and 2024 Oct: We see a clear trend where VI is being edged out of its leadership position</figcaption></figure><p>So <strong>recommendation 2:</strong> BSNL is doing very well in the 3G sector and should continue edging into the 4G segment as everyone else moves into 5G as 4G becomes a low cost inferior service.</p><p>And <strong>recommendation 3: </strong>We see that Vodafone Idea has lost much of its leadership in this space, so it must step up its advertising and compete with BSNL by using its historical positioning as a leader in the customer service space, one area where BSNL is weak.</p><p>And finally, I want to speak of horizontal integration.</p><p>I remember, about 10 years ago, we were just starting out with WhatsApp and a lot of people used to ask us for our “WhatsApp number” even though they already knew our mobile number. What this meant, was that we were treating the mobile internet or data category as separate from the mobile telephony category.</p><p>However, it soon became clear that almost everyone uses just one mobile sim card for all the functions including accessing data and calls, which means we do not have a “separate” WhatsApp number anymore.</p><p>For our data on internet speed, it means we must combine it with the call quality data, if we want to understand the full picture and identify what the opportunities are, and where the market is going.</p><p>And in this context, I faced the same challenge that I have been mentioning repeatedly earlier, that most of the datasets are now being gated and made non available, even on the free government portal.</p><p>For example, I used to be able to access the MyCall datasets month wise earlier by downloading in parts — now, however, it is made into one large dataset, with restricted access and its large size is an additional problem.</p><p>In this case, I keep getting taken to a downloads history page, where it keeps saying my download is in progress and it never happens, maybe a day later it says my link expired.</p><figure><img alt="Unable to download datasets easily compared to earlier" src="https://cdn-images-1.medium.com/max/1024/1*u2asST4vJfpOk_O1xZtr6g.png" /><figcaption>Unable to download datasets easily compared to earlier: datasets are now being gated more and more</figcaption></figure><p>So finding granular data easily for analysis is now a bigger challenge than analysing it, and I believe it is important for us to have control on our data if we want to use it, so I will put my efforts into this area to ensure I can get access to relevant fresh and useful data.</p><p>Anyway I hope you liked reading my analysis.</p><p>I hope to come back soon with a new dataset in a different domain and get better insights and make useful recommendations.</p><p>Please do share any feedback for my learning and improvement, I will be happy to implement.</p><p>Until I post again then, keep reading!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5ce7dbf6b9e3" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>