Drill to Detail - Drill to Detail Ep.36 'Marketing Analytics, AI and the Rise of the Omni-channel Consumer' With Special Guest Mark Madsen

Episode Date: July 24, 2017

Mark Rittman is joined by Industry Analyst Mark Madsen to talk about marketing analytics and the rise of the omni-channel consumer, the use of AI in analytics and personalization and what this all means for brands, for advertisers and for marketers.

So my guest in this week's episode is someone I've known for a number of of his writings and presentations around the topic of marketing analytics, given the product management work I'm currently doing for a startup in this kind of space. So I'm therefore pleased to welcome Mark Madsen onto the show. Mark, it's great to meet you at last, and welcome to Drill to Detail. Hey, thanks, Mark. So Mark, do you want to just introduce yourself properly? What's your history, what's your background, and what's the kind of work that you do now? Okay.

Well, history and background is that years ago, I started off getting out of college in AI, which is sort of ironic given where the market is today because I couldn't find work doing AI stuff back then. So I ended up programming and doing other things instead. Eventually I ended up at Carnegie Mellon University doing research in cogsci and then robotics for self-driving, actually, excavation equipment at the time,

which is also ironic because you couldn't really make a living doing that either so um you know off we go to what can one do with behavioral economics and cognitive psychology and stuff like that and the answer was decision support you know which then sort of morphed into business intelligence and then i stayed there for a long time. And eventually, after a series of jobs in IT and vendors, I went out on my own about 2004, 2005. I just decided that I wanted to do research and do work. And the only way to do both of those was to work for myself. Okay. Okay. Okay. So the paper that I read of yours recently that I thought was interesting was a paper, a presentation called A Pragmatic Approach to Analyzing Customers.

And I'll put the link to that, I think it's on SlideShare, in the show notes at the end of this. And what I thought was interesting there was you were talking about the topic of marketing analytics and customer analytics. And that, as I say, that resonates with me at the moment because of work I'm doing in a sort of similar sort of area. And one of the challenges or one of the kind of the directions that the industry is taking at the moment is this away from this idea of what's called conversion optimization towards looking at the lifetime value of a customer and the analytics around that. And I guess, first of all, let's kind of take a step back. What prompted you to write that presentation? And what were the changes that you were seeing in the industry that led you to kind of to do that and led you to your interest in that area?

Well, actually, what led me to the interest in the area is, I mean, it's always sort of been there. There's a sociological aspect to a lot of anything that you do when you're selling products, whether it's business consumer or business business or whatever. You're still dealing with people. And so a lot of the BI work always had a customer focus to it, you know, way back when. And then, you know, when the first version of the web came around, I wrote with a couple of friends, what was the first book on web analytics, which hit the market just in time for the entire web startup market to collapse. And then nobody wanted to buy books on web analytics because this stuff was all going to go away so yeah it's it's the best book on web

analytics you can't find um actually i think there's some really excellent ones out there now but uh um yeah so you know i was kind of involved and then i was doing consulting in that space and a lot of it was, there's the basic stuff, but every single time you get past the boring metrics, which are page counts and unique user counts and things like that, you start to ask questions about people and what they're buying and which people and you start talking about segmentation and pretty soon you're back in a lot of the old lift modeling and consumer behavior modeling and you know prediction and so I was working on circuit 2007 I just got completely out of bi and most of my work had been in marketing support in fact I left a director of warehousing and I ran

a consumer analytics team and some other groups as well all around this stuff. And I felt like that was the biggest pain point in the industry at the time. Marketing has lots of money. They're data poor and they've never. Well, at the time I wrote that talk, they didn't have access to really any useful internal information. They had no systems. Everything was done via agencies outside the company. And so, you know, it's been a long, slow crawl for the last 10 years to get kind of get marketing

instrumented to the level of the rest of the businesses. Okay. Okay. Okay. And I guess other things that have happened is the amount of data we have now is huge. The amount of data we all emit is huge, buying signals and social signals and so on. And so I suppose most marketing companies aren't short of data these days. So where do you see the innovation happening? And where do you see some of the interesting things happening in this kind of space really you know what's being done with this data that's that's good and i suppose how has it changed the selling process and the marketing process to your mind um yeah there's multiple questions to unpack there i i'm just

thinking through you know um what has evolved is the level of view or possibly an interaction with customers that they never had before. And this takes place right at a time when all the retailers had pretty much finished firing all of their statisticians because every retailer used to have a statistician on staff. And now they're all hiring them back again because there was a period of 15 to 20 years when they all got laid off and um you know they punted the responsibility for consumer analysis to either big data

syndicators like you know nielsen and iri secondarily to companies that are now in the market like axiom um or they or they gave it over to the consumer packaged goods or the product manufacturers who would buy up that data and had the modelers and could see across multiple retail channels and could provide consumer insights. So it was a very highly fragmented market, and it's still highly fragmented. But where things started to change was that everybody suddenly had visibility into this. And, you know, I spend most of my time or have spent most of my time working with companies that sell to other companies after I left that job that was consumer facing. And I was surprised at how naive most of the programs were. Like email marketing, most email marketers have never heard of RFM,

something that's been around since I think the 70s. You know, direct response modeling and understanding how these things work. And so I feel like in some ways it's still very backwards. And then you asked about how has all of this online stuff and all of this data, and how has it changed things? And it's kind of changed marketing practices for the better in that there can be more informed processes and practices and a retooling around measuring outcomes over measuring process you know the old model was

let's measure how much reach we have and how many unique visitors and page views or listeners and that shifted uh to um you know what well because there was some ability to make some connections lift and other things and and feeding the funnel and all that kind of stuff. So that's a good point, actually. You mentioned the obligatory funnel at one point in your presentation. So what were you kind of getting to with that, really? It was a dig at online marketing. Yeah.

I mean, so I guess on that point, I suppose the change I've noticed is around, I suppose the focus away from conversions and more towards looking at the lifetime value of a customer. Is that something you're seeing or is that just talk or what really? Well, that's really what, you know, that lifetime value stuff was really what we were working on 10 years ago. I think it just never became common practice because it's not that easy. You have to have somebody who knows how to build a mortality calculation that can give you some level of predictor

of when a customer stops being a customer, because there's the current lifetime value, which is how much they've spent with you to date, which can be measured in a lot of ways right margin and revenue and so forth but then there's also the predicted lifetime value which is the the potential future value and whether you can move the needle on that is the interesting question because people become customers and then they stop being customers and there are natural life cycles to a lot of this and so you know the holy grail is obviously doing it at the individual level as opposed to doing it segments and recommendation engines you know they make recommendations based

on all sorts of things and you're trying to do that you're building out models that are making assumptions but sometimes wreck engines cross-selling for example have negative impacts on lifetime value and so usually things are too narrow and and so lifetime value is really one of the key metrics and then in startup world because software marketing has shifted to subscriptions you're seeing a very heavy emphasis on exactly that right they might call it something different but lifetime value calculations over the value of a subscription to be able to forecast financials and growth and you know all, all the other stuff. Okay. I mean, again, in that presentation, you talked about post-purchase data and, I suppose,

non-transactional data as being useful to understand, again, the kind of the intent of the customer over time and so on there. Do you think this profusion of data that is beyond the kind of initial conversion, is that being used kind of usefully? Have you seen that being exploited and used as much as you'd hope to, really? No, it tends to be very fragmented in its use. Like the funnel stuff that you talk about,

I mean, that's been in there for a long, as an idea for a long time that, you know, there's, I mean, that's really a very old idea. But at least it positions the idea that the purpose of marketing is to generate, particularly when we go into the software marketing world and the dot-com world, loses sight of that core function and thinks that it's something entirely different, like generate leads. And generate leads is the front part of the funnel, but lifetime value is the important thing. So if you're building advertising campaigns, one of the interesting things is when the cost cutting hammer comes down, which campaigns do you stop? Affiliate marketing, SEO spend, you know, what is it that you're not going to spend money on? And most of the time, it's the one that costs the most or it's the one that generates the fewest leads.

And usually, if you analyze the data through the transaction pipeline, and particularly if you have a lifetime value that's reasonably reliable at the end, you'll find that there are campaigns that for their spend generate significantly higher margin or longer lifetime value customers and so that's kind of an end-to-end view of that process even though all you're measuring is something that's an input at one end and an output and there's kind of a big black box in the middle which is the rest of your organization. And that's where all that intermediate data comes in. The conversions, the engagements, the calls, if it's things like downloads, that, the direct sales activities that all are part of that pipeline.

And those have been highly fragmented systems in most enterprises. Okay. So there's quite an industry now around things like personalization and influence and so on. Do you see that having much of an impact, really? I mean, there's a lot of, I suppose, every CRO company then wants to become an A-B testing company and a personalization company.

But you could argue maybe sometimes doing things with data doesn't have the impact that you would expect. I mean, how successful have you seen personalization and that kind of industry being over the last few years? Narrowly successful, which is another way of saying probably not that successful. Is that in comparison to, say, A-B testing? Or how would you qualify that? Yeah. So, I mean, you know, a b testing is is just a technique it's a very good technique when it's done right

for which works better a or b the hard part about all of this stuff is that you know you start out with simple things like let's make a landing page for an email marketing campaign and when people click the link they'll land on that page and what's the optimal page so you you do a version and you divert some of your traffic to that and you do these controlled experiments which is all a b testing really is you know is a better than b and and over time you refine and evolve to a better, say, offer, landing page, whatever the thing is that you're doing. And, you know, and it's being driven by the data. The hard part about it is that there are really interesting, very subtle, I don't want to say side effects, but things that happen in controlled experiments. Like, I'm trying to remember what it is. I think it's called, no, that wasn't Twyman's law.

That was something different. There's this idea of novelty, right? So the effect of things like new give incremental bumps and they may last for six to eight weeks. And so it looks like it's doing better but in fact if you measure it for long enough it comes out doing worse and so you have you have these these techniques that you use but then as you get better you start to measure more and wider and you if you're good you begin to integrate more different perspectives on things. And that generates a

lot of complexity over, you know, the initial, let's do A-B testing on, say, webpages or campaigns. And so then you move into the more sophisticated things. And that's where, you know, I mean, personalization, which has been a holy grail for a long time and the trick with personalization is that most of the time you don't have enough data about an individual person to be able to personalize in a mathematical sense you can indicate interests but behaviors can diverge from interests very easily. And people also are reticent to disclose things that may be embarrassing. So you have all sorts of weird stuff that we learned in survey theory years ago in market research that comes home to roost. And so that's where I feel like you can often say a group of people will on average behave

in a particular way but if you try to take a particular person and say because this person is a member of this group I can predict their behavior if you follow that line of reasoning it's a flawed line of reasoning because it's not you're now trying to say this person is, is behaving in a predictable and deterministic way. And that is almost never the case. Yeah. So,

I mean, I guess that's why you have segmentation and so on there. I mean, it's, how have you seen, I've noticed the other side to it is, is doing this app, doing this,

like you said, in a repeatable way, such that the, the uplift you get is not just going to kind of run out after a few weeks. It's something that is a more scalable thing for a retailer. Have you seen this stuff being done at a scale or in a programmatic way where, you know, the kind of the influence and the personalization is done in a consistent way over time for these segments you know rather than being opportunistic i mean i guess the question to you is who is in your mind you know what are some good examples you've seen here or where do you think the industry is going

in this sort of area i i think it's when when you start talking i'm sorry this whole focus on user experience is a great example right design thinking and user experience these were actually before marketers really got a hold of the meaningful terms for research you know they came out of human computer interaction research years ago. And really, the core there is starting with what people need and want. And the other is like a particular user experience. It's end to end. And when you take things like A-B testing or personalization, you're taking a slice of a customer interaction over a life cycle of their interaction with your company you know whether it's a mobile phone plan or um you know buying

a toothbrush right there are cycles and patterns and re-engagements with things and And, and so when you talk about a lot of this, it all kind of, it's all related in some way. And so if people start talking about user experience, they need to see what's the very first point where a person ever heard of you, you know, was it an advertisement? Did they read an article on the newspaper? And, you know, does it, you know, all the way down through customer attrition at the other side of it. And so when companies start thinking about it, they start thinking about, well, where can you meaningful pull a lever and actually influence something that matters like market share, competitive acquisition of customers against your competitors, or it's a new area

and it's a new area and it's a new product and you're doing a product launch and what levers can you pull to make that product launch more successful or how can you keep people using your service rather than slowly you know falling out of that service um and and so that that means you're really taking this kind of business and business process focus and then coming to it and saying, well, what's possible? Is it possible to do A or B? And great examples are companies that have done a really stellar job in loyalty marketing. I used to do loyalty marketing and loyalty analysis. And that's an interesting area for me because you really have to define what you mean by loyalty and how you measure it. What exactly is it that you're measuring? And usually loyalty is a measure of individual behavior or aggregate group behavior of customers.

But there's a brand aspect to it, right? People are loyal to a brand or maybe a product, but that product is usually heavily branded in that sense. I love my iPhone. One of my favorite examples is Gucci. Gosh, it's been years now, but they were one of the early ones into trying to tie their loyalty program in with mobile technology and other things. But, you know, from a customer brand and loyalty perspective, they did a stellar job. For example, if you look at, I assume it's still true now, but if you look at the advertising they were doing five, six years ago, maybe seven years ago, they don't have QR codes or barcodes on these one or two page spreads that they do in magazines. They just have pictures because their view is that the artistry of the photography that goes into generating that image shouldn't be tainted by this big, ugly QR code stuck on the corner of it so you can scan it with your phone. So they built a mobile app that was tied in with their

systems where you could take a picture of the magazine or the billboard or whatever it was that you saw and then they would run image recognition against that identify the products that were in that spread which is a fairly nicely bounded kind of image recognition problem it's a limited set of products and advertisements that they know they have and then you could take a picture of it and it would immediately put it on an interest list. And then the next time you walked into a store, you could give it permission and your phone would immediately tie into the SOAR system and potentially the people in the store could have the things in your size ready for you to try on when you got there.

And they were very careful about how they approached this too right a very detailed understanding of human behaviors required because otherwise it seems like spying and so the application of these smart technologies and techniques is equally important but that's a great and stellar example of a company that really understood loyalty and branding and exclusivity and all these other things so so i mean you mentioned mobile there as well and and and how much do you think that i suppose the kind of omni-channel nature of marketing and being a consumer these days is effective things because certainly it's hard

to tie identities together you know across different devices um i guess the conversion rate on i don't know mobile is probably less than desktop and so on i mean i suppose we've got both both kind of opportunities and threats really as people start to do more mobile how's that affecting kind of marketing and marketing analytics in in your mind really or marketing technology it's um it's it's multiplied the number of marketing channels you know when i when i would analyze these things i would look at sales channels you know where do you sell and how do you sell you know and there are other sort of direct sales models and the indirect models like selling through third parties online selling through other people's stores and and those all started to blur to the

point where you can buy products in so many places and then the other is the marketing channels you know it used to be tv and radio and print right those were your big three lovers and and that's how you got your brand message out or how you got your promotions out and obviously there was telemarketing and i worked in direct marketing so we would send out catalogs as well as doing radio spots and deep spots and and then we had uh you know blow-ins the kind of things that you get in the mail circulars uh all sorts of stuff and then the website introduced web advertising affiliate marketing and then of course seo is actually a marketing function it's trying to figure out how to get search terms to land at your listings or products or information

and then you start throwing in the social aspects which you know we can see where all that's gone and marketers put advertising dollars or promotional dollars where the people are, and so it causes other things to dry up, but it also creates this problem where you had very few things to pay attention to before, and now you might have in total, if you looked at detail, 60 to 80, maybe more, and then you have to aggregate those up into groups and categories and this is you know and then try to optimize across this and it's an incredibly daunting task and so i don't even track channel share anymore there were some really great uh places out there you

could find online that would generate where marketing dollars go by various channels or techniques within channels. Like they would lump all of the various things into online and then search ads versus just basic banner ads versus affiliate programs versus this versus that. And then mobile really changes things because the mobile ads are slightly different. A lot of mobile tech is trapped inside of apps rather than going to websites, which I feel sort of sad about being an old line web person. But because of that, it's closing out the web and it's making things proprietary. And so you have the rise of the social networks, the Facebooks and, you know, the LinkedIn, you know, various sort of walled gardens, which then take their own advertising and further cause difficulties with how and where you analyze that. But the mobile thing also gave you a method of engagement as a consumer where before all the

online stuff was focused on you sitting in front of a computer and it's taken a really long time to escape the idea that i'm sitting in front of a computer seeing something and then i'm going to the store or i'm measuring a transaction by looking at click-throughs on a computer because I might save it. I might see it at work, then look it up on my phone or see it on my phone, look it up on my laptop, or if it's a considered purchase or not. And then you throw into it the fact that most people take their phone with them everywhere, so they're in the store. There's all these scanning apps that you can do price comparisons oh this is 20 cheaper on amazon i'm going to buy it there instead of my local store which um you know it actually this is interesting too it creates the negative dynamics

in the market that actually cause the market to become both less efficient and worse for consumers but it's the village greens problem of everybody grazes their sheep on the village greens until there's no grass left and then all the sheep starve. And that's retail today. You go into the store and you shop with your mobile app and you price shop and then you order it off Amazon and you get same day delivery. Well, that physical cost of that store and that retail display were fronted by that company on the hopes that you would buy there and the moment you do that and a hundred other people do that it has a material impact on that retailer and so you see the decline of of in-store formats particularly

with easily commoditized products like consumer electronics. And that dynamic kills off that, and then that's gone. And then you can't do that anymore. And then what happens? And so the instability and the constant flux, I would hate to be a marketer of these kinds of products today because it would be so, so challenging to deal with this, which is why there are so many more marketing costs than there ever used to be yeah yeah what was your take on um on amazon

buying a whole food whole foods i mean that was uh certainly uh that was quite an impact on on the industry at the time um yeah why do you think they did that and what do you think the impact that will be it was kind of, in a way it was expected, but there were people who had been saying they have been eyeballing that market for a long time. And it's funny because Whole Foods and Wild Oats used to be two customers of mine years ago, back when that was a new product category

that was kind of disrupting retail because it was all this natural and organic and then they all consolidated into whole foods which became sort of the biggest standalone out there and um you know along comes amazon and amazon had been looking for market segments right because you can't they can continue to tap the incremental gains that they get competing with you know the walmart's and other companies and the small outfits right as category after category shifts online and amazon gets a chunk of that their next growth area had to be in food and there have been a lot of people trying to do food and home delivery and they probably have better delivery logistics and planning than

anybody else. So that's the real strength. And so enter the home delivery market where they already know the frequency with which things are going on. And it seems like a natural foray because Whole Foods is a profitable standalone business. It doesn't have to eat a giant investment in delivery that some other delivery firm has to do. And it doesn't have to eat a giant investment in online that, say, a Safeway or Kroger's or, know whatever your favorite grocery chain is Tesco or something and so it's an interesting we'll see how they do but it gives them points of presence in local markets it gives them very what I would consider high end in terms of product selection high ends in the higher priced higher margin categories of grocery fresh grocery and package grocery it also also gives them a certain minimum i suppose there's a minimum level of coverage you need to have um to be delivering groceries around the us you've got to you know

because it's because the the everything spoilable it i think it was ben thompson talking on the on the stratus trajectory um podcast or on his website about uh it gives them a certain it gives a certain level of business that means they can go and offer fresh deliveries around the country which you wouldn't then have without that kind of whole foods market and it also then means you can then leverage that into supplying for example i don't know restaurants and hotels because you've got that level of business you're going to get almost it's their first and best customer really yeah the other thing is that if you look at store distributions a whole foods tends to be upscale urban. And so their locations are primed for a customer density that also tends to be, you know, the call it upper middle class segment of the market or the aspirational market segments

where people have more money than time, right? Which is a key aspect for any home delivery business. And so it makes a lot of strategic sense if you are in the grocery business to go that way. And I'm very interested to see how it all pans out, but it definitely, it gives them something that the other competitors in the delivery space or the grocery space are going to have difficulty challenging and they have a scale of purchase which gives them you know cost advantages in the same way that walmart typically beats mainly because they buy more stuff than anybody else. And so their prices tend to, or their margins tend to reflect that. So do you think anybody can compete with Amazon now? I mean, and I suppose the link into what we've been talking about is,

can analytics and maybe kind of AI and data and so on, can they help other e-commerce businesses compete with Amazon? Or is it really game over, do you think, in terms of that game has been lost now? There's always room for niche players. But in most markets, especially these unregulated capital markets that we have today, you know, there's almost no monopoly law anymore in almost any country. And so when you look at Amazon's share across category in almost any business, they're pretty much the dominant player. And the same in Walmart, they killed hundreds of different categories of business that had regional and local flavor to them. And, you know, this is just a natural ongoing play in the market

towards heavier and heavier large corporate involvement in ever smaller niches. But that, you know, you look at, well, yeah, I was going to say one thing and then I changed my mind halfway through. Here's why. I was going to say, look at Zappos, right, which was a shoe company that had great policies, really interested in customers. They bent over backwards to give you a great online experience, but also a great offline experience when you received your product. You could order three different sizes not knowing exactly which one would fit you and send the other two back and they would take that

and so people really loved that company and they were very very much built around understanding the customer base and doing this amazon bought them right i think there's there's never a price when companies have deep profits there's never a price. When companies have deep profits, there's never a price that's not too dear that everybody's got a price in mind and if they can exit at that price. I look at companies that are doing interesting things now, and probably my favorite, you could argue about its effectiveness, so I won't go down there, is Stitchfits. I read their blog religiously. I watch what they're doing.

I know a bunch of people who work there, so maybe I'm biased, but they're all really great analytics professionals, data science professionals. And they're doing all kinds of interesting stuff with style matching and style development and building all these different attempts at recommendations because personal clothing is very idiosyncratic on the one hand i like what i like but also on the other side of it there are definitely trends in styles and colors and you know other things in fashion that cause some level of conformity and and uh you know it's within that level of conformance that you have differentiation which is a weird thing i used to work in fashion years ago do you want me you've

everything so so the data so so i i you? So I mentioned AI a minute ago, and the last thing I want to talk to you about really is, I suppose, the emphasis and the marketing activity and the noise around AI in marketing analytics and analytics in general. So obviously, recently there was the announcements around Einstein from Salesforce. There's obviously Watson from IBM.

What's your take on the impact and value and effectiveness of AI in analytics, particularly marketing analytics, the last few years? AI, it sort of depends on how you use these terms. You know, AI tends to mean to a lot of people this sort of general intelligence concept. And we're very, very far away from that. I mean, you know, we went from what were called, you know, big world problems or open world problems, where there's a lot of uncertainty and ambiguity, which is what living organisms deal with, to solving very small world problems, small world closed problems like recognized cats and images online, which is great.

Actually, if you're a fan of the show Silicon Valley, there's a great take on it, which is hot dog and not hot dog. And so the food thing can recommend hot dogs and nothing else it's the ketchup isn't it it's the ketchup it recognizes i think yeah so it takes a lot of training the interesting thing is that training and machine learning is just like training a person you know you you don't grow up knowing cat from dog you learn it by experiencing many, many, many different things. And so when you train a person, a child, to recognize cat from dog or hot dog from not hot dog, you don't really think of this as training AI,

but that's essentially what you're doing when you build something like that. So you feed thousands of images of different things in. But what have you built? You've built an image recognizer. And what does it recognize? Images of things that you've labeled, but somebody had to label all those images the first time. And so discovery of these things, automatic categorization of them, you start to get into these uncertainty and ambiguity worlds. And these are the researchy interesting areas that start to touch on the open world problem. And so what you're really talking about then is applications of machine learning with statistical techniques.

And, and those, those techniques, there's lots of them and they, different techniques are useful for different problem domains. And so the art of data science or analytics is I have this problem or this goal, because typically your method of working is translate the goal that you want to achieve into an analytic context so that you can state a measurable outcome. Because when you build a model, you have to test that model and that model's got to do something. And so you measure the performance of that model in terms of, you know, accuracy, reliability, something.

And that means you have to have a number or set of numbers against which you can rate that model against just toss a coin right because your worst case scenario is that you are random or worse and so um and and so that uh that that means that the problems are very narrow a person translates them into an analytic goal and then says i can use any one of these five or six different techniques or a collection mixed together of them which is called an ensemble and you mix those things together or you pick one whichever set is the best that's what i'm going to use but you you have that goal that analytic, the numbers that you have boiled it down to, that you're going to measure against. And the other part, which people seem to forget a lot,

is do you have the data to actually solve that? And if you have the data, do you have enough of the data? Because it takes a lot of roll counts. Yeah, that's been my experience. I think people tend to underestimate the amount of data you need to get meaningful predictions or meaningful kind of results out of these things. Is that something you find as well?

Yes, absolutely. That's a key thing right there. So, yeah, so going on from that then, you mentioned Einstein. So what I think is really funny is this whole market, right? We had IBM with Watson. And well, what's smart? Well, you know, Watson, I like it because they didn't say Sherlock, they said Watson, right? Because Watson is the smart guy that works with Sherlock. And so he's your intelligent assistant who helps you along. But you know smart guy and then and then along comes um salesforce and he says well einstein because einstein's smarter than watson and he's

a genius and so we're going to call it einstein and then they do really rudimentary stuff with it and then along comes sap just this last week with leonardo well who's smart too that hasn't been taken by some other vendor well we've got leonardo here. So now you've got a three way battle between Watson, Leonardo and Einstein. Who's going to win? There used to be a great celebrity fight club claymation show, and I looked at, you know, I used to do this analytics work. And one of the areas outside of customer analysis and direct response was consumer fraud and business fraud. And I did work on, you know, stock market fraud and stock trading fraud. And that was back in the 90s. And that was one of the few areas that had enough data of the right type. And we had techniques that could be executable on the hardware that we have.

Hardware is no longer a problem. Software is no longer a problem. Techniques are no longer a problem. And data is only problematic in that somebody's got to sift through it and figure out what you can use. And so all these startups kicked off about seven, eight years ago, making easy to use analytic tools, which is exactly the wrong problem to solve because it presumes that you and I are smart enough to sit down with this analytics tool and say, do you want to use this kind of neural network or that kind of neural network? Do you want to use a Bayesian classifier or would you prefer a different classifier? Tell me what you think. So they tried to boil it down by saying,

well, we're going to use K-means clustering and we're going to use, you know, I don't know, some logistic regression. They picked a set of techniques and threw them into a tool and then they let you just dump data in. But did you have enough data and was it the right data? And is there any bias in your data?

And have you prepped the data properly? And is there sampling problems with your data? None of that is solved. That's what the experts know. So my take on that was that market will slowly collapse in on itself, because we don't need easy to use analytics tools. We need those tools for the experts to define the models that we want to build where we build custom models. But in reality, where you have a narrowly defined data set and you have a business application that has a process in it and you can optimize that process, why not just embed that thing inside that?

So you're going to see two things, a rise of verticalized analytics or data science or AI or whatever you want to call it, where you have some service you can feed your data to. It's a well-understood data set. You get the data, they do the thing and out it goes, or they sell you an adjunct piece of software and you install it on premise, or it gets into oracle and sap and salesforce and you know great planes right any of these packages in very small pieces and the economic model of building those software applications means that it's very hard to make money on small amounts small sales for a hundred different things.

But an ERP vendor can easily do this. And so those announcements are a very logical and almost predictable endpoint for a lot of analytics. The real question is, for a company, while it might make you more efficient or more effective, it makes all your competitors equally efficient and effective because once SAP builds it or Salesforce builds it, everybody's got it. So the true value in analytics, just like the true value in BI or other data is how you apply these things in novel ways in your context and so i i see leonardo and and these other things as you know the einstein as being great and necessary but they do not become

sources of differential value for a company the custom it still came there fantastic fantastic so so just to round off mal there's a couple of conferences and events that company. The custom, it still came there. Fantastic. Fantastic. So, so just to round off, there's a couple of conferences and events that you're involved in speaking at that are quite pertinent to this really. There was the Strata one and there's an event called the Accelerate Digital Conference. Do you want to maybe explain what they are and just tell us where they are and what your involvement is with those?

Sure. So yeah, I'm on the committee for conferences with O'Reilly for Strata. And there's a couple of conferences in sort of our realm of the market with data. One is the really big one, which is the Strata Data Conference. And that one is just, it's more,

I would say it's more technology focused. And so the Hadoop market, the Spark market, the big data and market in general, a lot of analytics, a lot of new world analysis, as well as BI and databases. And it's all things data, essentially, in that conference. There's the O'Reilly AI conference, which is focused on AI and machine learning and, you know, targeted at sort of the practitioner crowd in that field. And then the other, the one that I'm chairing right now is called Accelerate, which is an affiliated conference with the data warehouse institute and there the focus is more more you know not the hardcore data science people like a like an ai or a predictive analytics world would be but the the average analyst the people working at companies who are trying to make themselves a little bit better right going? Going from BI to learning statistics,

or you're new into a data science team where you're trying to build a data science team and you need to understand how to do that. And so the Accelerate conference is aimed at kind of the education and onboarding of people going into companies as practitioners, not necessarily working as researchers or vendors. So those are the things that I'm involved with and spend a lot of my time.

Sounds really good. And just to round up then, Mark, I mean, how do people get hold of you? How do they find your website, your details, and so on? Well, the travesty that is my website that hasn't been updated since it was hacked five years ago is thirdnature.net and then on Twitter it's Mark Madsen

and I'm at a lot of conferences so you can usually find me out and about somewhere in the world Excellent Well it's been fantastic speaking to you, thank you very much for coming on the show, Some really good insights there and thoughts on AI and so on there. So thank you very much. And it's been great to speak to you and have a good rest of the day. Thank you.