David Boyle ABC Always Be Clustering Sept 2020

Great. Thank you so much to check that this is working and that the comments feature is working. Can somebody please tell me that you can hear me talking? Can somebody please say hello? I did not want to be talking to a blank PowerPoint page for the whole time, so I really want to make this as interactive as possible. My eyes are going to wander up here to the comments section regularly and the question section. And so the more comments you give, the more questions you ask during this presentation, the more interactive it will be. And there is a presentation. It would be silly not to have one. But if you want to distract me from it and ask lots of interesting questions. Hi, Joanna. First question or first comment? Thank you. If you want to distract me from it, ask questions, go down rabbit holes. That'll be a lot more fun for all of us. It will not be distracting me for the presentation. It will be just adding lots of fun to the session. So does that sound good? Be back in the comments. How many of the 519 are actually watching and how many are just logged in? Yes. Okay, great. So we're going to talk today about one of my favorite topics, which is statistical clustering. And if you still stick around to hear more after I said that, then you really deserve a prize. But seriously, I think that this is one of the most exciting things that is kind of vaguely understood, but underappreciated and certainly underused. It's an incredibly powerful tool. Hi, Ram. Hi, Sophie. Incredibly powerful tool I think everyone should be using. I mean, I really think we should always be clustering. So by the end of this, I hope to convince you that that is true and give you some clues about how to go about it. We're going to have some fun along the way, particularly if we get some more good questions and comments in. I'll try and make it fun anyway. So by way of background, everything we do is about audiences in business. Whether you're thinking about audiences, whether you're reporting on audiences, whether you're planning business activities that will affect audiences, analyzing audiences, strategizing or targeting, it doesn't matter. Anything you do in business really requires some level of audience understanding, some level of audience data. However, I think that there's two ways that many businesses go about that. Hi, Jessica. Hi, Sophie. Hi, Ram. Again, more questions, more comments, please. I think the first way businesses often go about it is really individual targeting. And then the second way about businesses go about it, I think of this being on a spectrum, it's more about mass targeting. Let me show you what they are, and let Me show you where clustering falls in that spectrum. Individual targeting is clearly looking at each individual in isolation. It's predictive models at an individual level or it's storytelling at an individual level. And that's awesome because it's really precise and accurate. But it's just not very efficient to have to build a model for every person. It's very overwhelming that you can't carry that insight around in your head to. So it's good for some activities but not for others. The other end of the spectrum, many businesses think about consumers in mass as overall big lumps of people. That's very efficient because you just got one big lump of people, one average behavior. It's very simple, but it's not all accurate and it has huge wastage if you're using it for targeting or planning purposes. So most businesses sit on one of the ends of the spectrum, I would argue. And audience segments, which we'll talk about from a clustering point of view in a minute. Audience segments for me is the middle ground where you take the beauty of the individuality, but you group people into manageable number of clusters or a manageable number of segments. And that for me is much more efficient. It's precise, simple, it's representative because each type of person has an archetype, has a segment that they're described by, and you can use it for really strategic thinking. I'm going to show you how that works in a minute. But basically the spectrum from individual to mass and audience segments right in the middle. The first example I'm going to give you, which should be familiar to some people because I love hearing this often and it still stands out to me as the best way to describe the situation here is comparing Prince Charles and Ozzy Osbourne. And if you just look at subtotals like demographic groups like age, how many people are a certain age or country of growing up, or number of children or wealth, then you end up putting them in exactly the same category. So subtotals, just basic totals, drawing a line under the data and adding it up. Everyone who behaves in that way, absolutely rubbish. Clearly not very useful in describing the difference between these two people. You want to act on them very differently as a business, but they come under the same categories. So subtotals are rubbish. Forget subtotals, we're going to move on from that. And clustering, even if you add some more information about these two people, married a second time, successful in business, enriched data maybe still absolutely rubbish. Demographics just don't work. Subtotals and demographics, terrible way to describe people. However, the thesis I'm going to try and persuade you of today, clustering, not subtotaling people based on demographics, but clustering people feels like magic. I'll show you that. I'm going to persuade you of that. But what do I mean by clustering and clustering on? What if we're not using demographics, what are we doing? Well, there's lots of words you can use to describe how we should be cluster, what we should be clustering on. I like the word needs. That's the word I use most, someone's needs as it relates to the category that you're studying. In some situations, the proxy for these might be affinities or behaviors or interests or aspirations or attitudes. Some people like the word psychographics. It's way too fancy for me, but I put it on there because it'll resonate with some of you. But something to do with needs or wants or something that's deep inside people. If you cluster on that, then instead of demographic subtotals, you've got these clustered groups and they are really exciting. That's what I'm going to show you today. And the first example I'm going to give you, the example where it works really, really, really well, was from a few years ago. Now. I used to work in the music industry. I had the pleasure of working with a number of very cool artists, including this chap, David Guetta. And when I met David Guetta, he was incredibly successful already. I can't take credit for that. Of course, it started off niche, everybody does, and he'd become very popular, but he plateaued. He'd reached a point where he wasn't sure where future growth was going to come from. And that's where the clustering really came into effect. And so what we did is instead of saying, on average, you're very popular, or across all people, you have a popularity of X. But we broke people down into statistical clusters based on their needs as it related to music. And then something really interesting emerges, which is much more precise language around the type of people that he has engaged, where the growth had come from, and also the type of people they had not yet engaged but were excited about his music when they heard it. That is where the future growth could come from. I'll show you that in a second. Pop Idol and Casual were two of the statistical clusters where future growth was going to come from. And by naming them, by understanding their individual needs and behaviors and wants, David was able to put together an amazing program of music promotion, marketing that really engaged this new audience, new pop idols, new casual audience, this growth audience. You can see it here on the vocal album that he released. Up until then, it'd been super hardcore, relatively hardcore electronic music, some pop stuff in there. And this is where he goes big in the pop world with duets, a number of people that brought new audiences to his music. But important for him was he didn't want to lose his core audience. So in this example, the second disc or the alternate version of the album was the electronic album with none of this vocal pop nonsense. Completely designed for his core audience and maintaining his core audience while also on the top, growing his new audience. This is one simple example of how he did that. This strategy permeated, as I said, marketing and promotions as well as music and videos. So this is part of a much bigger strategy, but a really simple, neat example. When you have this very clear thinking, very clear language about consumer clusters, you can suddenly go do everything differently and go reach new audiences. And sure enough, as I said before, we'd already reached the point of growth and then stagnation. We were now able to say, that's the Nighthawk audience. That's one of the consumer segments. You've reached them and you've maxed them out. And that's it. By and large, that's your audience interested in nighttime music, music to soundtrack and night out, mostly electronic music. That was their need. But once we had reached the pop audience and the casual audience, to other clusters, to other consumer segments, David was able to achieve this remarkable extra growth which dwarfed the original market. But he built it on top of his original Nighthawk market. And so he grew the market using consumer segments. And I don't want to pretend that this was my strategy or that the strategy came from consumer segments, because of course, going to a pop audience, going to a much more mainstream audience is a really great way to grow. But this language and this clarity and these analytics helped to prove that that was the right strategy, helped chart a course through those audiences with the media they need to consume and the messages they want and the artists that that's worth collaborating with to help guide him through that journey and to measure success against it. So sure enough, you measure again in research six or 12 or 18 months later and you see him building these new audiences. So even when you think you know the strategy, clustering and data really help you clarify and go after it in a much bolder way. And so there are 11 consumer segments at EMI Music, and for David Guetta the story was really about those three, but for different artists there are different stories, David from Nighthawks to Pop Idols and Casuals. For Katy Perry, here's her Unplugged album, which took her from a pop idol audience to a cutting edge audience, an audience who really respected and craved authentic, from the heart quality music. And for Pink Floyd, another totally random additional genre, going from a cutting edge audience, which they'd been satisfying for a long time, to really deeply satisfying the collector audience, who they didn't think was interested in their music. But with research we showed them that actually they could be much richer and more elaborate in the box sets they produced. So really deeply engaging the collector's audience. So for each artist, really different journey through those segments. But for each artist, using those segments really helped clarify or challenge their thinking and really made a difference to their career. It's not just some fancy CDs that got produced in new directions. EMI Music, where we did this work, became the most profitable record label of the four by a substantial margin as well. I think that's because everything we did was smarter and bolder and quicker and clearer thanks to having these consumer segments. So when you get this right, it's absolutely transformational both to the culture and to the economics of the business. That's when you get it right. But it'd be boring just to show you good examples where it always works, wouldn't it? So we're going to look at an example where this did not work at all with consequences that we all feel every single day, all right, if you take that out the picture, take audience segments out the picture and if you just do this individual perfect person level targeting and if you measure things on an aggregate level of whole populations, what could possibly go wrong? Let's talk about the 2016 election, shall we? Now, I didn't work on this election, but I worked in American politics for a few years before this, so I know a lot about the tools, techniques and methods that were being used in the election. A lot of my friends worked on the election and I read a lot about the election. So here's what I think happened. The Clinton campaign was the most data driven political campaign that's ever happened, for sure. At that point in time, for sure, Obama did push the needle hugely forward. Clinton pushed it forward even more. Data driven to the maximum. But being data driven isn't enough. Hillary says in this quote from her book, if our data or anyone else's had shown we were in danger, I would have torn up my schedule, which was designed based on the best information, the best data we had, I would have camped out in Wisconsin, one of the three states that caused her to lose. Absolutely data driven. It's not enough. What they were really, really good at, though, were these two ends of the spectrum. It was individual level targeting. Every voter got the right door knock or email or online advertising. They got the right messages, they got the right frequency of contact. It was personally targeted and optimized brilliantly. Really, really good. On the other end, reporting was at state level or at demographic level maybe, or at national level. Really good based on great data. But what did we say about totals and individuals? They're rubbish for strategy. They don't work for strategy. And so you miss important trends. Like if you're David Getty, you think I've maxed it out, now I'm done, what else can I do? And if you're Clinton, you think I'm either winning or I'm losing in a certain state, you don't say to yourself, which segments am I winning and which segments am I losing? So I would bet based on everything I know and what I've read, I would bet that if Clinton had used audience segments as well as, as well as individual targeting and as well as mass measurement and mass targeting, if they had used audience segments, I bet they would have seen the pattern in the three states that they lost and they would have known they were at risk and they would have, as she said, camped out in Wisconsin and seized the election. So again, what did I say? Something with big implications for all of us today. They thought they were being as data driven as possible and were the biggest and most data driven political campaign that has ever happened in the history of the world. Data driven isn't enough. You've got to do it in the right way. And I argue that's clustering and segmentation. To help explain that, let me talk about efficiency, effectiveness and growth. On the efficiency side, knocking on the right doors in the election, maybe, or maybe like producing the right number of units of a product or a flyer in a product in a business or a flyer in a campaign measuring marketing effectiveness or efficiency, Things like that. They're really interesting. They make 1 or 2 or 3% difference to your costs or your revenue or your impact. Next, you have effectiveness, though, Edward, I really hope Biden Harris understand that one. I tell you, I'm doing my best. I'm doing my best to sing that song in the middle. You have effectiveness. It's better than efficiency, it's bigger impact than efficiency. Maybe 5 or 10% change in revenues or in costs or in impact of a campaign, how should I price a product or how much marketing should I budget for any one state or organization or group? That kind of thing. You can use data for that as well. But here's where I think we need to spend more time using data. And here's where I don't think the campaign did. And here's where we did with Snoop Dogg and with the Pink Floyd and with Katy Perry and with David Guetta, which is to help make strategic decisions in business. That's like, what product should I develop, which countries should I target via, which audiences am I for? And the same kind of thing in the election. And the problem is it's easy to do the things on the left hand side using data. It's relatively easy with some challenges in the middle, it's hard to do the things on the right. So people do the things on their left. They think they're being data driven, but they're missing a trick. Just like Clinton. On the left, I think you largely need algorithms in the middle, I think that's where analytics plays. But on the right, that's where insights play. That's where things that you, the candidate or the product manager or the artist can carry around in their head and act on day in, day out. On the right hand side then is where clustering lives. And that is where if you don't do it, you're missing out. Just big, big growth opportunities in insights that you can carry around in your head from clustering or otherwise. I think of them as being about helping you to make leaps, inspiring you to make creative leaps, giving you the confidence or the measurements or the inspiration to make creative leaps. It's not algorithms and analytics, it's insights. And again, I think the campaign optimized for those two, but not for giving Hillary the insight into the type of voter that she was winning or losing and what was trending that she could have carried around in her head and optimized for. We gave it to David Guetta, Snoop Dogg, Katy Perry and Pink Floyd and they acted on it. She didn't have that in her head. I think that's the cause. Clustering is beautiful, for giving people the kind of insights they can carry around in their head and use to make everyday leaps. So I think again, Clinton algorithms for optimization and efficiency. Brilliant, brilliant work. Amazing investment analytics for tactics and effectiveness. Really, really good, great work. But not what David Guetta had. Not insights for strategic decisions, which is where in business the real growth comes from or in campaigns. The real winning comes from. So let me just talk a little bit then about how you do some of this stuff. And then I really hope there's going to be some good questions coming in and some good comments because I'm going to pause and we're going to have a conversation about this. So how do you do it or what do you look for? What does this even mean in practical terms? First, I think there are three types of data you can use to do clustering these days, and you probably should do all three. But depending on your need and your business problem, one of these might be the one to focus on. It's either what I call owned data. So that's your customer list, your transaction history data on your products or your services, something like that. There's asked data, which is survey data or qualitative data, but it's something you're asking people questions and answer of some kind. And then there's gathered data, which I call gathered data, which think of a social media data. You're gathering it, it's not yours. You're not asking it, you're just gathering it. It's there, go get it. There's a bunch of that that you should do as well. So owned, asked and gathered independently should be clustered, I think, and in a way designed for every different type of business decision you're trying to make. So whether you're trying to innovate on products or brand or marketing or distribution or partnerships or market intelligence, you should come up with clusters that work for all of those outcomes and then have clusters have segments that everybody in the business can use, everybody can refer to that, become language and second nature again. That's what we got to in music and not what was happened in politics. It's what's possible in every one of your businesses. That's what I would aspire to. So one or more of the data sets cluster them with a goal of helping every part of the business. And some of those data sets are more useful for some things versus others. You're clustering your transaction data, your sales data. If you're a retailer, let's say it's generally more useful on the kind of tactical execution side because you've got customer names, you could email them or you can send them products after you cluster them. Also useful strategically. But that's where you really get tactical benefits. If you're doing survey data, that's where you get much richer data, where it really helps you to challenge yourself and think about big opportunities for growth. Kind of hard to tactically Execute against against that in a specific way. You have to carry those learnings and apply it back to other data sets to act on it. But it sets the direction, it sets the strategy and Twitter data. I find somewhere between the two, you can cluster that relatively easily and you can kind of act on it tactically by directly or indirectly through other platforms. But also it's rich enough, because what people follow on Twitter is a rich enough signal of what cluster they should be in. Rich enough that it's useful for strategic decisions as well. So different data sets, different places on that spectrum in terms of where they focus, but they all have a role in strategy and they all have some kind of role in tactics as well. Just don't miss the strategy. That's the lesson. I want to give you just one simple example of Twitter data being clustered. So this takes everybody who follows BBC Earth. I used to work at the BBC and I pulled this from an old deck. I remember it to this day and I love it. This is a couple of clicks in two or three online platforms you can use to do this a couple of clicks away, an insight like this. And it clusters everybody who follows BBC Earth according to what else they follow. So a bunch of people cluster around photography because they're photographers and their main interest in BBC Earth is, you guessed it, likely to be photography. A bunch of people follow science and biology things on Twitter and so they cluster around the science and biology cluster. Why are they watching BBC Earth, I wonder, huh? Maybe for the science and biology angles. You get the pattern here. A bunch of people, news and information, earth loving, as different from sustainable and environmental. These clusters fall out relatively automatically with again just a couple of clicks. But my God, you could base the whole strategy on this. Your content strategy, your distribution strategy, your marketing strategy, your communication strategy. You could base everything off this. We didn't, by the way. We also did lots of research. We also analyzed lots of viewing data, lots of owned data, transaction data. What I'm saying is you could, and that is absolutely remarkable that a couple of button presses and a semi automatic process can give you insights rich enough to carry around in your head and to set your whole strategy on. So I want to just end on celebrating how bloody remarkable that is. I try to remember the last time I was in an organization that had spent £100,000 on doing a clustering in politics in 2005, there were quite a few of these, then a bit later. So 2010 is the last time I remember somebody spending £100,000 on a big clustering Exercise. But you used to have to if you wanted to do it properly, which is lots of surveys, the only way to do it, and big agencies who had done this before. Fast forward 10 years. Not that long a decade. Well, not that long till you're old like me. Fast forward 10 years. The cost is not 100,000, it's zero marginal cost of doing one of these. Once you've got a subscription, the cost is zero. The BBC Earth thing costs zero to produce. It uses social media data, which is not quite as rich, but as rich, close to being as rich as survey data. And instead of using agencies, an intern can do it by pressing a couple of buttons. That is bloody remarkable. In 10 years this thing has gone from mystery, expensive gamble that you need ninjas for to something that anybody who's sufficiently curious can do in a couple of button presses at a zero marginal cost. That is bonkers. So that's it for presentation. I see a couple of questions. I hope there's going to be more, but I hope you leave this. As excited as I am, or vaguely as excited as I am about this, the power of it is absolutely immense. I showed you that in emi that was using survey data, though nowadays you don't need that. You can do it on any data set you have where you've got personal level data for close to zero marginal cost. And if you're vaguely curious, your intern could do it for you. I don't think there's any reason to not always be clustering. Incredibly powerful. Easy. Do it, do it.

Fantastic, David, Absolutely fantastic. It's really, really fascinating stuff. Yeah. Yes. I think we've got a question come through from Edward Bass. Where would data external to an organization such as an open data, or whether data sitting is a very cool model? I think he's referring to the BBC slide you talked about.

Well, I think it might be this, maybe this slide he's referring to. But Ed, feel free to clarify. I mean, I think that the to cluster, what you need is. It's like record level data. So we're talking about audiences here. So you need a person that's a unit and you need some stuff about that person. So for Twitter data, you've got a person and you've got the things they follow on Twitter. That's one way of clustering them. For survey data, you've got a person and the questions they've answered. For transaction data, you've got a person of the things they bought. So that's some examples at the person level. That one. Okay, is this one yeah. So if you've got external data, the problem is for external data it's hard to get person level stuff, which is why we mentioned Twitter and not Facebook, because Twitter gives you person level stuff, Facebook doesn't. And so if you had open data from any source, as long as it's person level stuff, then you can do person level clustering. That's great. You don't need any pii. So let's say you got a list of anonymous user IDs and the movies they'd reviewed from a movie review website. Totally cluster it. See which movies come into different clusters. For weather data, the thing I'm thinking about to your question is like what the unit is. So you could cluster locations. If each location you had all the weather data for the year, you could cluster the locations that would be your unit and say, well, which locations have similar weather patterns during the year? And you might spot some interesting groupings of locations. It's hard to imagine how the weather data would link back to being to a person and be very useful. I would probably cut in terms of this framework, I would probably call it gathered data. Something you've gone and got, someone else has got it, you bought it, or you've gathered it and you can use it. So yeah, you. No reason you can't cluster things other than people. We talked about places that have similar weather. You cluster stores. If you're a retailer with hundreds or thousands of stores, you can cluster products based on which people bought them instead of clustering people based on what products they bought. So as long as you've got a unit, a thing and then some behaviors of that thing and you've got a long list of them, cluster away lots of different applications.

Fantastic. I've got another one here from. Is it Mikel? Michelle, Sorry, I saw the Finio graph. Are there any free solutions to cluster data from social.

Yeah, Michael's referring to this graph and which just came from Affinio. I don't know if there are free off the bat solutions available. There are lots of places you could go to do clustering. This is again Twitter data. I don't know if there's ones that are free off the bat. So my caveat before was once you pay a subscription, then it's free to run and I think there's a few services that will give you that ability. But in terms of just walking off the street and wanting to do a clustering, I personally don't know of any. I bet that there are some out there. It might be a bit hacky But I bet there are some out there and if anyone's got an interesting problem that they would like me to take a look at for, I'm really happy to run this and show them the results on their data set. That'll be a fun thing for me to do. You learn something new every time you do this, every single time with every brand or artist or product I run this for. Please just get in touch and I'll take a look for you. But walking off the street, I'm not sure. I'm sorry. If anyone does have any ideas, fill up those comments.

Yeah, there are quite a few questions here, David. I've got another one here from Luke. Does this apply for smaller customer numbers, social media followings in hundreds? What are the first steps from this size?

Yeah, that's a really good question. I mean, in my experience, if you have, if you have a reasonable number of Hundreds, you have 3, 4, 500, you can get something interesting out. You don't want to break it down into too many clusters. Otherwise obviously each cluster is quite small and therefore potentially unrepresentative. But it doesn't take, in my experience of using, it doesn't take more than 3, 4, 500 to get something really, really, really interesting out. And then you have a separate conversation with yourself about, well, how representative is this? So the way I think of it is you've got an insight, hey, there's a bunch of people, maybe it's 50, maybe it's 100, maybe it's 300. There's a bunch of people who do this thing and they have these behaviors and therefore they want these things. You then have a separate question to ask yourself in the real world outside Twitter. How big is that? Is it 10x1000x is that it is 100% of my audience that 50 people. So the way I think of it, clustering them will give you an insight. You then have to think separately about kind of sizing and validating it. But you're still like a hell of a step forward. And the beauty of that is, which I didn't mention the presentation, it's completely driven by the algorithm and so you don't have to say, oh, I wonder if there's photographers in this data set. If there are, and if that's important, they will appear as a cluster. And if photography is not an important dimension, and whether you love follow BBC Earth or not, if it's not a differentiator or just fade into the background, it'll be a subtle behavior of each of the clusters. But it won't be the main feature. So I think do it on even small data sets. Don't be wedded to the results. Know that they're true within that sample. But you'll learn something that will then make you think differently in the real world. When you talk to people or do more research or get more data, it will move you forward and help you to be just one step in the right direction. It'll tell you things that you didn't expect because it's not hypothesis driven. So yeah, go for it.

Good stuff. David, just staying on that same question. I have one for you. You mentioned a lot of examples of music artists and politicians. I wasn't sure if it was because you were working at emi so you got to happen to work with all these celebrities and music artists, David Guetta, Katy Perry and all that. Does this model work better for individual brands like personal brands, like politicians or influencers, better than for a business, a retail brand or, or do you have any examples or just curious to. To find if it works better for personal brands rather than a company for organization, for example.

That's a great question and I think it goes to Sophie's question on the thing as well. She asked what if you. What if people don't feel passionately about your brand? Maybe don't follow you on Twitter, but maybe financial services or other brands. So the question is then you know, where does this work and where does this not work? I think you know, there's nothing about the method that doesn't work in any given situation. So they. Plus if you have the start there, if you have the data, if you have some data on people, then clustering it will tell you interesting and exciting things. Almost always in my experience, if you're in financial services, maybe your weighting of these three different data sets is slightly different in terms of how much importance you put on them. Maybe the gathered data, so Twitter data or other social media, maybe that's bit less important, is still really, really useful because you'll still have some audience and it's still useful to know who they are and which subgroups there are still useful. I'd still do it, but maybe that's a bit less important for you. But you know what your strength is in your own data because you'll have wonderful records for each person of their behavior in your product, how many times they log onto your bank maybe or the amount they spent each month or their balance or the retailers they spent it with maybe. So you'll have incredibly rich own data, much more Useful than other companies. Maybe social is a bit less important. And survey data, ask data for me is the one always useful, almost always useful. So I'd say in financial services, maybe is owned and then survey and then gather in that order. I think for I worked in luxury goods and at the top end of luxury goods, don't try surveying billionaires, they're not big into online surveys. So the ask data becomes not very important, not as social. And so again, the own data becomes the big one. And in music or entertainment brands or politics, even gathered data becomes really important. And asked data, surveys or polls become really important. But you don't have that much owned data, you don't have that much richness of what people have actually done in the past. And so, you know, that becomes a bit less important for this exercise. Bit less useful, less insightful. So I think the answer is yes for financial services or for retail? I've worked in both of those. And yeah, absolutely. The lessons here are 100% apply. You should have a different level of importance on these three data sources on the left hand side. And so, yeah, so do it. And then ask yourself, how useful are those results for this decision or this decision or this decision? They're definitely useful, but how useful are they and what relative priority do I put on those different data sets? That's what I would do.

Fantastic, David, excellent stuff. Sophie says thank you for that and she's on to build their presence on Twitter.

Now, Sophie, I think that's an absolutely brilliant point and I've worked with a number of brands, brands and over the last few years and they said, oh, I'm sorry, we've invested a lot of time and money in Instagram. I'm like, well, that's lovely and it might be where your big audience is, but you ain't going to learn a lot from that. So why don't you just put a little bit of time and effort into Twitter because you can learn a lot from it. So it might not be your biggest revenue generating channel, but boy, it's going to be your most insight generating channel. So I think you're absolutely spot on, Sophie. Invest in Twitter just for the purposes of really understanding deeply who your audience is, who it's not, who your competitors are and the relative strengths of you versus your competitor in different segments, all that stuff. Do that for sure.

Right. Without being biased. I'm actually put a link to a social consumer segmentation tool called Audience. So do check it out. They're our event partner as well, so they're really really cool tool if you're looking at building clusters. So a shout out for them and yes, if you've got any more questions, do put it on the Ask Question tab. But if not, you can connect with David on LinkedIn. Twitter is on Twitter, of course, but also he's on the Inside Network, so do come on to the Inside Network, sign up to it. We've got some really good members there and we'll be running some Ask Me Anything sessions in the interview in the coming weeks and months. So a session is similar to this, so David would be on that as well. So do sign up to the community and connect with David if you've got any more questions around clustering, so he'd be happy to help out.

Yeah, I've got all the time in the world to share more examples, go down weird rabbit holes of different industries and run analysis and play with this. Like, I love this stuff. And so if you're curious in any way, feel free to reach out. We'll have fun looking at it together. Promise.

All right. Fantastic, David, the passion does come through. Thank you so much for this and your time and enjoy your rest of the day in Connecticut.