The DataHour: Data Science in Retail

Key Takeaways

all right so we are ready to begin hello and welcome everyone to another session of the later hour series we are thrilled to be here with you this evening um that is going to be full of action-packed learning i am samridhi chaturvedi part of the data science team at analytics video and the chat moderator for this session is going to be sahil so for those of you who have joined us for the first time a brief introduction to the data r series so it is basically a series of webinars conducted by analytics vidya and it is led by top industry experts it is a fun way to understand the concepts of data science from the leading players in the data tech domain and as the name suggests it's one r dedicated to data so we are hopeful that these sessions are going to be a great source of enrichment and value-adding for our community members now on to a session today which is data science in retail so retail refers to the selling of goods and services to customers for individual consumption as you all might be aware and the key to retail is that after purchasing purchasing from the retailer a customer consumes the goods and services and does not resell it so in this data are chaitanya will explain some common retail terminologies that you would have already heard about during your last visit to the data to the big bazaar or something and then proceed ahead to discuss some of the applications of data science in detail i hope you are excited to attend this data hour session with us and before we kick things off and i hand it over to a speaker um a couple of things that i would like you all to remember that we are recording this session and we'll make the recording available uh in a few days on a youtube channel also please do use our q a section for asking any questions that you might have during the session and thirdly we will share a poll about the feedback of the session towards the end so i'll request you all to fill it up now on to a speaker in this session of data r we have chatena agarwal with us so chetan is passionate about solving complex real word problem statements using data science and he did his bachelor's and master's from iit madras and is currently working as a data scientist at shell india he has six years of work experience primarily in the retail domain a very warm welcome to you sir so over to you chattanooga the world the virtual stage is all yours so this session is about use cases of data science in detail i am chaitanya agrawal data scientist currently working with shell business operations in bangalore so what is retail so let's talk in the common terminology i believe almost all of you would would have visited shops in your neighborhood from your childhood only so if you visit to a shop to buy a grocery from a local grocery store you are taking part in the retail transaction so basically you are buying some goods from a business owner for your own consumption so this is simply retail and what is retailer the shop owner who is selling the goods to you he is a retailer who is retailing the goods to you so retailing is it includes all the activities involved in selling goods or services to the final consumer for personal non-business use so retail retailer retailing are almost like [Music] interconnected so let's go to the next slide so types of retail here i'll cover three important and i would say most commonly known types that you would see in your daily life first is traditional trade so traditional trade is a mom and pop stores that you find in your neighborhood whether it be a grocery store whether it be a farm pharmacy store or any other store which is operated by a single shop owner he owns that store so that is a traditional trade it is also known as general trade and traditional trade basically is available in almost entire country in india if you talk about india it's present in tier one city sphere two cities tier three and villages as well so traditional trade is ubiquitous but now if we talk about modern trade so uh let's take the example of india it is d mart in big bazaar but if you go to us you'll find walmart kroger or if you talk about uk there it is costco in australia it's wolfsworth so these are few of the examples of modern trade so modern trade means they have a chain of stores from which you can go to any of the store if it is present in your city and you can buy any goods and services which they are providing so modern trade basically hasn't penetrated deep into our country in india at least it's mostly present in metros tier one and some of the tier two cities and mod so traditional trade has been present since attendee modern trade came into the picture later on and latest entry to the scene is e-commerce so any website like amazon flipkart alibaba they all are e-commerce websites and e-commerce is also a type of retail where you can directly buy from the amazon so amazon is doing retail trade with you here here again you are the consumer and amazon is procuring books from the seller basically amazon is a marketplace but yeah it is kind of a retail only so let's move the next slide here let's briefly go through our use case of data science in detail uh retail so why data science needs to be used in detail why we uh like what was the use case that was first envisioned like why people started thinking of how we can maximize our efficiency we can maximize our revenue so there's several use cases which are there in retail so if you talk about your daily life i believe uh everyone would have used google photos so google does image classification there google can tag your images according to which person is there in the photo or you would use language translators so language translator can translate translate any language x to language y so similarly data science also has use case in detail so think about the case if you are a shop owner or in traditional trade and you want to store some products in your inventory now what happens if some of the product goes out of stock and a customer comes in and he asks for the same particular product so you don't have the product you can't sell the truck to the customer so you are facing out of stock problem so data science can help predict out of stock so [Music] uh [Music] so you are not audible your voice is breaking i think there's some wi-fi issue at this end [Music] upon this again guys i think there's a lot of uh power cut issues with today in bangalore due to heavy hands so uh am i audible yeah so do you want us to share the screen yeah please you can share the screen otherwise it will be an issue again it has been raining heavily since quite a few days uh so there are a lot of foreigners today no worries so sir he will be sharing the screen yes foreign so i was talking about uh auto store problem with the retailers so basically this is one example where retailers can stock up the product based on the recommendations they get and they can avoid auto stock issue and thereby reduce the loss of revenue that might happen second is recommendation engine so if you shop on amazon you see you search for a product you get certain recommendations and that's a recommendation built built in within amazon website so it's also a part of use case of data science in retail so you see there like i would say infinite number of use cases i couldn't list all of them so here i will talk about four use cases firstly use cases space planning market mix modeling and order book preparation i'll give a very brief overview of this three problem statements and the fourth problem statement trade promotion optimization i will focus much deep on this topic which is i'll explain in the end so we can start with space planning can you go to next slide so space planning i suppose all of you would have visited any mall in any of the cities so what do you observe when you visit a mall how the layout of the mall design so let me ask a question i believe uh i can see some answers from the chat so why uh you think uh food court and multiplexes are placed on the topmost floor of the mall can i see some answers from the audience so uh yeah correct so bjorg has given the correct answer so people with it all the floors so basically that is the funda behind space planning what a retail store designer retail store layout denier wants to do is that he wants to visit he wants you to visit each and every floor of the mall so that there is maximum visibility given to all the stores present in the mall thereby increasing the chance of a purchase so if you are just going suppose you wanted to go to the mall and mostly highest footfall is received by foot code and multiplexes so there are high probabilities that if you enter a mall you'll go to the topmost floor where foot code is situated once you go there you would you would have to pass through each and every floor so you will observe all the surroundings of the mall and that way it will maximize the sales for the mall owners so what is happening here is uh objective what they are trying to solve here maximize the time spent by a consumer within a store so similarly to mall you can think of any modern trade store like uh costco walmart big bazaar dmart if you visit any of the stores you will find that the layout is designed in a very data driven process you won't find things placed here and there without any part given to it so there is some part given to that and based on the planning all the excuse the place on the stream so space learning is basically which you can use to design the layout once this techniques are once the designs are implemented the layout is implemented uh you'll see that uh there should be some increase in revenue and thereby they take into consideration how much was the revenue gain and then they adjust the retail loan accordingly so this is a feedback loop which they follow and let's see some of the techniques which are followed in space planning so as soon as you enter the store you will see you first see decompression zone so what is decompression zone so decompression zone is a space which is situated just inside the door as soon as the customer walks in why a decompression zone is necessary so many people would think that it's a waste of space there's no products there are no products placed there nothing is present that space is blind so what is the use case of decompression zone so decompression zone is required because retailer wants you to enter the store and forget everything about the outside world he wants you to focus on your shopping experience you he wants you to observe all the products place in his retail shop or retail store thereby increasing his margin because if you are focused on the shopping you will buy things and it will lead to increase in revenue for him so that's why the decompression store decompression zone is very much needed if you are planning to create a retail store second thing is a food code and multiplex example are already given third one [Music] it so the product would be placed at the top floor he wouldn't be have been able to see the product so that's why the placement is important and it also comes under space planning fourth is uh most action which is very attractive in the store and the fourth one is complementary products so um complementary products uh sorry is my voice breaking just a second thanks thanks for the feedback oh uh okay i'll turn off the video i believe that will prevent the bandwidth okay so uh the last example is complementary products are placed within reach for increased chances of impulse buying so example swing gums are placed near cigarettes bread is placed nearby butter so these are the like uh techniques using space planning team to maximize the revenue and it is one of the use cases of retail can we go to the next slide so a second use case is market mix modeling so basically what is market mix modeling uh suppose uh you want to advertise some of your products if you are a brand you want to advertise some of the products you do advertisement on tv you do advertisement on print media you do it on like uh youtube ads so there are various avenues where you can do uh advertisements of your product you can do various kind of marketing and promotional activities to promote your product so market mix modeling is a decision making tool used by brands to estimate how much worthy was that marketing initiative which you took so suppose i'm spending in three different marketing initiatives and i want to see what was the return on investment on each of the marketing initiatives so market mix modeling is a tool which helps you does that so it breaks down the business matrix to differentiate between contributions from uh different drivers so the drivers can be like a print and media ad or digital spans or any other marketing activity that you would have done and second one is the base drivers so suppose a brand is doing no activity no promotional spend nothing then also that brand gets sold why because that brand has built some brand value in its market so customer knows that that brand exists so incremental drivers helped to propel the cells further but base drivers can help business sustain the sales even though there is no marketing if the brand has enough equity in a brand equity the is the brand is enough brand value in the market so why mmm which is market mix modeling is required what is the importance of it so it helps the marketing guy to do better allocation of the marketing budget it helps identify if i am doing marketing using three different channels which channel is giving the most revenue or how much should i spend on this channel so that my return on investment is maximized is also helps the marketing person to better execute the ad campaigns so it all like uh how much i should spend on each of the marketing channels to avoid saturation and third it also helps marketing person to do business scenario testing so suppose if i inve increase my spend on this channel by 10 how much will be the increase in sales so marketing market mix modeling is also very common use case in retail industry you will find each and every company doing it for their products because they want to maximize their spend they want to maximize their roi for whatever spend they're doing so market mix modeling is a very i would say must to have product for every company who is doing retail service retail sales uh let's go to the next topic so retail order book preparation in traditional trade so have you wondered how the order book is prepared of a grocery store which is present in your neighborhood how that owner of the grocery store places order for the products how the products are delivered so let's uh imagine the scenario if suppose today is uh 5th of september it's the starting of the month so i already have some product stock in stock in my inventory now by end of the product by end of the month some of the products stock would have got depleted so what will happen now a person from the wholesaler team will go to the retail shop he will ask the retail shop owner for what items you want to place the order for what excuse within each category you want to place the order so retail shop owner will give him all the quantities of each of the excuse he wants to purchase and retail sales representative will give some suggestions based on the uh trending excuse in the market within that region he will give some suggestions it's totally on the retail store owner whether to buy those uh suggestions or not so this is how the cycle work that it then the sales representative goes again back to the wholesaler and makes the delivery of whatever products that the order book had but this process is not very optimized you can see there are many like inefficiencies shopkeeper may not know what all things are happening in the market which product is in demand retail sales represent you may not know what is the total uh like how much sales any product is making making for us in the same geography but in some other nearby region so there are lot of inefficiencies in the process so retail order book preparation is a tool which can help sales representative to suggest to the shop owner that this much amount you should purchase for this all this excuse so suppose some xqx is very hot sold item in some other region so this retail store order book preparation tool can give that insight by making use of data science it can build all models and it can basically give a good basket of products with how much quantity that the shop owner should purchase so that the revenue of shopkeeper shopkeeper is maximized and also the revenue of the wholesaler is maximized so this problem statement is an optimization problem statement in which the revenue is maximized by optimizing the number of sqs within a category to be sold and quantity of each of each of the sq to be sold by each store so it suggests right mix of sq based on seasonality so if it's a rainy season he will tell uh show up owner that you should stock this much item x of odoma's mosquito repellent you should also stock this item which is a very hot selling item in this region so you should also stock it it also can help shopkins to avoid out of stock in-store issue and it can improve the portfolio breadth of the shopkeeper so this is how this tool can help shopkeeper maximize itself create a good order book so that the decision process is very much data driven there is no like a gut involved basically the sales representative will give some suggestion this process will not totally replay the sales representative and the shop owner relation but it will be a decision augmentation tool it will provide some support in the decision making that they are doing on day-to-day basis so let's go to the next slide yeah so let's talk about the promotions in modern trade so can you guys identify what each of these promotions mean can you identify can you name all of the promotions that you see on the screen for each of these four feet pictures can i see some answers from the chat has correctly identified one of the pictures buy one get one shory also so let me explain so here on the screen you can see four types of promotion on the left most picture you can see it is a display promotion so basically some items are placed on a secondary uh display apart from the regular shelves that you can see in a store so uh it is a part of display promotion this buy one get one everyone has correctly identified it it is a multi by promotion third one here half price which was seen in the middle bottom it is temporary price reduction which is a discount of 50 percent and the right most one is feature promotion or catalog promotion so here you see various uh products are featured in a newspaper advertisement or a template template sorry template so let's go to the next slide so what are these types of promotion let's understand in detail temporary price detection is self-explanatory so for a temporary period of time for a small period of time shop the retail store provides some discount on the items like it could be 20 discount or 50 discount or any percent of discounts so this temporary price reduction it is temporary in nature you don't see it price half price throughout the year because if store owner is pricing all its item at 50 discounts throughout the year it's no more a promotion it's the price it's the new price of the products so for the uh promotion to work the promotions should be offered in silos they should not be offered throughout the year or very long continuous period of time because otherwise then the promotion will lose its value so this holds true for all kinds of promotion so first is temporary price reduction i hope everyone would have understood it second is multiply this is also self-explanatory so you buy two brush you get one brush free buy two get one buy one get one any type of combinations uh third one is feature and catalog so you see templates newsletter emails any of this mode of communication where a product gets featured about which invo which informs customer about the promotions running in a store comes in the feature promotion and the fourth one is display so basically display the secondary placement of the product in addition to placement in ielts uh so display basically what it does whenever customer enters a store if it sees that some product it's placed on display it catches attention of the customer so that's why displays place so if you visited any uh modern trade store you would have seen many products placed separately in a corner or in some distinct location where you can very easily identify that product is placed on a promotion because some special treatment is being given to that product let's go to the next page please so this type of promotions can happen with or without price decrease so feature without display display without feature feature and display this three kind of product promotions can happen with or without price decrease so if a feature promotion is happening it can happen along with price discount or it can happen without price price discount as well so prior discount is not a mandatory condition for this three feature without display display without feature and feature and display kind of promotion activities in the fourth one price decrease only here only price decrease happens no display no feature no multiplier nothing happens so multi buy is also one of the promotion types here let's go to the next slide so trade promotion optimization so basically whatever we have talked about right now about the different kind of promotions i wanted to give some background before we discussed trade promotion optimization so that's why we talked about all the different types of promotion so now let's talk about what is trade promotion first so trade promotion is a marketing campaign which directly influence product performance so it occurs in store to target direct shoppers actively making purchases so market mix modeling is bit different from trade promotion so trade promotions have you will see only when you visit a store so if you are visiting a store you see a product 50 percent on off so you will see that promotion only when you visit the store so that's what the second line means so this promotions runs within the store and they target direct customers who are making active purchases within the store so why trade promotion is needed this multi by 50 of display feature because you want to boost performance of products where some of the locations are underperforming so there's a location where some products are not performing expert expectations so you provide some discount to improve the sales or you want to shift some established product into new markets or you want to introduce new products into an existing market so for various sorts of reason you do trade promotion now again what is trade promotion optimization then so to design the promotion calendar of an year you need to know what kind of promotion i need to place in which month or in which week i need to place what kind of promotion this is not a random activity that you can place multiply on the fifth week of the year and then place display promotion on the 20th week of the year it doesn't happen randomly there's a science behind it which is very much data-driven so data trade promotion optimization is basically a data driven process which can help brand maximize roi on their promotion spend so brands spend a lot of money while doing this promotion spend so this money whatever promotion happens is generally paid from the pockets of the manufacturer so if you see any buy one get one or display promotion there is a very high chance that for that promotion money came out of the pocket of the manufacturer not the retailer sometimes retailer can also do promotions on it itself by itself but generally the promotions are paid by the manufacturer because they want to boost their sales and thereby increasing sales of every uh entity in the sequence like wholesaler and retailer as well so basically trade promotion optimization tool helps key account manager also known as cam to decide at what price to promote and when to promote so tpo helps cam to create price promotion calendar or i would say promotion calendar so according to promotion optimization institute 2019 state of the industry report cpg company spends between 11 and 20 percent of revenues on trade promotions so 11 to 20 percent is a very big number so if they're spending so much of their percentage of the revenue on the trade promotions therefore definitely expect a very high roi on their investment so while doing trade promotion optimization we need to quantify the impact of each of the promotion types on total sales for so for example if i'm doing any sort of promotion what is the uplift in sales generated because of that promotion you want to know that because you can't randomly do the promotion you are paying money for it so you have to really you have to be very much data driven so whatever type of promotion that you are going to do you should know how much is the expected increase in sales that will happen which will kind of be a return on investment for their money which i am i will be spending on the promotion so to do the uh quantification of the impact we study about sales decomposition which is an essential part of trade promotion optimization so trade formation optimization itself is a very big tool and sales decomposition a very is a very important module of the straight promotion optimization so let's go to the next page so what is sales decomposition so suppose if i am the key account manager if somebody asked me no so sorry not keep on another if i'm the product owner or i'm the marketing guy or i'm the person who is responsible for the sales of a product x so if sales of product x is down let's suppose some person within the company asked me why the sales of this product is wrong i would say it is due to drop in advertisements or but the advertisement guy will say maybe the distribution of the product itself has come down means the product is not readily available in the market so maybe some other person will say that maybe the pricing and promotions weren't good enough so how do you get answer to all of this questions so you can clearly answer all of this question by doing simply doing the sales decomposition so you get the data you would see how much uplift was generated by each of the promotion activities you see who is the culprit here what is the advertisement or was it the fault of wrong pricing and promotion strategy or the distribution itself was not present so these are the answers which we get by doing the sales take up transition so since there are multiple business drivers that impact sales we need to understand how volume impact of each of these drivers can be quantified right so uh if we talk about mainly cpg industry which is consumer produced goods so it in this cpg industry you have products like your hand cleaner bathroom cleaner sanitizer soap so all this products comes in the cpg industry so uh we have a database which is provided by nielsen so nielsen is an agency or is a company which collects this data from all the retail stores about the promotion activities that are happening in a store about the distribution and the current they collect uh i would say hundreds of variables they derive hundreds of the variables from the data they collect from the store and nilsen database help us to quantify all these retail drivers let's go to the next page so let's talk about distribution so if you are purchasing the database from nielsen you would expect that volume so volume sales will be definitely present because volume cells is a number one measure of how the product is performing so in week 1 was it just 100 units of sale over the 200 units of sale so volume sales is definitely number one measure but the second most important measure is the distribution so if your product is not available in the market at all you give whatever promotions anything there will be no sales because the product is not there on the shelf at all so for customer to buy the product it has to be present on the shelf right that's why distribution is also very important aspect of this analysis you should know how much is your product distributed in the market in which all retailers stores it is strong otherwise there is no point of doing all this analysis if distribution data itself is not available so how distribution is calculated by nelson basically nielsen supplies a variable named percentage ac distribution which gives us the distribution information so it tells us percentage of stores selling our product but this percentage is not simple weighted it is uh it is not simple percentage it is weighted average based on the product size uh sorry store size so let's take an example you see a table here let's say product x we have three stores store one store to store three and we have product x product x is being sold in store two and store three but it is not being sold in store one so if you want to calculate how much is the acv distribution percentage seo distribution of product x you will do 60 plus 80 divided by 180 which comes out to be 78 percent so how this is calculated percentage asu distribution as i said will be calculated for only those stores where the st product was scanned at the counter what it means if the product was not sold at all even though that product isn't shelf or it's not on the shelf but it wasn't scanned at the final counter billing counter if there's no data for the sale of that product it will be counted as uh zero distribution from that store so that's why uh here for the store one when there was no sale of the product or the product itself was not available on their shelf uh it was not translated for the acv distribution calculation and that does how the suv distribution came out to be 78 so let's go to the next slide please so what is sales decomposition [Music] your total sales can be divided into two components one is base sales and second is incremental sales so what is base cells so base cells in estimate of cells that would have happened anyway if there was no promotion so if there is no promotion whatever cells are happening those are base cells but even though when promotion is happening the sales that would have happened without even the presence of promotion is based still so i will clarify later on the next slide i have a plot to clarify better but to simply understand base sales is a sales that would have happened irrespective of whether the promotion would have happened or not so it is similar to whatever base driver that we saw in market mix modeling so what are the drivers of base sales there are mainly four important drivers first distribution second is regular selling price which in case it is known as mrp in india seasonality and computer impact so if there's a good distribution of your product in the market it will affect your base sales your price point of the product will affect base sales the seasonality will affect base sales like in the rainy season uh sale of umbrella or mosquito repellent goes up so seasonality affect bases and computer impact also affect base cells how your computer is pricing let's go to the next slide so incremental sales so if let me first explain the base cells again uh through this plot the plot you see on the right side of the screen here you see that uh week three has some sales which is quite higher than the week one and week two so in week three we did a promotion that's why the sale was very high again in week seven the sale was very high and again the last week the sale was very high so this three weeks were the weeks when there was a promotion happening in the store and whatever is the extra sale which was seeing is known as incremental sales so even for the promoted week 20 units of sale would have happened even if there was no promotion so this 20 units of sale is a base sale for this week when the promotion was there in the store correct so incremental sales happens due to the four promotions that we already discussed so total units now can again be explained by another formula as well so total units equals to incremental sales plus base cells or incremental factor into base cells so they the reason why we have defined another formula so incremental factor basically is calculated by doing a regression which i'll explain in the next slide so how it is calculated incremental factors equals to exponential of promotion type into its uplift coefficient what is promotion type so for the temporary price reduction discount percent is the promotion type so you will impute the value of discount percent and you will multiply by its uplift coefficient of the discount to get the incremental factor raised to the power of e to get the incremental factor for the temporary price reduction for display feature and multiply you have acv display only acv feature only acv display and feature combined in your acv multiply so corresponding acvs of display feature and multiply you will input here and that will give you give you incremental factor for each of these three promotion types right so let's go to the next slide so uh this equation is very important while doing uh sales decomposition so what all things we see on the scale i think the formatting has got messed up so on the left most side you see log of v log of v the volume so basically you want to estimate how much each of the drivers will affect the volume sales of your product so p is the price of my product cp is the computer price d is the distribution percentage suv distribution pr is promotion uplift pr is sorry promotion it can be any type of promotion beat uh multiply discount percent any kind of promotion and cpr is computer promotion so basically uh this all are the drivers we believe will affect the volume self of a product so for example if you see uh i think many of you will get the reference uh maggi noodles is sold in india by nestle so i think a very long bag 100 gram of pack of the maggi noodles used to cost used to cost around rupees 10 then slowly 70 gram of the packet used to cost for rupees 10 then same 70 grams packet was sold for rupees 12 now it has been sold for rupees 14. so you see what has happened here nestle company kept on reducing the size of the maggie product by keeping the price constant so we can have a coefficient a variable for pack size also or the size of the product also if you think that is also a driver the change of the driver can change the volume cells so you can include any kind of driver in this equation if you believe that will impact the sales of your product what this regression equation gives us it gives us elasticity coefficient so uh i think the formatting got messed up so b1 coefficient is the own price elasticity b2 coefficient is cross size velocity so what is elasticity let me explain that first elasticity is percentage change in volume for percentage change in price so you everybody knows that if brand increases price of a product by let's suppose 10 percent it will directly affect the sales volume sales of that product right so how brands make decision how much price increase they should do so that their volumes are not affected to an that their revenue goes down so this is a very uh data-driven process brands look at the elasticity of each of their products and then they decide how they should increase the price of their product or how much promotion they should do basically own price elasticity means when you do changes in the price of your own product and you want to judge the effect on the volume you will use one size velocity now if suppose a computer product of the same type is doing any changes in its price it will also affect your sale because for example if a computer is pricing its product it decreasing the price of its products from uh 10 units to 9 units then uh that product is now has gotten cheaper so definitely yourself will go down because he is providing a discount on his price so his product sales will go up so cross price velocity helps us calculate that distribution velocity i have explained many times like if you increase the distribution or decrease your distribution it will affect the volume sales of your product similarly promotion uplift so here you can do any add all types of promotion you can judge the effect how much uplift each of these promotions are doing so uh promotion also you can add and also the computer promotion so this regression equation you see it's a log log linear equation so log of v which is volume equals to log of price computer price log of distribution and uh but promotions of computer and the own product are not log why it so so if you see if you take partial derivatives of the volume with respect to price first own price it will give you dv by dp uh so basically it will give you the formula of velocity which you see on the bottom right so when you do a partial derivative of this regression equation with respect to each of the variables you will get the coefficients and to get elasticity you need to do delta v by delta p into p by v so this calculation happens automatically if you do log log linear regression here so v1 is the elasticity coefficient so uh i will give one more example so let's suppose uh milk is one of the products so elasticity can be can vary from minus infinity to plus infinity mathematically but in practical terms it varies generally from i would say minus five to plus five there's no uh any hardcore range i can define but average or you can say it differs from minus five to plus five so if you decrease your own price volume will go up so uh b1 coefficient which is own price velocity the sign of it is negative but if computer increase its price your volume will also go up because your product has become comparatively cheaper so the coefficient of b2 is positive so like this you can derive the science of each of this coefficient as well and that will help you understand how each of these drivers are affecting your volume cells right let's go to the next slide so this is the last slide so uh let me present a scenario let's suppose there is an item for which we are doing this regression analysis and there was no price change in the entire modeling period history of that product so if there is no price change you cannot calculate the elasticity of that product right because to calculate the elasticity you need to know what was the price change and corresponding to that you can calculate volume change so because of the uh constant price of a product throughout the modeling period history you cannot calculate the elastic coefficient so how does cam will make its decision on price changes for this product how he can make an informed decision if there's no elasticity present for this product x so to solve this problem what you can do let's suppose you buy a hand cleaner of rose fragments of 250 ml and uh for this product you see that the price change happened quite frequently in the past so you see earlier it was sold for let's say 100 uh rupees now it being sold for 80 rupees and now after some months you will you would have seen it was sold for 120 rupees so they the uh fluctuation in the price throughout its history but another product maybe uh lime fragrance was sold sorry i got it opposite so rose fragments there was no price thing at all in history for but for the line fragments there was constant change there was continuous change of price in the history so what happens then you see that both kind of hand cleaners are exactly same for you you can buy rose fragments also you can buy lime fragrance also of the same pack size right so the price is also same just the fragrance its difference so for you or for the key account manager the elasticity of the line fragments will almost be same as the elasticity of rose fragrance right so if you do not have the elasticity coefficient for the rose fragrance you can substitute it with use of lime fragrance elasticity because in its price history there was a continuous change in strice so that's how you can copy variables you can copy coefficients from the similar products which are there in the market so if suppose this product would have not also seen the price change you can pick another similar product so there's a hierarchy of products first comes the country another uh then comes the retailer manufacturer segment brand so the hierarchy if you don't find any replacement for your coefficient in the same hierarchy you go of one level up if the replacement is not present in that level as well you go one level up again so uh i agree that as you keep going one level up and up the elasticity will not be a totally match uh match for both the products but definite but it will at least provide you some decision point to make a conscious data driven decision so this is how you can uh do substitution of the coefficients if you haven't got the coefficient from your equation so uh this concludes my presentation i i think uh i explained all the concepts uh clearly apologies again for the continuous disruption that happened because of the loss of wi-fi so i believe this session was informative and useful to all of you so any questions that i can take up okay sorry to interrupt sir so before we take up the questions i'll be sharing a poll and i would request you all to fill it thank you sir okay sure oh yes guys you can please ask the questions if you have any there are two questions i guess in the q a section so sarah was asked why we are using log log regression and why using logs for elasticity so basically uh as i explained through the equation log of volume equals to let's suppose a constant plus a into log of price this is the equation right so if you do a partial differentiation of this equation you will get exactly the formula of velocity and that's why log log regression is done so basically last city is defined as percentage change in volume divided by percentage change in price so the coefficient that is present along with the price variable that coefficient gives you the velocity if you do log of volume equals to log of price plus constant so this is how this is why log log regression is done [Music] any other question uh yes there are three more questions in any section uh okay i was looking the chat sorry okay uh we cannot classify b2b business as a retail business no so in the regression formula why are we taking log to adjust the scale i think i just answered that which data science algorithm apart from regression can be applied in detail sector and how they are implemented so in the recommendation engine which i talked about in the very beginning uh you can apply like any algorithm exe boost random forest so uh there's no like uh i would say limitation of the algorithm so this another very important and very common problem statement is demand forecasting so the big bazaar in india is a retail stand so they want to know how much demand they want to estimate the future demand right they want to know how much stock they should stock up so to do the demand forecasting again you can use any type of algorithm depending upon the quantity of the data and the relationship between between the variables within the data so there is no limitation on the type of algorithm you can use it depends upon the problem statement for this sales decomposition regression is used but if you talk about any other problem statement you could use any other algorithm as well depending upon the use case any other question okay guys so are there any more questions or should we wrap up okay i guess so no more questions okay so thanks a lot sir on behalf of analytics vidya i would like to thank you for your time and for delivering such a wonderful session um i'm sure our audience found it insightful and hopefully we can conduct more such sessions with you in the future