 Good afternoon, I'm Shikhar, working for Capricade Technologies. I'm expertise in integrating business analytics, statistics, with the technology to empower the business managers. So today's session will be more focused on how your favorite retailers can take money out of the analytics. Now this is the dicey question where is it the technology helping the business team group, or is it the statistics, or is it the analytics? My answer would be like it's a combination of all the three which is making the business more prominent and more technical. Since yesterday we have been having a session on banking, finance, media, and this is more concentrated on retail. Retail business where day to day we go to a shopping, we have a food and grocery, we buy apparels, we go to a restaurant, we order a pizza. So all this comes under retail analytics. Here we are going to focus on the retail analytics area where the whole outcome of this presentation would be or take away three insights, possible three things which can go out of this room and just implement it as immediately as possible and improve the favorite radius of business. So this do include the techniques what has been implemented and how that needs to be implemented, how that needs to be executed, and what is the impact due to implications of these techniques. The biggest breakthrough in this new era of retail analytics is integration, integration between a business person with a statistics and a technology. I'm not sure how many of the technicians, tech geeks are here and analytics geeks are here or business managers here. Can I have some hand raises for tech geeks, statisticians? So this session offers a whole set of audience where it will be concentrated on both of them. The first breakthrough is a real-time engagement model. Have you ever imagined the step in your store, the store staff, the retailer has received you very well with the name, with the purchase what you have done, what is the impact of it, he takes a feedback of the products, he takes a feedback of the suit you have bought in the previous purchase and then he recommends what is the best product for you. Seems surprisingly very easy with the unorganized states. When we go to our type recipes or type recipes or the villages, the retailer knows the name, what we do, what we like, do we like suit, this is like his very, his up-to-date on all the information. Can we implement the same kind of a thing on a bigger retailer, like a Big Bazaar or Shopper Stop or a Lifestyle where the store staff is more attracted to the consumer with the real-time technology. So this real-time technology has been improved the retailers by 4%. Then going in the further sites will understand how that 4% has been increased using this technology. The second part is, we know the data is coming from social media, the data is coming from e-commerce, the data is coming from online, lots of data, tons of data. Does it make sense if we integrate the whole things? Now, how it can be helpful? As we heard in the yesterday's of your sessions also, that if you are making a transition in your credit card, EBOE, 1 lakh or 2 lakhs, there's an immediate call comes to you saying that as the purchase has been done by you, or is it a fraud? The banking industry is very much mature enough to have this technology in place. But the retail is very new. It's very new due to two reasons. One, kind of a scope, kind of a data which was captured in the past. Now, due to the technology improvement in the current decade, you have started capturing the data. How that can be helpful? As simple as that. If a customer stepped into your store, has given you a mobile number and an email ID, and he logged into your Facebook after a day, where he has actually allowed the credentials to share with the, let's say, the Puma, or a jewelry like, jewelry like Ketanjali. So, he shares the information with the Ketanjali. And after the 10 days after that, he changed his marital status to an engaged. How do you feel if someone calls you up, he interlinked your email ID, he interlinked your mobile number, someone called you up, so it's good to have, good to have that your engagement has been fixed. Can you have some shopping at a jewelry shopping at DTanjali store? So, the quality of information is there. The events has been tracked. The consequences of these events can be tracked very easily. So, here we will be talking about some integration between all these techniques. Well, we are talking about personalization and customization. This is on the basis where customers as an individual, not a group in it, will go further into this customer as an individual, how the Bayesian prediction models and linear regression models has helped retailers to increase their sales. To give a brief introduction of the retail industry, retail accounts to 15% of the Indian GDP, which accounts to 470 billion. And there are 11 shops out of 1000 people. But of course, having said that it is 470 billion GDP, there is only hardly 3 to 4% as an organized sales. Now today, as of today, there is the data which we are capturing of the whole population in India is only 3%. Now, as the organization sales keeps on improving, that kind of data will be captured will be very higher. 60% comes from verticals other than food and grocery. True. So, always the analytics is more constantly in food and grocery. How we can implement to cut up? How we can implement to cut up? Is that business? How we can implement that to cut up? How we can implement that to an apparel business? How we can implement that to footwear sales? And as behind the scenes, it's very important that analytics can be helpful everywhere. So, what is the challenge in retail industry? One is, is it really the data hygiene is the issue? Or the lack of it? I mean, how many of you feel that the data hygiene is the issue in the retail industry? Okay, lack of it? So, you are seeing that there is a 6000 TB of data has been captured in the year of 2011. Compared to internet data, hardly it seems like internet data has been captured on a day. But it's much very valuable information where the complexity, calculations and complexity predictions will matter in the retail industry where we are going to focus today. Now, connecting the customer online, offline and social media is very important because seeing an innovation has been making the transactions across. The other thing is playing with the state of data. Now, coming to an example like a grocery industry where there are 1000s of products, 1000s of products. It's very difficult to understand which product is moving with which product. Always we do in a categorical manner which category is moving with other category. But is it, is it, is it, there is a scope of improving, a scope of improving the business through the retail data? Yes, there is a huge scope of it. Now, why is it a limited tool to analyze data is more tools have been developed by the big MNC for a banking industry but there are very few less tools for a retail industry. So, what are the typical business problems in a retail? Typical business problems in a retail hardly have a campaign conversion up to 2-3%. There is a campaign, there is a promotion which is executed in the retail industry hardly 2-3%. But really speaking, a banking industry have a higher scope of conversion state because the fact that there is less competition, there is less general rate, the customer going away is very less because he stick to the brand. If he changes to other bank he need to change his loan account, he need to change his investments, he need to change a lot of things. Whereas I buy a footwear in a woodland and tomorrow itself I can go and buy footwear in a puma. Whereas in an Indian industry, that's a problem in an Indian industry where no person is more loyal towards any brand. Can any people is agreeing with me that no person is a brand to any particular brand. So, it's loyalty loyalty comes into the picture where Indian people are more relaxed and will keep on shopping across the brands. So, that was the biggest problem. Now, how we can tackle that problem? Can we do something when the customer is in the store when he is already purchasing? Can we do some in-store engagement activity? How that can be achieved through analytics? How that can be achieved through technology? It can cover in the next couple of slides. The second problem is high communication cost and less relevance. We have seen in the most of the developed countries the communication cost is up to $1 for the direct mailers. And even the text cost in the EDM cost is very higher. Yeah. We have heard about a propensity model. Can you lift up to 2, 3 times of the lift? Yeah. So, the propensity model is based on one objective saying that how many set of customers how many set of customers can respond to a beauty way beauty products campaign? How many set of customers can respond to an EDM? How many set of customers no appliances can be? No. As we keep on counting there will be hell lot of production there will be hell lot of propensity models which needs to be built. Can we build something which can cover whole set of marketing initiatives? In our experience we have built our 15 propensity models we are going to detail afterwards. But if we have built 15 propensity models to tackle the marketing initiatives for the whole one year it's not an ad hoc but it's a whole one year they have made a prediction models to make sure how they are coming in the future. The third problem is the customer cooperate. 80 to 85% it's like huge hardly 50% of the customers are your loyal customers. Where are the remaining customers are going on? So, need to engage its customers neatly by understanding the customer in a better manner. Just to show the prediction algorithms we have categorized. We all know marketing. The kinds of marketing and the kind of analytics we do right now the kind of analytics what we do right now is we define a marketing activity which needs to be planned in the next couple of years for the next couple of months. What that marketing activity should be? What do we do? We do logistics, we do predictions we do discriminant models to predict when the customer is going to come what the customer is going to purchase. Now there is the after marketing activity where we understand the kind of hypothesis what we've considered before marketing activity is valid or not. Use text mining to find out to concise the feedback feedback analysis, to concise the customer complaints and we all thought about during install activity which is the execution has happened the plan of the execution has done in between the execution and before execution after execution there is one more level of technology came into picture due to the cloud based solutions due to the real time technologies which is during install activity we will engage when the customer is in the store so that's a real time power of the real time results how does it works now taking example as a grocery store we went to purchase on a grocery store and he purchased a basket of basket of goods analyze the basket of goods what exactly is present in the basket what exactly is not present that fits into the recommendation where the current recommendations in general the typical traditional way of processing is we compare what product is going with what as simple as that which is we are currently covering what really goes with what to marry it with a business objective there are two other points which is getting missed out is what the customer is missing in the basket that's something the customer need and it is getting missed out in the basket one, two have you understood have you ever understood have you ever thought about does this customer need it or is it a push from the marketing so that's two major points actually driving the business to the next level so here with the four factors which is what is missing in the basket what really goes with what customer past purchase data and business objective so all this happens in in a store where the billing happens so in the meanwhile the scanning of the scanning of the scanning of the whole items is happening whole this real-time algorithm works out what the customer needs so the system throws up with the instant suggestions and it shows off what is a real-time institution how that needs to be executed that's an independent business impact which is the retailer is making money how it flows this customer stepped into a store he purchased so after he purchased we understood what he needs we understood what is missing in the basket and then we value the patronage of the customer and we offer you something at the store so where the conversion rates can go up to 5 times or 6 times really that makes wow to the customer but how it works so this is more on the statistics way and more on the approach how we are but let me tell you in a very simple way so we have considered the products in a grocery store what are all the products available in the grocery store pick a just grocery what is the percentage that B will be going with A if A is present in the basket as simple as that as all the statisticians love to work on the two levels of layer now the complexity comes in where I want to define the three for layer 4, layer 5 to understand what exactly is there in the basket so this is more on the using the base probability but the same kind of a thing can be done using the appropriate algorithm but because the pros and cons of both algorithms are much different here in this context we are using the base in probability to understand if the what is the probability of C buying C if the A and B are already in the basket what is the probability which is in the past history of the customer and past history of the other customers we can get to know what is the probability of ABC together by probability of A multiplied by probability of B saying that A is already in the basket so this can be extended to the n layers which we call it as a n layer basket where it will be keep on moving if the customer purchase today he purchase A and B and D so immediately the A flows A to B, B to D and then the layer 4 comes in the picture what A, B, D and what else can go with the customer's interest so all our information I am sorry we have one of our product where all the information the customer is storing at the store this is the customer's current basket or the other company so this is currently analyzing the data which is currently presented in the basket current basket now what can be done so after this the second step of the algorithm is where it will check now we have studied about the missing basket now the second step would be what customer needs it really the product E is given by the customer is a need for the customer when the product E has been purchased how many days back whether it has purchased 10 days back or 20 days back if it is a 10 kg of Ata which has purchased 10 days back there is no point in giving initiative on the Ata again right now so what will happen is it will learn about the past history whether when that particular last product has been bought is it is the real need of the customer or not the second step of it now so we can build up to as many rules as possible and at the end of a bill or end of a slip you get as a bill that the whole missing basket can be printed out of these could be the possible missing baskets in your can in your del no that is the system rules which has been taken care of so once should I answer it right now or at the end of the session once it is done I will take that question so there is a selection rule and a rejection rule which we are talking about the past purchase and behavior is very important very important in some kind of rejecting something or selecting something and all this happen on a click where the scanning is happening so now we have studied about how we can through the real time technology we can increase the conversion rates up to 2-3% we have seen in the studies it has actually went up to 20% and raise the customer's retail this is the 4% now coming to the interesting part would be the interesting part would be actually there is a flow problem so we have studied about the philosophy stuff how we can change the things in a departmental store right now the departmentals should have a different set of different varied categories men's wear, women's wear and home appliances, electronics that has different set of products in a departmental store so can we do something which is suitable to the whole market for the whole one year not just for one of the month or one of the day or one of the week so there comes a big complex data along with the marketing offers so more or less the marketing has been in the saturation state for the last 50, 20, 50, 100 years everyone knows what actually needs to be executed everyone knows what needs to be done and what should not be done so taking those into consideration can we mingle, can we merge with that with the big and complex data so how that can be done we have approached in a different manner where there are 15 propensity models there are 15 marketing different offers so typical propensity model will give a result of we will make sure we are targeting 20 to 30% of the customers but this kind of approach will make sure one customer we get two or three offers at least and the best offers best marketing offers to them so 15 propensity models has been built based on the 15 marketing marketing initiators using logistic and disconnect and then the primary challenge in terms of statistics is merging all the 15 propensity models into one and assigning the best one offer one, offer two, offer three against each customer and that has been tagged for the whole one year and it can be executed at any point of time without any lag lag in the execution so after applying this we can do a design of experiments how this impact can happen where the random way of doing a propensity model with one propensity model actually can give a lift up to only two to three gaps but when actually multiplying this up to 15 propensity models and make sure all the customers sets of customers come in the picture 20% conversion where the profit would have been multiplied by 25 times so now interestingly departmental store we have studied and now can come into a pizza business how that works we all like a pizza and are allowed to have an order every week or every week so pizza loyalty program is more than a group more than a group rather than an individual needs to be changed into your family loyalty program internally what we have done was to tag in the customers so instead of treating as a customer we treat it as a family the whole set of four people belong to one family how this can impact the business is when I'm doing a direct mailer I don't know the customer one and customer level are the same but while using this all this regates algorithms and phenotic and edit distance we have merged the customers the whole merging has actually merged 20% of the customers which actually decrease the marketing cost by 20% simply without any impact, without any changes it has actually decreased the 20% marketing cost coming to this pizza business who is your customer so typically we do a clustering or segmentation and say there is my high level customers there is my high standard customers there are my discount seekers, there are my potential customers can we do something more than that are we missing something do you like to have a chicken takeout pizza every time so this is not actually taking care of the segmentation here this is not taking care of any chicken takeout pizza or your favorite pizza or your garlic bread or your pasta what we have done is we have took a taste clustering also into picture which is saying you tend to have the same pizza every time or you tend to have three different pizzas alternatively how you change your pizza preferences comes in the y-axis dimension and then further we have built down into 17,000 segments taking the significant variables from the price to a model price to a test with seven dimensions now interestingly in pizza we have understood that the stickiness of the pizza is up to 60% the same person ordering on a Saturday evening last week and he tends to order the Saturday evening again so that's a very brief insight which actually made us to make a high level of segments where the customer will be treated as a unique individual rather than a group so how it works in a week so how it works customer made it first purchase second purchase and third purchase now after this last purchase after this last purchase when is the next order can you predict that using the number of customer visits the last purchase micro segment is the way it belongs to customer frequency using linear regression models all this actually tend to make a continuous engagement with the 17k so this is actually avoiding a marketing person to get away saying actually there is a change in that there is a difference in do the campaign today don't it's a different way of approach where we see that the customer is customer customer is actually a customer not a job businessman he actually adds a value to it so make customer engagement plan rather than pushing something to the customer so this linear regression technique actually takes care of external events like cricket and football matches also so how we can grow the retailer's business so if we have more data decades back 20 years back we don't have any data in the retail where we see we acquire the customer engage the customer retain the customer and grow the customer how the lifetime value is to be increased it's like negligible what happens if non analytics data driven marketing happens where there are lots of bulk things that's happening to you bombarded to you without any analytics so it can grow up to a certain after that it can drop off but we ask how the how the if you are doing with an analytics way it is going to help your business and retain your business quite a long time so here we have a very good examples where all customers are treated similarly $1 investment has given a $8 of ROI written and when the customer has treated some greatly without 17k segments $1 investment has given a $9 simply by introducing a linear regression technique to understand predict when he is going to order a visa make an improvement like $1 investment has given another $20 so this is a real case study of how we have grown the business to the retailer in the past thank you any questions one of the problems so regates is a regular expression it compares the strings so it can extract the house numbers it can extract land mask for you so regates expression so this is more on the statistics side and the technology side where our regular expressions can be used one of the problems I see is integrating with augmented views so how do you manage so we have we are majorly in India where we have integrated most of the power systems because of which the whole online data and the offline data can be integrated across the whole platform so that is actually helping in making this get an engagement plan very well so you tell us about your behavior of the user yes yes are you guys doing the cross promotion no, no, no we do for one of the business in India only one editor just wanted to ask are you specific to tough examples like visa and repair house names like online shopping so real time is very easy so I will go to my question in this case you are associating a cluster number when you actually associate this with the online online shopping how do you associate the end user attributes so the end user attributes can be captured or can be registered in the store can be registered in the e-commerce website or in the social media whenever it could be the registration can happen in one shop so currently the real time association what I am talking about is in the store yeah, so this is going to be about point of sale yeah, point of sale because point of sale is a very difficult part because the integration comes into the picture with the different power systems should I so simply your and your brother and your father would have an order but the address would happen see yeah, so if someone wants to a direct mailer campaign mentioning over a postal letter and send you a cost instead of sending three direct mailers to your home we will send him only one so in our database currently using that matching the address algorithms we figured out that there is a 20% set of customers who tend to order from a different number where simply the marketing cost got down by 20 years I think we can take the question