 Just talk a little bit about how we can better use data for whatever we do in our lives. I'll focus on that question in the context of SMEs, the small and medium enterprises. The first question to ask is what are SMEs? I mean, it's something that we keep hearing about it, a lot of development organizations, governments keep on talking about it. There is sort of a definition by government, which is about any company with an investment of, say, less than 10 crores. If it's a manufacturing plant or if it's a service company, there's a certain definition. There's a slightly different definition. But essentially, these are small and medium enterprises, say, with a turnover of anywhere between 10 lakhs a year to about even up to, say, 50, 60 crores a year. When I was growing up, we had this very textbook picture or at least picture of SMEs as a sort of a hand-loom kind of thing. The reality is, I don't know how many of us have sort of gone deep inside there, but things have changed. A lot of them are fairly automated, fairly mechanized things. In this, here you have the analogy from moving from a hand-loom mill to a more automated or mechanized mill. The other aspect is that the fact is they are an important part of our economy as of now. Now, a few realities about SMEs that sort of make them a little different from very large businesses. One is most of them do have some kind of digitization of the workflows happening. But let's say if I run a manufacturing unit, I would have tools to keep track of each of the workflow steps. But the tools that they use are very local custom-made tools. They aren't the standard ERPs that you would see around the industry or in a large corporation. What that also means is that the SMEs don't, no, I mean, the SMEs are also usually typically family-run businesses run by a single entrepreneur who may not necessarily understand data well. As a result, the decision-making is very, very intuition-based. The traditional business intelligence tools that you keep hearing about it are not really used in that context for a bunch of reasons. One is that expensive. They have a very steep learning curve. If you have to figure out how to customize SAP to a 50-crore textile manufacturing unit, I'll probably end up spending my life doing that. And obviously they have a huge cost there. These enterprises prefer somebody, even a local IT vendor, who can build the tool for them. And I'll just take a quick jump here. As a nation, we have learned how to digitize workflows, digitize processes. We know how to take a paper process and map it into an IT system. Even a local small software vendor can do that for you. But the next level of jump, learning what to do with the data, which is what this conference is about, that skill hasn't become the common place yet, which is why we sort of struggle with how to better use data. I'll talk about SMEs in a specific context, in the context of sewer textile industry. But whatever I talk about, I strongly believe what I sort of, yeah, is something that can be applied in all domains. It just helps to have a vertical domain to think about, to think through when you're talking about solutions there. And then you can easily adapt the solution to a broader context. So quick again, D2. SURA is about a 400 plus or 500 plus year old city, a large trading port. The reason why all of these factors might be important to understand is it's a psyche of people there, small families and businesses. Lots of them, about 15,000 wholesale trading firms. So you don't have the advantages of a large company which can sort of invest in a large tool. And most of them there have digitized workflow. Again, just so you understand the historical context. This is a map where you don't have Bombay but you have sewer. It used to be a trading port back in. We don't have a Bangalore here either yet. I don't know, it's 1700 somewhere. Again, sort of an illustration of the typical synthetic claw that you wear, the manufacturing process. Don't need to go into details but to illustrate the point that there is a fairly complex process that has to be sort of looked into to figure out how data can play a role there. This is a six-step process that I talk about. You begin with a petrochemical byproduct. Make it into a proper yarn as it is called, the threads. Give it into a fabric. And then dyeing, printing. And then a lot of embroidery that goes into it to get you a shirt that I'm wearing right now, for example. Now, what role can data play there? A lot of the use cases that I'll talk about aren't sort of in any way. Actually, all of the use cases that I talked about are fairly well-documented algorithms that are used in, say, if you go to a typical business school, people will talk about supply chain management. There are these standard algorithms there that are used there. I think what my key premise is that you can use these algorithms in a context that they haven't been used with a little bit of customization, a little bit of adaptation in that context, and result in something which sort of significantly improves efficiency of what's happening. So some sectors in which you can use data. And the data that I'm talking about is data that businesses already collect. A typical Sari manufacturing or a Sari wholesaling unit will know what is in what step of production, how much of a fabric they have, and say how much of a POY they might have bought, how much of it has drawn for sizing, and how much of yarn that they have when they go down right now, how much has gone through each of these processes at any moment. Now, how can you use all of this data to plan? You also have the sales information, the seasonal, annual, daily sales information. Now, how can you use this information to run your business better? A few use cases that people talk about forecasting. I think they've heard a few talks about it. It's a little different in this context that I'll dwell a little bit on. Production planning, pipelining. It's a pipeline manufacturing process. How do you sort of ensure that the pipeline is optimally used and sort of meets the sales requirements? Product costing and pricing. So if you typically go to the SMAs, they have a very thumb-off rule way of planning each of these things. Hey, my sales last year was so-and-so. Maybe this is this time of the year. Perhaps my sales would be a little bit like what it was last year. That's where the decision-making about sales forecasting ends. Planning. I have stock, which is I need to sort of have this much sales in the next two months and maybe sort of my production turnover is say two months and hence am I overproducing do I need to produce more for the next two months of sales? Costing and pricing is very, very thumb-rule based. These are the X levels. These are costs at various steps and hence my product cost would be X percentage more than that. It's all very, very thumb-rule based. You have multiple channels of sales and marketing or grand-building. How do you evaluate which of them is giving you the best ROI? Well, a little more on each of those. Time series analysis, for example. It's fairly well-established research concept that you can use for historical sales prediction. A lot of standard algorithms there. But the domain, the applicability or adaptation to the domain becomes important here. If you talk of a synthetic clothes sale, there's a peculiar pattern to it. This is the flight data, but fairly increasing the percentage of flight data. The sales sort of pick up sometime and say April and May when you have the first sort of marriage season normally. And then you have another pickup in October November when people are purchasing for the marriage season. So when you sort of do a seasonal prediction of this, you sort of have to say, this is the sales for last year and I'm sort of trying to adapt it for this year. I sort of have to map the dates. So Diwali is here and Diwali time is sort of a peak time. So next year Diwali is in a different month or a different date. I kind of have to stretch that timeline there. There are these sort of peculiar adapts you have to do to sort of get to something. And in this context, as with I think all seasonal sale predictions or prediction models, you will have a sort of a linear trend which is bringing in the macroeconomic factors and then you have a seasonal variation there. There are other things that you could do here. You could sort of cluster products or classify products into various categories and then track predictions there. Again, in the context of SARIs, I'll talk about there are two kinds of broad kind of synthetic fibers of SARIs that you'll see. There are printed SARIs and there are embroidered SARIs. And printed SARIs are normally lower priced, daily wear kinds. Embroidered are more heavy work and more expensive types. And the bandit season or the festival season, the embroidered ones pick up as you would expect. Whereas the other time of the year, the printed ones pick up. So you might have to... In this context, I do feel the need to sort of go into this level of classification or clustering of various product types to get to a more accurate analysis. And in any domain, I think the point that I will continuously emphasize is that whatever domain you apply these skills to, you have to sort of understand the domain well enough to extract these patterns or extract these patterns. Sales forecasting was one thing that we talked about briefly. There is something about the production planning and pipelining. Let's say you have an end stage production process and you know the turnover time for each of these end stages. So based on how much of what you have in each step, you can sort of predict how much of a... Let's say I have a two stage production process to simplify things. The first step has a sort of a turnaround time of T1 and second step has a turnover time of T2. Other way around actually. T2 is the first. So S2 comes before S1, whichever way. So I know how much of a sort of stock I would have cumulative production I would have in a certain time frame, say now to T1 to T2. I can easily compare that to the sales forecast and identify the fact that this time frame, as things stand, I'm under producing so I need to push more stock into the production pipeline to meet this requirement. Is that sort of clear? I know till time T1 the stuff that's there in S1 stage will come out as ready product and from T1 to T2 the stuff that's there in S2 stage will come out. So I know what I'm expecting as supply from my side I've sort of done some kind of demand forecasting and then I can see where, am I over producing or under producing? A few caveats that I'm sort of sort of things that I'm hiding under the carpet here. Obviously prediction you don't have a single line, you have a range, you have sort of minor error variations but those are details that I didn't want to talk about in a 20 minute presentation. If you were to do a live sort of a system which we do, you have those ranges and error factors built in as well. So this is for the finished sales. You could do the same thing for each level of production. See if you have, you want to engage your, keep your production pipeline completely engaged. So are you producing enough in the first stage for the production pipeline in the second stage to be sort of fully occupied? Costing and pricing, again, sort of if you map each of these production steps into a cost factor, and cost factor can be dependent on the product. So a certain kind of printing will cost different from another kind of printing. What you have from each of these end steps of production is that you have sort of the direct cost of the product. You have things about inventory cost if the product is tied up as inventory at various stages. Indirect cost are the organizational cost. Now, you can sort of, again, model these well and get a cost model. You have the sales model, which is some of the product sales at the price that you set. Some product goes at a discount, some product is just thrown away as wastage. So you can sort of equate the two and figure out what the pricing model would be. A little more logical and database than a gut-feeling-based model that you often see. Sales and marketing costs, channel attribution again. Let's say you would be, let's say you have N channels of distribution. You could easily do some kind of a heat map with this geographical distribution. Let's say Maharashtra is supposed to be a huge buyer of synthetic fabrics, and you are sort of gray there, which means that you're sort of only capturing a small fraction of market there. It means you've got to fix something there. Visualization could be geographical, it could be by various sales channel groups. So you have channel one, channel two, and you know the cost of the channels, you know how much business they're bringing in. You can sort of do a comparison and see, do a cost per unit sold kind of attribution there. The bigger picture that I wanted to mention is while all of that was in a specific context, the point is a lot of these analytics tools that we have, they are tools to solve various problems. Tools could be time series analysis, linear regression, machine learning, clustering, visualization, a bunch of tools. There are many more that I haven't even mentioned there. They could be used to sort of solve various vertical domain problems, forecasting, production planning, and the kind of solution that we're trying to build has a common core which is applicable to various contexts, but custom solutions which are adapted to sort of each industry in a way. Once you have the tool, if you come in as a client or as somebody that you need to use this tool, you need somebody that needs to use the tool, you might have just to just go through a simple level of parameterization, sort of mapping your processes, your data structures to what the tool expects in a way. Not trying to replace existing ERP solutions or local tools, it's something that sits on whatever tools that you use and just generates this interesting report which is a little easier for the entrepreneur to understand to act on than raw data that goes into the system. Pilot simple Excel, R where Excel breaks. People have been talking about what are the platforms that you use and these are some of the platforms you use. The point that I was talking about, these skills being sort of fairly well established, nothing new seriously there except for application to some of these domains. Some well documented literature about how Boeing aircraft under many serious losses because they couldn't manage their internal logistics while Walmart becoming a big player. These are fairly well documented examples I've taken them from elsewhere. Should I take questions? No, it's structured as a product. Something that will sit on, it's structured as a product. There's a certain level of customization that goes with it but at the end of the day it will be a software that sits on top of whatever system you have for workflow and give you this report. Does that answer your question? I think so, sorry, I'll just add something there. You know if you go to a lot of these SMEs while we talk about big data cloud computing the fact is most of these SMEs still run things off a single computer sitting in there sitting in their office. The internet, the network connectivity may not be that reliable. There's a little paranoia about virus issues about even putting the idea of putting the data on the web. So when you go to tap into that I think this you can't get away with the idea of having something running in-house. So here's a question. So my question is in contact with a startup in software industry and operating finally in V2C space. So now in V2C space it's expected likely to generate a lot of data over a period of time which will definitely be useful for the company. But then since startups are cost sensitive so when to decide like you start generating the data or capturing the data which may be used further because in the beginning the startups may not have that financial support required to do the same. So when do you exactly suggest to start focusing on this rather than focusing on the platform that they are developing? So I didn't get the question. So it's in context with a startup. So nothing to do with this, right? No. Okay, so carry on. So the startup is into the software industry and operating primarily in V2C space where a lot of data will be generated over a period of time but then initially startups may not exactly want to focus on capturing the data for the analytics purpose because that will require some financial support but since the startups are cost sensitive so when do you exactly suggest to start focusing on these data capturing and the analytics part rather than focusing more on the platform? I think it depends on the problem that you're trying to solve. At least here we're not trying to solve the data capture problem. They already have data capture tools in a way they have done but looking at here, somebody has already solved the problem of digitizing the data. All I'm sort of proposing is let me make you use your data better whereas if you are doing company which is into sort of data capture or if you don't have a system which has data and if you want to actually use the data more effectively then you go to capture it right at the start. So I think it depends on the problem you're trying to solve. If analytics from data is not your main problem you're trying to solve then you can take a little But over a few times in one or two years and we want to just monetize that part also but initially I do not want to spend more on that part that's the job I have because when exactly you start investing into that part also I can't give a very general answer to that question I think I would say when the fundamental problem I'm trying to solve is the fundamental of that problem would I start looking at another monetization option what I'm talking about is other ways of monetization so if I have a certain problem I'm trying to solve a certain key monetization plan in mind until I crack that or I have a decent comfort level with that I wouldn't distract myself doing other things there but some people could have different takes on that My question is how critical is it for you to sell this idea to SMEs and how usually they are ready to invest into something called analytics? The two aspects you do a startup or any organization either you are in market creation game or you are in market capture game market capture is somebody has a product and you are given a sort of a delta percent version of that and that's normally a slightly easier pitch I wouldn't say it's easy easy pitch but that's normally a slightly easier pitch this is sort of a market creation thing people don't currently use any tools so you sort of have to convince them about the value now when you have to convince people to buy something there are two aspects again there is capacity to pay and then there is willingness to pay at least in this context these companies have a company which might have a turnover of talking numbers here 1 crore would be easily happy to invest in 1% of the revenue 1% of the revenue in a solution that seems to have some improvement there at least in Surat textile industry there is that there but again you have to be more strict value there and the only way you can be more strict value is by piloting it out if you come up with a solution chances are the guy who looks at it and says hey I already know this you're not giving me anything new you sort of have to incrementally pilot it out and having this first pilot site sort of becomes important place where you can sort of fight it out discuss with the guys who sort of run the business and see whether they find it useful or not what else can go back into it becomes important and I think once you have sort of a key player captured the guy amongst these SMEs captured others are normally easier sale to make exactly our business is too dynamic and volatile to fit into these models there is this thing about piloting and sort of pilotating the other thing that I have learned is where do you draw a line I can be a full-fledged consultancy company and say hey I'll help you fix your business but that's a much harder sale to make because suddenly you're putting yourself in a position where you say you know more than the guy who runs the business on a 24-7 basis so at least the comfortable starting point not just starting point actually the comfortable skill point is also a product which gives you meaningful information from your data and lets you act on it so hope is that you will if not today if not tomorrow maybe within certain sort of time frame see a value form of this information to act on it but at least you can just use it there's no harm to be done I'm not saying that I'll sort of run the company for you but I'll just give you something useful you figure out whether you find it useful or not so if you were to sort of talk about adding manpower to these companies and especially an analyst of that you sort of lost the game already so at least my senses which is approach we're taking is build a product one time there might be there is customization there is adaptation but then once done it sits on top of your top of your workflow too and you're there to support whenever you have issues but no there's no content so I can't sort of answer this from experience right now yet because I haven't done that step but my hunch is so this talk about a lot of skill sets are similar so while sort of I'm adapting to a specific industry hunch is when I go to an industry too second industry the kind of customization I'll have to do is won't be won't be very different I might instead of say two stage production process I may have three stage production process instead of say a different measure of one measure of cost in one year industry I'll have a different measure of cost there instead of say three sales channel offered one time I may have different set of sales channel there so there's that kind of parameterization that will go in there will have to be some kind of domain expertise that you have to sort of go deep into that industry to get to get but the hunch is that it shouldn't be too bad but I will be able to answer that question better in about a year maybe most of these SMEs get their product made by a local vendor suppose SME one has a contract under one to get their product ready SME two has contract under two their data model might be entirely different in fact how is the solution for this so fair question that also links into the question that maybe was asking about scalability at least while the SME solution space the IT solution space is fragmented in major hubs you have to find a few major players like I can talk about through a textile industry again there are three major companies that are there one company has about say 2,000 clients which is probably 50% of all the SMEs that have been digitized so there is the space is fragmented but at least there is a significant fraction of them that can be captured by adapting your system to major data solutions that they use that's again a hunch that's sort of I've checked in this context but needs to be applied more because while SMEs also won't also invest in getting a very very custom IT solution made typically what happens is you have a place which has a N number of units, printing units and a local guy would have made an IT solution for all of them as opposed to making a solution just for one of them so that gives you some little bit of clustering there