Dinesh Parwani’s Post

𝐄𝐱𝐜𝐢𝐭𝐞𝐝 𝐭𝐨 𝐬𝐡𝐚𝐫𝐞 𝐦𝐲 𝐫𝐞𝐜𝐞𝐧𝐭 𝐩𝐫𝐨𝐣𝐞𝐜𝐭 𝐨𝐧 𝐝𝐚𝐭𝐚 𝐚𝐧𝐚𝐥𝐲𝐬𝐢𝐬 𝐢𝐧𝐯𝐨𝐥𝐯𝐢𝐧𝐠 𝐚 𝐜𝐨𝐦𝐩𝐫𝐞𝐡𝐞𝐧𝐬𝐢𝐯𝐞 𝐫𝐞𝐭𝐚𝐢𝐥 𝐝𝐚𝐭𝐚𝐬𝐞𝐭! 𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧: This project aims to provide valuable insights into various aspects of business performance. By leveraging data analysis, the goal is to identify key patterns and opportunities for optimizing business strategies. 𝐎𝐛𝐣𝐞𝐜𝐭𝐢𝐯𝐞𝐬: 𝐒𝐚𝐥𝐞𝐬 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬: Evaluate sales performance across different dimensions such as time, geography, and product categories. 𝐂𝐮𝐬𝐭𝐨𝐦𝐞𝐫 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬: Understand customer purchasing behavior and identify top-performing customer segments. 𝐓𝐢𝐦𝐞 𝐒𝐞𝐫𝐢𝐞𝐬 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬: Analyze trends and patterns in sales and profit over time. 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞: Assess the performance of different product categories and sub-categories. 𝐐𝐮𝐚𝐧𝐭𝐢𝐭𝐲 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬: Investigate the quantity of products sold across various categories and locations. 𝐓𝐚𝐫𝐠𝐞𝐭 𝐀𝐜𝐡𝐢𝐞𝐯𝐞𝐦𝐞𝐧𝐭 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬: Measure the extent to which sales targets have been achieved and identify top-performing categories. 𝐏𝐫𝐨𝐣𝐞𝐜𝐭 𝐋𝐢𝐧𝐤 :-https://lnkd.in/gHEgMc-4

Statistics is something I’ve been studying recently. I'm sharing a few things while reading Practical Statistics for Data Scientists by Peter Bruce and Andrew Bruce. Here are a few core concepts I’m working through: Say monthly Spending on Coffee: $15, $20, $22, $30, $35, $40, $45 1. Standard Deviation: Measures the typical distance of each spending amount from the average (mean). Tells us how spread out our spending values are around the mean ($29.57). 2. Variance: Similar to standard deviation but without taking the square root; it’s the average of squared distances from the mean. It shows the intensity of the spread in squared terms. 3. Mean Absolute Deviation (MAD): The average of the absolute differences from the mean. Gives a straightforward sense of how much monthly spending varies from the average. 4. Median Absolute Deviation (MAD from Median): Median of the absolute differences from the median value (which is $30 here). Shows typical variability around the median, less affected by extreme values. 5. Range: Difference between the maximum and minimum values: $45 - $15 = $30. It tells us the full spread of spending values. 6. Order Statistics: Ranking values from smallest to largest (e.g., $15 is minimum, $45 is maximum). It helps to understand data distribution by rank. 7. Percentile: Shows the relative standing of a value. The 75th percentile here is around $40, meaning 75% of values are at or below $40. 8. Interquartile Range (IQR): Spread of the middle 50% of data, calculated as Q3 ($40) - Q1 ($20) = $20. Gives a view of the main chunk of spending without extremes.

Our Commercial Data Science team does such a great job on these bi-weekly pulse reports. This time we delve into the retail industry. Each report contains key macro insights and industry research. Read now! https://bit.ly/3TNErDM #economy #experian #industrytrends #retail #CommercialPulseReport

𝗔𝗻𝗼𝗺𝗮𝗹𝗶𝗲𝘀 𝗮𝗻𝗱 𝗢𝘂𝘁𝗹𝗶𝗲𝗿𝘀 When preparing data for analysis, we often notice some unusual numbers that required investigation. They are known as anomalies and outliers. It's important to understand what anomalies and outliers are and to identify the differences between them before addressing them. 𝗔𝗻𝗼𝗺𝗮𝗹𝗶𝗲𝘀 are data points that occur outside the expected range of values and cannot be explained by the base distribution. 𝗕𝗮𝘀𝗲 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 refers to the normal pattern that data follows. Anomalies are often caused by invalid data. 𝗢𝘂𝘁𝗹𝗶𝗲𝗿𝘀 are data points significantly different from the rest of the data. These are values that deviate from the other values in a dataset but can be explained by the base distribution. The main 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝗰𝗲 between an anomaly and an outlier is that an anomaly is often an error or a rare, unexpected event, while an outlier is an extreme but expected value that still belongs to the pattern of the data. 𝗘𝘅𝗮𝗺𝗽𝗹𝗲𝘀: 𝗔𝗻𝗼𝗺𝗮𝗹𝗶𝗲𝘀: - A sudden spike in website traffic that cannot be explained by any known marketing campaigns or events. - A sudden drop in sales for a product that has been consistently selling well. - A sudden increase in the number of errors in a system that has been running smoothly. - A customer who is aged 200 years old. 𝗢𝘂𝘁𝗹𝗶𝗲𝗿𝘀: - A top student scoring 100% on a test, while the class average is 70%. - A house significantly larger and more expensive than other houses in a neighborhood. - A stock experiencing a sudden price change not in line with the rest of the market. - A customer who is 99 years old. 𝗣𝗼𝗶𝗻𝘁 𝘁𝗼 𝗥𝗲𝗺𝗲𝗺𝗯𝗲𝗿: Anomalies and outliers are not always errors or mistakes in the data, they can also represent genuine variations in the data. If you find this post useful, do like or comment on it. #DataAnalyst #DataAnalysis #Outliers #Anomaly #Difference

Get to know your data before you start analyzing. Before you can find insights in your data, you need to understand what you're working with. Take time to explore your dataset, look for patterns and oddities, and get a feel for what you have. Imagine you're given a dataset on customer behavior. Taking a few minutes to skim through the data, check for missing values or outliers, and get a sense of the overall trends can save you hours of headache later on. Familiarizing yourself with your data upfront makes the analysis smoother and more effective.

This is a case study I performed for an ecommerce household appliance company I interviewed at recently. While I have masked sensitive/confidential information, the rest of the work presented remains unchanged. I gained a lot of valuable experience performing this analysis. I was given a dataset filled with numerous null values, anomalies, discrepancies, and generally miscalculated fields. After cleaning, validating, and organizing the raw data I began my exploratory analysis starting with the big picture, uncovering trends and finding general patterns and outliers present in the dataset. From there I got even more granular and began a segmentation analysis, using the different dimensions available to me, to find what key drivers were possibly influencing these trends and consumer demand overall. I had to make sure I could communicate these findings in a digestible fashion, so I visualized them utilizing a mix of line, bar, column, area, and pivot tables/charts to help tell my data story. Finally, I offered what my recommendations would be to the relevant teams at the company using my analyses, assumptions, and hypotheses as support. At the very end I documented my technical process to counter the old adage "garbage in, garbage out". If you have any feedback or thoughts, please leave them below as I'm always looking for new methods, perspectives, and ways of looking at big data and how to tackle it!

📈📊 Beyond graphs and Charts, the analyst compares current data against existing trends and frameworks to give insights. And with some certainty, predicts the future… 😂Yes, we play “God” sometimes. For context, imagine what you can achieve if you knew with (92%) certainty that running a (10%) discount campaign in the first (2 weeks) of December would boost your MoM sales between (70% to 130%) by (January 2025). - Notice the brackets? Those are the value an analyst brings. Anyone can look at graphs and charts… The true magic is in the insights drawn from its interpretation. If you briefly describe a market data you’ve been collecting in the comment section. You might get a free review and brief analysis 😊 I’m sure that many of us are looking to read something interesting…

The goodness of fit in a media mix model is a tricky problem. Every business is going to have a different normal level of error: in some businesses, you might be able to achieve a super tight fit. In others, that may simply not be possible because there are too many other sources of variation beyond marketing spend or factors that you can include in the model. But the most important thing in an MMM is not the goodness of fit with the data that you see. The most important thing in an MMM is its ability to predict data that it has not seen – its ability to forecast new data. We do look at the goodness of fit for in-sample data (data that the model has seen). We use it as a filter, though. We use it to say if it's not doing this, something must be really wrong. But doing that well is just table stakes. The real challenge is to feed it data that it hasn't seen. Let’s say today is June 14th and I want to fit a model up to today. And then, in a month, I want to come back to July 14th and ask what it would have predicted having seen this data between June 15th and July 14th, and then compare that with actuals. That's the real challenge. And ideally, it should be able to get about as close fitting that data as it does for the in-sample data. That's a very high bar, but that's what we're always shooting for.

I used to think data told the whole story. I was wrong. Last year, I had a call with one of my clients saying "our sales numbers look terrible". I extracted the quarterly data, I dove in, ready to uncover the problem. The data was clear: our conversion rates had plummeted. I spent hours crafting the perfect presentation, complete with charts showing our decline. I was ready to be the hero who identified the issue. Then came the meeting. As I confidently presented my findings, the sales manager raised her hand. "Did you know our main competitor had a flash sale last month?" I froze. No, I didn't know that. She continued, "They slashed prices by 50%. We couldn't compete." In that moment, I realized a hard truth: data without context is just numbers. I'd been so focused on the what, I'd completely missed the why. Since then, I've changed my approach: 1/ I talk to people and understand their challenges and perspectives before diving into analysis. 2/ I ask "stupid" questions. Lots of them. 3/ I consider external factors that might not show up in our data. Now, when I present findings, I include both the data and the story behind it. The result? My insights are more valuable, and our decisions are more informed. Numbers are powerful, but they're just one part of the picture. The real magic happens when we combine data with human insight. #DataAnalysis #BusinessInsights #DataStorytelling

Ever waited for the "perfect" market data to make a decision, only to realize it never comes? Early in my career I learned that the ‘perfect’ market analysis is likely not going to happen. You'll never have all the information you think you need. At some point, you have to pull the trigger and make a call. The best analysts are the ones who are comfortable with uncertainty. Who can take incomplete data and still piece together a picture that's directionally right. And that's often all you need to make a smart decision – pointers in the right direction. Here’s my best tip to deal with this challenge: -> Define the "Minimum Viable Analysis": Get clear on what data and information is critical versus what’s just nice to have. Often 70-80%, or even less, is enough to make a solid decision. Sometimes you can move in days or weeks instead of months this way. ------------------------------------------------------------------------------- If this resonated with you, follow me for more.