 All right. We head into the next session, which is a special session. Now, this next session puts the spotlight on decoding artificial intelligence and machine learning in the context of marketing. Now, adaptive marketing is personalized. Adaptive marketing is segmented. Adaptive marketing is customer first. And adaptive marketing allows for flexible customer journeys. Adaptive marketing is also data-driven. But it does need that human insight. We at the Contlay, we are big proponents of sending the right message to the right customer at the right time. And that is why we've slotted this most insightful session next for you. Ladies and gentlemen, I'm privileged to welcome on stage our next speaker, Ashutosh Srivastava. Now, he is chairman and CEO of Asia, Middle East, Africa, and Russia, and CIS, Mindshare. He's the man who set up Mindshare India in the year 2001. And we are waiting to hear from you. It's all yours from here. Can you hear me? Yeah. Privileged to be back. I remember when I left, in fact, had done this exactly 13 years ago. Thank you, Anurag, for inviting me back here. What I thought I'd do is, after Tamara has spoken about how artificial intelligence drives creativity, is to look at it from another angle, which is to see how we can use artificial intelligence to find the right people at the right time, and the right place, and the right platform, and create engagement in a way which works for the brand. So that's the angle at which I'm going to take you through. I have a few examples, which I thought would be useful to share. There is a lot of work which is happening in this field. And it's obviously a lot of developments which are currently driving brands to experiment and get on with it. So let me begin. As you can see, the title of this session I have put as humans versus machine. Because a lot is said these days about how artificial intelligence is going to take over all the things that all of us do in this group. And therefore, all we need to do is put certain things inside a machine, and it all comes out, and we can all go and retire and live lives of peace and happiness. The way we look at it is certainly in our lifetime. If you ask people, the way technology is progressing, it's probably going to certainly not in a lifetime have machines replace the tasks that we do. You saw some of the examples of what artificial intelligence can do when it comes to driving creativity with Samara. But in essence, the way we look at it is this is a way of powering human intelligence to do even better, to amplify human intelligence, to be able to do things which otherwise are not easy to do, and making it all happen. So before I take you through the session, I thought it would be good if I can get everyone in the room given we have such a diverse audience on the same page as to what we mean by terms like artificial intelligence, machine learning, you heard about algorithms. So I'll spend a couple of minutes just getting everyone on the same page, and then we'll see how we apply some of these things to find the consumers that the brands want to target and how to engage them effectively and to know whether it is working or not in terms of delivering the results. So that's the point I was raising earlier, that the task with which we apply it is to find, make sure that the right message goes to the right people at the right time and the right place and the right platform, and it works for the brand. So these are the three buzz words which everyone talks about these days. Let's quickly just scan through what we mean by them, and therefore how do we apply it. So let's pick up the term. So machine learning, Samara briefly talked about how machine learning is applied, and I'll show you an example in the session. It's essentially reading complex patterns from data, and you can either supervise the machine learning to do a task which you want. So for instance, in the media world, it may be how much should I bid for an impression on which I'm going to place an ad, right? Or it can be unsupervised. There are lots of people who are clicking on my ad. Can I find something in common amongst all of them? You know what we call finding people and building profiles and segments of different types of consumers? Or it can even be reinforced. I mean, how many people in this room who have Spotify or Samara? How does it build a playlist? It plays you a song and tells you, do you like it or not? So that's how the machine learns your taste and starts curating the music to the kind of things that you like. So all this is learning from data, reading the patterns from data in different ways, and then applying that to simplify some tasks. Let's look at the world of algorithm. I mean, if you deconstruct what it is in simple terms, it's simply a set of rules, a mathematical equation. You tell it something and it gives you the output. But if you start training it with a whole lot of data, exactly like the example I gave you on Spotify or Samara, start training it with data, then it starts to predict because it is learning things about you. It's learning behavior. So it starts to build certain models. So that's why a trained algorithm is a model. And the more of these kind of things we use to take decisions which we as humans take, a whole lot of machine learning comes together and it allows us, therefore, to apply what we call artificial intelligence to take certain decisions which are assisted, decode the data, read the patterns, apply rules, and build some prediction. If I do X, the output should be Y. It's really the application of artificial intelligence that we are talking about in today's context. Of course, it will evolve as we go along, but largely the cases that you will see in the session which I'm doing are around that kind of application of AI. So to summarize, looking at reading patterns and being able to make certain predictions and therefore making your job easy to take the right decisions from a marketing perspective for the brand which will result in a certain consumer response is what this is about. Okay? Now the thing is, if you look at algorithms, there's a whole marketplace out there where people write rules. You can pick it off them, but you cannot make it work for you unless you use your own data from your own business. And you've got the talent which understands marketing to tell it what to look for. What makes a difference? Because algorithms can be written by any mathematician, any student, but it's your data in your business and the people who bring the experience to that algorithm which make it useful for you. And the examples which you will see are really where the differentiation is created by the talent and the data which has got something useful to come out of it. So one of the companies that we are working with as a partner in this 24 by 7 AI, it's basically specialized in application of AI in terms of doing better search and also in terms of personalizing, marketing brand messages at scale. These are the two areas which we work with them on. They've simply worked with us to create a framework when you start thinking about how you're going to apply these things. So understanding the customer journey, anticipating what the intent is, creating engaging experiences and then the outcomes which are there. How do you optimize for that? So there's application of AI in each of these areas and in order to make sure that this is done on scale from the marketer, brand and agency point of view, creating capabilities in each of these three areas. So being able to target, finding the right people at the right place at the right moment, make sure that the right message is coming at the right time in the right place on the platforms that you use and the ability to optimize in real time because how many of us sitting in this room as brand marketers, we sit with a team and we figure out, okay, in this campaign over the next one month these are the set of things we will do and then at the end of that period we will see whether it worked or not. But the real benefit of AI application in this is that you can decide every second of the duration of the campaign, the decisions that you need to be taking to get better results. So it's about the ability to take those decisions in real time and change the outcomes where the real benefit of AI starts to kick in. So going back to what Tamara had said earlier, looking at the human angle of this entire thing which is people, their experience and what they bring to the table to innovate and to be creative. And it's using the power of the machines to read all the data and the patterns to be able to assist you in doing things. Bringing the two together, stop working, bringing the two together, we start to create magic. So I'm going to show you an example here of a brand which doesn't want to be named. It's an example from Thailand which we did recently. So we looked at all the consumer moments at which this brand can be consumed. And we figured out through all that data that there are almost 7,000 different moments where right place, right time, where the brand can be consumed, where the brand can be triggered to be consumed. So imagine running a campaign where you've got 7,000 different messages matched to triggering consumption to consumers at any point in time, right? That level of personalization. Let's look at how this was done. Can you play the video please? BrandX wanted to connect its brand to specific meal moments in Thailand across multiple audience groups, locations and times in a programmatic campaign. To do this, we had to create a marketing strategy with 7,056 targeting and creative combinations covering 14 regions, 63 snacking moments per week, two audience segments and across both desktop and mobile in a world where nobody could manage more than 40. Human limitations when interfacing with DSP and DCA systems mean you cannot make this many inputs by hand. To address this challenge, MindShare together with YouGov, AppNexus and Givox developed a world-first AI solution called ANA. ANA uses a bespoke AI to construct an integrated media and creative algorithm that powers the operations of DSP and DCO platforms. This allows ANA to manage complexity beyond regular human capabilities. Once in flight, MindShare's fast team made previously unreachable volumes of data-driven AI optimizations. Using a high-definition media plan with over 7,000 line items and a high-definition dynamic creative insight dashboard resulting in over 3,000 optimizations, a 150-times increase in decisions made, all updating in real time. Results. 26% reduction in VCPM for desktop, 17% CPC reduction, and a 56% CTR increase. By personalizing creative, we achieved 2.5 times the CTR of an average campaign. MindShare, moving into a new era of programmable media. Can you imagine a hot summer afternoon and you're standing just near to a vending machine and that's where the consumption is triggered based on where you are, what the weather is, what your intent is because you were searching for something on a phone. Being able to do that at an individual level as a human being, possibly easy, but to think of all such possible combinations and to be able to execute this on scale is where the real application of AI starts to kick in. And suddenly it opens up so many opportunities for the brand to start engaging with people and driving these kind of results. Another example I wanted to show you, this is from one of our group companies, Axis, which has a program called Co-Pilot. Again, it's an application of artificial intelligence. So how do we find the right people? What drives certain behavior when they're exposed to brand messages? And how do we write great results? So let me just quickly take you through this example. Coming from there, it's basically applying again, artificial intelligence to drive performance. So let's look at a typical campaign. A lot of us in this room who buy, run campaigns looking at outcomes. These are a whole lot of things which we consider when we are doing a plan, which you can see on the screen. The consumer in this case is looking to buy an air ticket, right? So anyone who's planning a campaign for this is looking at all these parameters. What we are trying to do is to acquire a customer and then make them buy a ticket at a certain cost. And the objective for the brand is to try and minimize the cost at which you acquire customers and for you to acquire the targeted number of customers. So you start looking at this data. This is the outcome you're driving. So you first look at, when you're putting out that ad to make people look at you and click to start planning a journey. What are the different things which an ad has? So this is an ad you can see there, the sunny side of life. Website there, it has a whole bunch of features. So which browser was it put onto? What am I doing? Am I seeing it on a mobile phone? Am I out of home or I'm inside my study or at the office? What technology is being used? What time of day is it? A whole lot of features that go in, a whole lot of inputs which we look at. However, all of these things are not necessarily equally important, right? So when do I plan to buy tickets? Is it in the morning, in the afternoon or in the evening? No two people do this alike. The idea is to find you at a time when you're interested in booking a ticket. What machine or what device are you using? Where are you at this point in time? What is working and what is not working? Again, based on machine learning, what kind of message is going to attract me to start looking at you? So we distilled all that down, again, application of this machine learning into a whole number of things which we look at. These are the features which will drive the outcome. But some of them are more effective at driving people to buy tickets, some are less. So how do we find out which are more important? Because those are the ones that we are going to focus on. So broadly, there were four, which ad exchange are you buying your ad on? What's the size of the creative message that is appearing? What device are you looking at? And at what time of day are you watching this ad? This is going to trigger the sale of the ticket. So the data was reduced, clustered into these four, and even amongst these, some were more effective than the others. So again, the intelligence started to cluster them. So as you can see, there are some which are more important than the others at driving the sale of the ticket. These are grouped together. The next thing was, so what is more important? What is more likely to result in a sale? We started assigning some weights. Again, the machine started assigning some weights to the action. So as you can see, there are some in red. These are not as effective. There are others in blue which are driving the result far more effectively. Armed with this information, that's how the whole ad buying happens at this stage. So again, look at two scenarios. So as a media planner, I would know from the data, which I can digest, the ad on the left is someone who's sitting in front of a computer looking at Expedia, planning a journey obviously. So you are more likely to find such a customer. On the right, on the other side of the screen is someone looking at a mobile phone, reading the news on the BBC. And intuitively you would know that probably lesser likelihood for someone to click and buy a ticket. This is something I can probably sit in the office and do. But there are zillions of such combinations of occasions, place, devices, time of day, intent which all come together to decide whether the action is going to be done or not. So how do we make sure that we find all such occasions? So when the campaigns are bought, you obviously start putting money into each of these scenarios. You put more money in the scenario which you know intuitively will work for you. And you see, you put in some money in the other scenario to see what kind of result that's generating the machine learns. And it starts to tell you what you should be buying based on this kind of input. So there are lots of such scenarios, like I said. So again, as a media planner, I could probably think of two or three or five such scenarios, but the machine has mapped out hundreds and thousands of such scenarios and it started to buy ads in all of them and it started to track them. And what it's finding is, can you click this? Can you click it, we'll play the video, not playing? Okay, so what the machine has done is it's digested all the data, and it figured out the scenarios in which you are going to get the best results and immediately your cost per acquisition starts to come down. So which is the net result? Is the tickets are sold in the quantities that you wanted and they are at a price which is below your targeted acquisition price. So it's applying step by step machine learning and each time you're on this campaign, it learns a little bit more about what works and what doesn't work. And it looks at all those millions of scenarios and that's how it moves you to the next level. So an interesting example of how we can apply all of these. Another example which I wanted to show you, this is very much from here. Now how many people in this room who have kids who are below 13 years of age? Quite a lot. How many of these are on Facebook? Kids 13 and below who are on social media, Facebook? Or who are pestering there, you say, everyone, right? Does Facebook allow people at 13 years and below to register as on them? No, they have a policy, right? So if you as a brand are targeting kids, it becomes very, very difficult to find out because those of you who had kids who registered, typically they fake their age or they build a fake profile. Very difficult to find out and identify such people. So how do you find out such kids to target? So in this case, the brand, which is Boost from GSK, what it did is it went and firstly we figured that the best environment in which you could find some kids is games on mobile. So again, a proprietary machine learning algorithm was applied to figure out whether the people who are playing those games are really kids or not and who are the people who are the kids? Who are the people who are adults? And then to be able to target the kids with the message. So have a quick look at this video. Boost fueled its growth by high brand preference and best of power among kids with communication focused on taste and sports celebrities. However, in 2016, they faced an issue of declining brand preference. To build a stronger connect with the TG, we looked for insight from a kid's perspective. We got it in karate kid. Children no longer consider celebrities as demigods, but more as friends or mentors. What if a Virat Kohli became their friend instead of a larger than life celeb? With this in mind, Boost created three long format digital videos showing Virat coaching Rahul. Digital was the chosen media to disseminate these videos. Send me your videos at boostenergy.com and practice with me. The challenge was how to reach the kids as they surfed the internet in a disguised form. Our research showed that mobile was the first screen for kids and gaming the perfect way to reach out to them. But how do we separate kids from adults on gaming apps? Mobile gaming ad network pocket came on board and identified kids using their machine learning classification algorithm. The technology first identified kids using contextual relevance. The data was then vetted with analytics and repeat content consumption monitoring from audience platforms. We took user initiated video inventory on their mobile gaming apps to reach our TG. The core campaign was layered with presence on cricket influencer destinations. Virat Kohli released the teaser on Facebook to kickstart the campaign, followed by the first video the next day, which was picked up by cricket influencers. The results were phenomenal. Driving 18 million video views, the core campaign reached 6 million unique users. 10 million of the views were complete views. An achievement considering that most of the inventory was user initiated or skippable. There was a 7% lift in consideration among respondents. The ad received a high recall value at 86% and the brand registration stood strong at 62%. Impact was seen on the brand's performance. Pulse the activity boost registered a 2.2% increase in market share. With an 84% consideration top box score, the brand equity was at the highest levels. Right, so again, intuitive and things which you'd obviously do, but again, using the data and the technology to be able to then fine tune and target kids with this message as a whole new set of results coming in because of the ability by using machine learning to be able to target all your effort at kids like that. Here's another example. This is for Hilton. So again, when you as a consumer start doing this, how do you experience the technology? This is an example from there again from our partners at 24AI. So let's look at, let's say I want to book a room at the Hilton and this is now the machine learning algorithm working in the background for you. See how it unfolds for you. So I go to the website, I look at the city, I type it in, so let's say I'm looking for a room in London and there are certain dates which I'm going to go ahead and book for, so it goes, I go and click on the dates on which I want to share this event. What will it do? It starts throwing me options where I can stay, right? All of us have done these bookings before. So we start scrolling and looking at the prices for all the hotel rooms and what it looks. So it looks like an attractive rate and a good option in a room here. Come back to the site. I'm searching for more information at this stage. Already the model is saying that now there's a higher likelihood of this person to convert. So what is it doing next? Why don't you become a member and therefore I'll drop the price for you and I'll give you a really attractive lead. So I go there, I start entering as a consumer. I want to compare it with what the rate would be if I were not a member, right? And you can see that the model now thinks that you are more likely to go ahead. So let me quickly pop up a virtual assistant for you and convert because you've now become a hot lead. So the machine learning is working in the background. It's tracking your intent, your behavior and based on all the knowledge it's accumulated from the data before, it's trying to convert you eventually, pops up a virtual assistant in front of you to try and convert this lead into real action to recruit you as a member of Guilted. That's how you're experiencing it at the front end. Okay, one last example before I wind up. So all these examples you've seen so far is about brands applying it to campaigns in different forms by basically simplifying the data using algorithms which have been trained between learning and therefore large scale application of AI. There is another application of that and this is both in deciding the media as well as driving the creativity, which is churning vast amounts of data to create insights. Some of these insights are even used as a business application to see whether the brand is doing well or not. So I was going to show you this example of a tool which we developed for again in India for Hindustan Unilever. Let's have a quick look at that video. Introducing Purple, mind shares revolutionary decision making product powered by predictive analytics and artificial intelligence. Purple has four key modules each addressing recurring pain points that brands face. One, the media mart or the data layer. Two, autoleses or automatic insights. Three, the predictive simulator and four, voice enablement. Let's start with the media mart. Data across sales, brand track, media and macroeconomic factors lie in unbreakable silos and in varying formats and nomenclature. It takes analysts as much as three to four weeks to extract relevant data from different systems and bring them into common denomination for meaningful analysis. The media mart addresses this pain point and is the bedrock data layer of Purple. Extract, transform, load, routines enable consolidation of slow-moving data dimensions like sales to real-time data scrapping from social media. The media mart is the standardized single-view data destination for all possible data that could be required for any brand analysis. Coming to autoleses, brands often react after the hazard has stricken. It's often a post-mortem and seldom preemption of the hazard. Autoleses enables critical data visualization with AI-driven automatic insights and trigger rules. It comprises of a powerfully visualized early warning signaling system for metrics that matter. Autoleses enables proactively spot micro markets where key metrics have significantly dropped and highlights flow movements in these metrics. This warming mechanism flags volatility in investment variables at TV channel and print publication level as well. Autoleses also helps understand the strength of cause and effect relationship between 100 plus variables at the click of a button. And last but not the least, all of this bundled into an AI-fuelled autoleses report which enables prompt root cause analysis within minutes. The predictive simulator is at the heart of Purple. Powered by self-learning predictive algorithms, Purple predicts the impact of own and competition media on outcome metrics. The predictive simulator is a handy tool for planners who can assess the impact of their media plan and make prompt corrections on the go. Yes, Purple also talks. There'll be about a hand on these things. Sure, getting numbers for you. Here is a graph containing info. While HUL has grown by only 7%, your competition to be has grown by almost 27%. Hope this was useful. The technology queries the underlying data tables, joins the dots and leverages natural language processing algorithms to articulate the solution in English. Now we have Purple at work, crunching terabytes of data across multiple disjoint platforms, flagging business hazards and prescribing media solutions at the click of a button. We had a genuine business problem to solve on Vaseline. We had seen that spawned for Vaseline had dropped in the previous winter season and we actually wanted to identify what is that variable that is leading to this decline. And that's where Mindshear's Purple came in handy. It used multiple data points to check and analyze what exactly is the root cause of this issue on why spawned is falling on Vaseline. So it came up with a very clear set of recommendations on where we should be investing our media monies so that spawned comes back. We implemented our plans in winter of 2017 basis recommendations of Purple and we actually saw spawned come back sharply. Purple, a truly transformational problem. Right, so again application of vast amounts of data being churned here to analyze your business results to do real-time planning. Things which we wouldn't have even thought of a few years ago is all made possible by the application of AI. Digesting all that data and converting it into insights whether it is to understand who the customers are going to be, what is it that is going to make them change their behavior, fuel the creativity and also decide how you're going to expose those messages and how you're going to create that engagement using vast amounts of information. And this is I think just the beginning of this whole change or this turn in terms of deployment. So the way we go about doing it, at least the example which I can share with you the way our own community is doing it here is step by step doing a whole lot of small pilots capturing that learning and making it bigger and bigger. But a few things which I'll just leave behind as thoughts is one is that the data and the tech that we use, the platform needs to be consistent. The second thing, there are lots of applications which you can develop as you go along using data to decide when is the right time, using data to decide what the intent of a bunch of consumers is. So different applications, so these are all trained algorithms which are developed and left on that platform for the planners to use, whether they're generating insights to create the communication or whether they're generating insights to see where to place that communication to engage people. And again, when you apply it first, you see big changes and then it's all about learning and incremental changes and therefore consolidating on that. So that's how it works. So we use a platform called Domino. Again, for the tech people inside this room, you'd be familiar with it, it's basically a place where you can standardize creation of machine learning algorithms, putting them literally onto an application store. So here's an example of some of the trained algorithms that we've started using and planning to decide which channel or which platform is generating more results than the others. There's a number of such applications which are already in play. And again, a change in mindset at the end of the day because when you work with such vast amounts of data, suddenly planning for a campaign to planning always on becomes possible because you're monitoring changes in behavior and impact as you're running your marketing campaign. Creating collective intelligence. You saw a number of different algorithms and applications that are used. So how do I use all that? And enabling constant iteration. There is no one right answer. As you run the campaign, you start seeing results with technology and the AI enables you to constantly iterate and make it bigger. In fact, recently, anything is possible. Just last month, we worked with Google. We just went and bought, let's say, a whole year of cricket data, all the footage. And Google and us worked with image recognition software to map out from the ICC. Last 12 months of all the cricket match data with a view to what? To predict when the next wicket will fall because a lot of our brands, Pepsi and many others use cricket very heavily and having contextual messages at the right time means that can I make sense of the data to predict when is the next big action going to happen and therefore can I come out quickly and therefore put in a message to generate the result? And it's quite fascinating because I'm telling you, we are now, right now, tracking a series in Australia. And the first match we applied to, we got nine of the 10 wickets which fell right a few minutes before the wickets fell. It's as big news not just for the brand marketers but also for the betting industry, I must say. But that is the real power of AI. It's able to make sense of all that data and the technology exists to digest all that information and start harnessing it to that kind of effect for the brand. So that's really the future of where we are going to do. It's bringing the power of AI and human intelligence together to make things like this happen. So I hope you found this session interesting. Thank you very much.