Ep. 65 | Beyond CDPs: Customer Intelligence
This week Michael Greenberg, founder of Retina.ai joins Allison Hartsoe in the Accelerator to discuss customer intelligence platforms. Where CDPs are all about organizing and selectively activating the data, CIPs or Customer Intelligence Platforms are all about rapidly building a single source of insight. By definition, insight means continuous learning in a looped system. Hear Michael talk about how companies can benefit from a data system that makes recommendations to you.
Please help us spread the word about building your business’ customer equity through effective customer analytics. Rate and review the podcast on Apple Podcast, Stitcher, Google Play, Alexa’s TuneIn, iHeartRadio or Spotify. And do tell us what you think by writing Allison at info@ambitiondata.com or ambitiondata.com. Thanks for listening! Tell a friend!
Podcast Links:
Read Full Transcript
Allison Hartsoe: 00:01 This is the customer equity accelerator. If you are a marketing executive who wants to deliver bottom line and impact by identifying and connecting with revenue generating customers, then this is the show for you. I’m your host Allison Hartsoe, CEO of ambition data. Each week I bring you the leaders behind the customer centric revolution who share their expert advice. Are you ready to accelerate? Then let’s go.
Allison Hartsoe: 00:33 Welcome everyone. Today’s show is about customer intelligence platforms. And to help me discuss what this is and why you should care about it is Michael Greenberg. Michael is the CEO of Retina Ai. Michael, welcome to the show.
M. Greenberg: 00:48 Thank you so much, Allison. I’m really excited to be here
Allison Hartsoe: 00:50 Now, I absolutely love your physics background. I find it incredibly compelling how people who didn’t start out in marketing fields at all ended up in this space. So can you tell us a little bit more about your background and how you founded Retina Ai?
M. Greenberg: 01:07 Sure. I’d love to. Um, and you’re right, it’s kind of marketing is taking all types. I think that’s one of the amazing things about the field. Um, so my background starts actually next to NASA. It was kind of preordained that I would go into physics. Yeah, I always joke that if I was born in New York, right, I would have been in finance. So what uh, what did it happening was this, I was working on a project, it was called the Lego project. So for those listeners who might be following super nerdy topics, that was the laser interferometer gravitational observatory, which just means we were trying to detect gravitational waves. And what was interesting there was that was my first touch on anomaly detection. We were trying to detect something that was 1 billion the size of the width of an atom. And so for that, you’re obviously trying to filter out all the noise, everything that’s going on around you. And I think that was the first time I actually got my first taste of data science. So we just called it science, no real, no real
M. Greenberg: 02:02 extra append there. And as life went on, I found myself with a position at UCLA, and there was this program there called Startup UCLA. Long story short, I met a lot of entrepreneurs, and you know, uh, all sorts of things happens when you hang out with that kind of a crazy group. And I ended up forming my first company when I was 29 spinning it out of the university and was very fortunate that that was acquired by private equity in 2014. That’s sort of where I first really landed directly in the world of marketing. And that was with summit partners, a private equity firm out of Palo Alto. While I was there, I saw sort of a huge problem. And the problem was that we were taking these huge positions in businesses that were either e-commerce or digital retail, and we found that they were starting to go upside down. The cost of acquiring every new customer was a little bit more than what they were spending. And this term CLV, which I know everyone is talking about now, was still kind of being they need about because it was a metric everyone wanted, but it was kind of fiendishly hard to measure, sometimes the data wasn’t right, sometimes we didn’t agree on what it exactly was. And that sort of my first touch on how data science could really start solving this problem.
Allison Hartsoe: 03:14 I want to come back to what you were saying about a private equity firm being concerned that the marketing spend was too much because we have companies like Lyft and I don’t know how many more out there that basically haven’t turned a profit and our going public. So, where you just an anomaly inside this firm, you’re like, hey, these companies aren’t going to survive, where you the squeaky wheel?
M. Greenberg: 03:33 Yeah. It’s always tough to play chicken little in a private equity shop. That is a dangerous game. Yeah. In a way, because these were people with backgrounds that were usually from consulting, meaning they could make excel thing, but they didn’t really have the techniques that we’re going to be talking about today to really predict ahead sort of what was coming around the bend. So, I was kind of starting to see something that I would say was about a year or two out and actually, my cofounder Emad Hasan who was at Facebook, saw it from inside the shop where he saw the same businesses getting less return or higher CAC on their spend, and that was a huge problem.
Allison Hartsoe: 04:12 CAC being customer acquisition costs and that’s something we always look at is the ratio between what does it cost you to acquire a customer and how long are you going to keep them.
M. Greenberg: 04:21 Yup, exactly. I would say that’s becoming the all-important ratio for, for those who follow Mary Meeker, I believe she dedicated a whole chapter in her famous Internet trends report each year just to that ratio kind of signaling a warning belt, all the businesses that might be going upside down.
Allison Hartsoe: 04:37 So this clued you in and so you were at the startup group at the time, and then you decided, I’m going to jump in with a company, that’s a big switch from a private equity firm to jumping into a company.
M. Greenberg: 04:47 Yes, it was my second Rodeo. So, you know, after being acquired, I think, you know, Alison, I know you have a background in startup land as well. I call it, it’s like joining the circus. Sometimes you go in there, and you get a little bit feral along the way. But I enjoyed it, and I have done three years at this private equity shop. I learned a lot because to your point, a lot of startup businesses sometimes never turned a profit all the way through IPO. All of these businesses were going concerns. They had eVida. They were innocent, contributing cash flow. And so it was really a very different way of looking at it. And it helped me understand the problems of what I would say non-venture businesses as well as what we see in our world. Uh, so yeah, I decided to start it and jump back in, and it’s because of one gentleman, and it’s this gentleman Emad Hasan. So in 2016, he decided to leave his great job, uh, at Facebook. I decided he’s my great job at summit partners and we formed retina to really help businesses in a sense, start answering some of these core questions. We didn’t really know exactly how we were going to turn it into a product yet. We just wanted to see if it was mathematically possible.
Allison Hartsoe: 05:55 Nice. Well let’s talk about this definition for a minute because we’ve talked about CDPs on the show and I think most people are familiar with customer data platforms, and I always think of a customer data platform is something that binds your information together, all your customer records together. And then usually they have an identity matching piece, and they have an execution piece, so they’re able to activate whatever data comes in. And recently we’ve been talking in the show a little bit about the next generation of CDPS, which we were calling CIP is or customer information platforms. And I think that’s more of the intelligence or the brain inside or on top of a CDP. Does that sound right? Do you agree? Or how do you see it?
M. Greenberg: 06:39 That’s a really great point. You know, it’s interesting cause CDPs really we’re a function almost of what the limitations were down at the database or data lake level. And so they kind of would track all of these as you put it, customer points across all of these points solutions. You know, segment IO is a great example of a big CDP, and there are many, many others. And the problem here was if organize the data but it didn’t really close the loop, and it doesn’t mean that they won’t, it doesn’t mean that that’s not on the pathway. But the big, big tier one, firms like Facebook, Amazon, they were really deploying in a sense teams of hundreds sometimes of data scientist to answer the questions of the data. And I think what happened as a response to that was everybody wanted to be a part of this.
M. Greenberg: 07:26 Back to my days at summit partners, we couldn’t hire a single data scientist for many of our companies because A, it was very tough to get them to move away from the coast. And B, many of them were making more than our vice presidents made on the coast. And so the question became, could you start answering these deeper questions around what would a customer do next, why would they do it? And more importantly, could you start building out automation loops around it. My cofounder stacker and actually was originally in building autopilots for helicopters. He actually helped to contribute to the autopilot from Marine Force One. So I think you’re seeing a lot of that idea control loop theory coming over into this and CIP to me mean one thing really. And that’s helping across the entire business build a heartbeat, a heartbeat monitor if you will. So that means everyone from the CX team all the way to the acquisition team have not a single source of truth, but a single source of insight, and it should be delivered in a way that’s actionable. So I like to say it’s kind of what data science teams are doing today. It’s doing a piece of that puzzle.
Allison Hartsoe: 08:32 Yep, that makes sense. But at the same time, I think there is this sense that AI itself is really hypey. And if I put all this data together, I can, you know, munge the number around and I can come up with all sorts of different things, but it doesn’t necessarily relate to the engine of my business. So how do you cross that gap where the data that’s coming through, you know, if I have a hundred data scientists, even if I ran them alone in a silo, I’d still need to match them with a subject matter expert. How do you close that gap that you know, relates to what a CFO might care about or what actually moves the business?
M. Greenberg: 09:08 Yeah, great question. You know, the way we think of it as this, and it’s actually interesting that you mentioned CFO because our initial customers, we’re actually CFOs. They knew something was wrong in a business, but they didn’t quite know what. And we were in a sense brought in to help them out. And then marketing was brought in after the fact. So here’s what we do. You have to have innocence, and this is where the art comes into the data science and ability to know what exactly you should be doing. And I think, you know, one of the big problems, let’s just talk about CLV for a second, is much like machine learning has a lot of hype. CLV has a lot of definitions. For certain businesses, you could have a data scientist to your point, run a classic package. Let’s say they want to run a five-year CLV analysis and they use a certain model for it, but he did at the business when we have two years of data. Or maybe they didn’t window the cohorts. So they’re building that model on a cohort from a long time ago. And so that doesn’t represent the customer of today. You can have these problems when they don’t have that subject matter expertise where they’ll deliver something that’s mathematically correct but not business. Correct. And I like to tell the old joke, you know, a data scientist goes for six months and analyze is something around what’s driving our best customers. And they’ll come back to you and say the ones that’s been the most, you know, that happened, that happens a lot.
M. Greenberg: 10:30 That happens a lot. And so to your point, how do you met new ones, here’s how it works, and you have to have, similar to what we had with tools like looker where the subject matter expert could go in and apply filters, you’re now seeing non technical or quasi-technical people interacting with these new CIPs where they can pre-select and set up the problem with their subject matter expertise, and then allow the model to run underneath. So that’s kind of how you close the gap.
Allison Hartsoe: 10:56 So then in what you just said, I always think of looker as a data visualization platform. Are you saying it’s more of a CDP?
M. Greenberg: 11:04 Ah, no. So what I’m saying is I think of it as a classic Bi tool, right? So CDPs also have a lot of actions to your point in your earlier preamble about them, right? So they might be managing your email, they might be managing transactional events. This is, like you said, a visualization or data exploration tool. So in a sense, you have to have something that you know you’re going to look for. You’re looking for a needle in the haystack, these new bi tools, these new CIPs. It’s kind of what the needle finds you. And so that’s kind of what I’m talking about where imagine now again, an ability to go in and ask questions of your data, but also have the data make recommendations to you. That’s really where the CIP sets as opposed to just these dashboarding tools and filtering tools.
Allison Hartsoe: 11:49 Well that’s a really interesting concept. So talk to me more about how the data makes recommendations to you. Is that in the cleverness of the model then?
M. Greenberg: 11:58 Hmm. Right. So that goes back to how they’re set up. And to your point, how do you solve the subject matter experts problem? So whenever we work with a business, I’ll give you kind of an example. The first thing we do upon setting it up is we do a little sanity check, and we, you know, have the model spit out some analyses. Maybe it’ll say, here are the top five things that we find your best customers do and how to predict them. We’ll then have all the data visualized for the in customer, and they can say, well, you know what, I don’t really care about this, this and this, and they can deselect those things, and then run the model again. And that ability really allows them to tune it in a sense to what they want. I always think about, you know, sort of this study, we give a dollar shave club where dollar shave club was trying to move off the razor and blade model and really starts selling a panoply of other products.
M. Greenberg: 12:47 Well, the first thing we had to ask them is what kind of products do you want to look at? And so, by ingesting those products and they were able to pick it, and then they said, you know, here’s the area we want to look at the demographics we want to look at, in a sense that’s what data science is called data munging. After that, the model was able to spit something out, and it actually was able to find about $16 million of revenue because we found that there was this $3 shave butter that for some reason and full disclosure to listeners, I have a pretty healthy beard, so I had to look at this objectively, but uh, they found this shave butter for some reason was extremely correlated with high LTV, and so this allowed them to run an experiment. Now if you looked just at a dashboard, right, it would just say, hey, these many people are buying this product, you wouldn’t know. It’s also attached to other buying behaviors. And so they actually started adding this product to cart. It turns out that it was called them not just correlated, and drive a lot more sin. And we found it was a prediction of about $16 million of untapped revenue just from changing that product mixture and putting that product higher up the funnel.
Allison Hartsoe: 13:48 Wow. That’s a pretty impressive example. Especially when we’re very familiar with the whole dashboard idea, and I oftentimes find it frustrating when the dashboards come in, and they just fall flat. I’d go back to the idea that you use the right tool for the right purpose. Yeah. A dashboard is not an alert, and a dashboard is not an analysis, but yet we try to get dashboards to be the be all and end all of everything. We try to get them to tell us, you know, like reading tea leaves in the data. You know, if I just turn it this way and that way we’ll like get an answer.
M. Greenberg: 14:20 Right, right. You know, I always say dashboards are kind of like the banner ads of marketing in the past, right? There’s a little bit of blindness to them. You start looking at them and eventually they just fade into the background, and I’m sure for, you know, anyone that’s a CIO or a leading up analytics that might be listening. I’m sure you probably sometimes or frustrated with how few people really are looking at the dashboards daily as a way to inform what’s going on. It’s really not human nature, especially because they only may change a small amount of day over day. Again, my co-founder Ahmad, when he was at PayPal, they looked at, oh my goodness, I think he had to look at 50 dashboards for 120 different markets looking at something like 50 different product variants. And so he ended up actually building a product that would look at the dashboards for him and every day he’d wake up, it would write an email saying, here are the things that changed that you should look at. So I think in this, that was almost a Prodo CIP. Uh, but that’s, that’s kind of perfectly explained how the relationship with our data are changing, and how this is allowing really businesses to get smarter. You know, machine learning. In this case, this is not one of those horror stories of it taking jobs. This is one of those stories of it really supercharging your super-powering up whoever the executive might be on the other end of the data.
Allison Hartsoe: 15:32 Yeah. And I want to push a little more on what you said a little bit earlier about CLV and how it’s measured. And Pete Fader at Wharton often talks about there’s no real one CLV model. You really have to kind of tailor it to your business. How does somebody figure that out? Do they just pull the model in and then they start looking at how behavior and customers interact in order to get that holy grail together? Or is there some touristic that they can take?
M. Greenberg: 16:01 Yeah, that’s a great question. So I’ll take you deep inside what’s happening, you know, in the big data science teams, and then we can kind of talk about how I’m seeing this cascade through all the way to smaller e-commerce businesses that might be on Shopify but scaling rapidly. So the big teams, you know, your Uber is your Amazons. What they’re doing is actually, they’re running an ensemble method, so they’re trying everything. I love that you mentioned Peter Fader, by the way. Uh, Dan McCarthy, who was one of his most impressive acolytes, he actually wrote the bike till you die package in R, which really in a sense productionalized, Ah, I know you’re really familiar with it, Allison. Um, it productionalized a lot of their work. He’s actually sitting on our scientific advisory council. And what you can do is you can take a package like that which like you know, those are publicly available, they’re written in our code, and you can deploy it for your business, but let’s say you have a business that is both subscription and ad hoc so someone can buy any cadence they want.
M. Greenberg: 17:00 Well, picking the wrong model, it may not be actually even purpose-built for that. So the bigger teams, what they’ll do is they’ll run five, six or seven models, and more importantly, they’ll make tweaks to it. This goes into something that I think is actually a little bit dangerous that’s going on in the market right now. You mentioned CDPs earlier. People think that, you know, if my product, whatever it is, spits out a CLV. That’s great, I’ve got it. It’s going to work. Well, you better know what that model is, and actually what data is using for it. I’ll give you an example. The big teams, what they’ll do is they’ll exclude certain cohorts. For instance, maybe your early adopters, the business is five years old, and those first two years, that customer looks very different than the customer today. If you’re running a TLC model, it’s going to miss score.
M. Greenberg: 17:48 The new customers might over buy it towards the old ones or maybe your CLV is actually only predicting one more year. So it’s a one-year prediction, but we’re calling it customer lifetime value. That’s not really the whole lifetime of the customer necessarily. Uh, there are so many big decisions that you have to make that those big teams will actually test and they’ll test all variants until they really dial it in. C|Ps now do that automatically. And so what they’ll do is they’ll run an ensemble of all models and could, including some of the most bleeding edge models to make sure you’re getting something that can be 10 20 30% more accurate. Then just doing these sort of out of the box features.
Allison Hartsoe: 18:26 And I remember when the Zodiac guys were around, they used to oftentimes talk about how the precision of a CLV model is about 10% difference. You know? So the larger the company, the more that 10% can be, but I don’t know any company that wouldn’t want an additional of 10% of revenue just because they did the math. Right,
M. Greenberg: 18:45 Right. And you know what? It’s sort of how do you use CLP. So I’ll tell you how it can get really dangerous. One of my favorite ways people are using CLP, especially in e-commerce, is to use it to build a lookalike audience. And if you think about how that was done before, oftentimes what people would do is they take the last three months of data, and they’d say, okay, Facebook, build me a lookalike audience on the customers that we’re spending the most. Well, the problem there is if a customer just came in and only had a few transactions, you may think, oh, that’s not a good customer for me. Uh, I’m not going to build a look-alike on that. Well, with proper CLV models, you can actually find those great customers earlier and find more of them. But if your CLV models a little bit off, even a little bit off, you might mislabel a multitude of customers within your database.
M. Greenberg: 19:34 And then when you try to build lookalikes on it, you’ll actually suppress ads to potentially some of your best customers. So we’ve seen this happen where people who are using CLV, their CAC gets reduced by 10 or 20% when they just tune that CLV and actually use the proper one. Also, they bring in a better cohort finance also for reporting. If you’re off by 10 to 20%, that’s meaningful when you’re thinking about your investors, and you’re thinking about potentially even reporting to the market. So you’re right, that matters. And I think for mom and pops, it’s obviously good enough to just get some of these smaller out-of-box solutions. But as you’re, you know, going past that $10 million revenue mark, it is pretty important to start dialing it in.
Allison Hartsoe: 20:17 Yeah, I completely agree with you. And I’ve also had zero shoes on the show in the past, and one of the things, he had expressed was a lot of frustration with the Facebook model in the way that they had loaded it up and they weren’t able to get more customers back in. And your description reminds me a little bit of a butterfly effect. It creates waves of magnitude out in terms of inaccuracy if you seed it with the wrong information in the first place or if you seed it perhaps in an incorrect manner.
M. Greenberg: 20:47 Yes, and again, you know for everyone listening, I know there will be those who have major data science teams. The other problem here is we want data scientists, and they’re amazing, right? My team is made of them. I come from that background. Ahmad’s from that background. We want them to do everything. CLV itself, you mentioned Peter Fader is a specialist problem, right? There is still a big art to it, so making sure that the data science team is reading up all the current literature or that you are speaking with someone who knows like ambition data, a little hat tip to you. But that really understands what CLV means. That’s critical because again, I see this problem too, not just of CDPs but of data scientist teams where they might be doing great work on personalization, let’s say, but not necessarily dialing in and using the most cutting edge techniques for CLV. So I do think that’s another interesting issue out in the market that this is not a simple thing to calculate, but when you get it right, it’s transformative for the business.
Allison Hartsoe: 21:47 I agree. And you know, the interesting thing here is we haven’t said much about AI, and yet that is in the name of your company. Where does data science and an AI pickup?
M. Greenberg: 21:59 Well, now you’re mentioning the hype train. Uh, so you know, my mother thinks by the way that I’m building sky net.
M. Greenberg: 22:07 That’s what she thinks AI is. You know, I like to use the term machine learning. Uh, whenever I think AI, I think of artificial general intelligence and something that really is learning different things all on its own accord. What we’re really doing here is, and in this sort of the quote on quote AI point, the machine learning models that we’re using for CLV are where we start. So we actually have cutting edge models that sit around it that been tell the business exactly what behaviors are influencing the CLV of the business so that they can make strategic and tactical decisions like product mixture, like triaging customer service tickets. So for instance, I love saying this, you know, when I wait on hold now for 45 or 50 minutes, it might be because the CLV that I have for that business, it may not be because of the order in which my issue came in.
M. Greenberg: 22:55 And so all of these things are where that extra layer of machine learning comes in, we can rank order things like customer service tickets give a dynamic intervention. So the customer service Rep could say, hmm, this person deserves a 20% discount, and this person may be 5%. All of that extra layer is sort of that extra AI component in addition to that core CLV model. And that’s how we think of it not sort of where the AI component is, but again, for those listening, I do think machine learning is a great way to think about this, and you’re going to see a lot of businesses that say, we’re AI, and we’re a black box. I would say challenge them, gently challenge them to say, well, what model are you using? Because most of these models to your point on Peter fader started in academia, and then were improved upon, so if they can’t tell you that, then I would say potentially run for the hills. They should never be telling you it’s a black box automatically doing work.
Allison Hartsoe: 23:52 Yes. In addition to the precision of the CLV calculation. That’s probably one of the key differences between a CDP and a CIP, and ,when we come back to this topic in this definition, I oftentimes think that technologies are really built for a purpose from the beginning, and if you take a tag management tool, and then you pivot it into a CDP, how well is that going to pivot up into a CIP? In a sense, all the technology and all the hiring of staff has been aimed at one particular goal, so it’s almost like you really have to go back to what was the core of how this tool was built. Was this an email tool? Then they’re probably going to do email really well or was this an intelligence tool than the intelligence would be top notch.
M. Greenberg: 24:39 Allison, that’s so true. I always like to say you can’t just say that something like CLV is a feature. It’s way too deep and way too difficult, and so I think you’re right. These were workflow tools or email automation tools that then said, you know what? We probably need some machine learning or some AI on it. And there’s nothing wrong with that. I mean there’s definitely can do some good work, but like I said, when your business starts to grow a bit, you really do want something purpose built that’s building these insights across the entire business, not just for the marketing team though they benefit mightily from it, but also customer success and the finance team as well.
Allison Hartsoe: 25:14 Yeah, exactly. So you’ve shared a couple of examples with us from dollar shave club and the other example about CAC and CLV. Are there any other examples that you want to share?
M. Greenberg: 25:25 Sure. Well, I always like to say, and this is a great example of how it’s working in other areas because CAC and LTV I think are probably pretty familiar for your savvy audience. I think the example in customer success, you know, I think customer success is having their own renaissance the way marketing did, you know, five years ago. And really all of these techniques that were developed in marketing run optimization are now coming over into the CX team. So one of the companies we’re working with, Madison Reed, we’re actually going to be launching a potential customer success trial with them. Again, with this idea that you can take your Zin desk tickets and reorder them with CLV, and then actually dynamically have an intervention. We actually have a slack Bot so that a CSR Rep can actually understand and ask questions in a real time of what they should do for a customer interventions.
M. Greenberg: 26:15 So I always like to say Comcast ventures is actually our lead investor. You know, when you wait four hours, for instance, for a truck to come up to fix your broadband, imagine now a day where, because of the quality of your relationship with the business and your sin, maybe that goes down to one hour. And this is where CLV is now starting to really power personalization. And I think that’s where this is completely going. I hate to use personalization because that term has been wildly bandied about, but your CLV defines your relationship with the business and allows the business to, in a sense, improve its relationship with the customer. So everything like customer success, that’s one way where we’re working with a business to optimize the relationship and really move from net promoter score to popping up CLV based on giving people a better experience. So now they’re not a cost center right now they’re driving revenue as well.
M. Greenberg: 27:07 The other one, and this is the one I’m most excited about, we’re actually pulling out of R and D and starting to productionalize and market a new predictive CLV capability so that you can start to predict it before the customers first transaction. And this is really exciting. So now a customer lands on your website, and you can start to figure out how should I think of them, how much are they most likely going to spend with me? And this is where CIPs again being purpose-built or are able to improve the customer experience prior to transaction. So you could dynamically show a discount or a certain product based on that CLV and the products predicted that they might buy. And that, in turn, gets moved over into the retention team once they convert. And then the CX team. So that’s kind of how this customer journey thread is being tied together with CIP.
Allison Hartsoe: 27:57 And you know, it’s interesting kind of behind your comments. Uh, I read a story this morning about face to face human interaction is starting to become a luxury good. And when you talk about the ranking people by CLV or their level of service by CLV and then adjusting for, do they get the call center or do they get a personal visit or do they get some other element along the way? I think this is a really interesting and somewhat dangerous course that we’re on and that it makes sense to cut the way that you operate by CLV because businesses are not charity organizations. But at the same time, I think what you were just talking about with your new capability of understanding somebody before they are actually showing the transactional value helps prohibit that CLV only view because you’re not just concerned about how many transactions did you do in the past.
Allison Hartsoe: 28:52 You’re also thinking about what’s your propensity or what’s your future capability. And so if I can tell that you look like a CLV customer, a high CLV customer, and yet you haven’t made a purchase from me, that’s a really nice way to welcome someone in who could be a very valuable customer to you. It’s almost like the person who pulls up with the Rolls Royce and you know, you can see that coming in, and yet half the time when you walk into a store, you kind of want to say, Hey, don’t you know that I love your brand and I haven’t bought it in a long time and in here I am again. And these things can be really helpful for the way an organization can not just operate by CLV but can operate in a more sensitive, more human fashion.
M. Greenberg: 29:36 Oh, it’s so true. And by the way, I love the Rolls Royce reference only because there’s not old adage job you’re at a car dealership and see car salesman judges you immediately by the car you drive in, hop in in, and then what you’re wearing. I would almost argue that with these machine learning techniques, you know, oftentimes it’s the person that comes in right with the Hoodie, and maybe the Subaru is the one that has potentially more spend or more brand affinity. And so this also allows you to innocent find those hidden customers that maybe didn’t look like what you thought, but because of these deep patterns in the data, you can also find that customers didn’t come in in the Rolls Royce, but you should be giving that great warm welcome to. And so that to me is, is the power of this. Oftentimes you’re going to get a lot of counterintuitive output, and that’s because there are so many deep patterns beyond human scale in this data that you’re finding.
M. Greenberg: 30:26 So I do think that in the very near future, this will be table stakes. All businesses, when mapping the customer journey, they’re going to have to map it not only with CLV, but their interventions will be formed around it. Because you’re going to know kind of how much you’re losing in the future if you don’t capture this customer or if you don’t solve their customer service problems. So this puts a real number on how much you’re losing based on what they would be predicted to spend, and I think that allows you to start applying dollars in a more intelligent way.
Allison Hartsoe: 31:00 Yes, exactly. We’re clearly on the same page here. I was struck by an example someone shared with me recently about a coffee company and said that this coffee company knew that every customer was worth a CLV a $14,000 and change. And I was like, wow, because you think coffee, oh, it’s three to $5 something like that. But over the time if I’m losing that person, I’m not losing three to $5 I’m losing $14,000, and depending on what type of product you’re selling, that’s really how everyone should be thinking is the future opportunity cost of that person, and the 10 people they tell when they’re not happy with your business.
M. Greenberg: 31:36 Absolutely. I mean I think that’s probably the most exciting moment tea ha moment I love in my job is when we go in our speaking with our CMO and a CFO, and they’re in the room and we show them a notable map of their entire customer journey and we show going from one note to the next, here’s how much their CLV is increasing or decreasing, and immediately they go, wait a minute, you’re telling me if I make this one action, if I spend x or I try to win this customer back to where they switch to a different product, it’s worth x amount of dollars. It’s kind of this eyeopening moment because everything was instincts before that. They know that product’s worth more than that, but they didn’t put a number on it so they can say, oh, we’re willing to spend a hundred or $150 on that intervention because we know it’s worth 200 if we keep them on this one particular pathway. I think that’s a huge game changer. What will happen next will be automations around that though, and I think that’s what you’re going to start the CCIP to do is automating in a sense parts of those customer journeys or informing those CDPs. Oftentimes we work with a lot of businesses that use CDPs. They’ll use our segments, they’ll use our CLV scores and then plug it into, like you said, that email marketing automation or that acquisition automation.
Allison Hartsoe: 32:49 That makes a lot of sense. So let’s say I’m convinced and I love this idea, and I want to have more intelligence in my business. What should I do first?
M. Greenberg: 32:58 Well, I’m not going to say go to the retina AI website, but what I will say, because we do have a marketer checklist on this, but I would say the first, make sure your data’s in the right place. Is your data prepared to answer these questions? So the first thing I like to say is had you set up your database or your data lake so that you’re capturing at the customer ID level, all of these attributes, the transactional history, the correct timestamp, even better, are you getting clickstream data next to it? Are you getting certain behavioral data next to it? So that’s the first part. Make sure you set that up correctly.
Allison Hartsoe: 33:33 I can’t emphasize how important that is because it’s not just, Oh we have some customer information in all of our order data buried underneath. No, it’s the customer record, and it’s just a totally different way of structuring your data that older companies oftentimes don’t have in place.
M. Greenberg: 33:53 Right, and honestly again, going back to old attitudes and tales about data scientists, even though really this has been a field we’ve only been called data scientist for five to six years, one of the things they talk about is about 90% of the data scientist’s job unfortunately is cleaning the data. And so the more that your data is clean and organized at that customer Id level so that you can quickly pull it the factor, you can start answering these questions. So that’s part one. Part two, you mentioned it earlier, do you have a customer identity resolution strategy? So have you put in some techniques, attract touchpoints and know that one customer when they came back with the same customer or that you’ve reconciled all that purchase history with their customer service interactions, all of that is really critical as well. And then finally, I think this is a big point.
M. Greenberg: 34:42 Are you able to then build machine learning capabilities on top of it? So not to get too technical, but your data might be sitting, let’s say an Amazon S three buckets, very cheap storage, but then when you want to start asking questions of it for machine learning standpoint, you sometimes need to transform that data. It’s people use things like spark clusters. So if you are doing your own data science internally or you want to, making sure you’d have that orchestration layer where you’re spinning up these nodes because these models, oftentimes they’d run in parallel. They’re pretty heavy computationally. And so you want to make sure your data is in that state where you can start using machine learning, or work with someone who can take that data out very clean way, in a safe way, uh, usually in an anonymized way and run that analysis and push it back into your database. I always say, is it extensible? It’s as simple as this. If I add one more column to my database because have a new product or tracking something new, am I going to break the rest of the database? That’s what we call extensibility. So is it ready to grow with your business?
Allison Hartsoe: 35:45 And that’s so important with event data, and clickstream data and other data around the customer is, it’s very, very wide and you have to be able to handle more and more. But I like what you said before about the storage element. It’s almost like the S three bucket is the refrigerator, but you’re not asking your refrigerator to be a stove. You know what cooks the data and creates the meal is not the same as where you store it.
M. Greenberg: 36:10 Right, right. And again, you know, I’ll, I’ll give it a little hat tip to a company out there called data brick. You know, there are a lot of businesses now that use Amazon web services as cold storage like you said. And then that hot compute layer, that layer that where you have to organize the data in a parallel format. There are businesses like data bricks that helped you automate that orchestration the way that Amazon web services does for your cloud storage. So can these companies are coming to market. We kind of sit in that layer right above it, giving you those insights just to kind of conceptualize us in our relation to the data.
Allison Hartsoe: 36:43 Very nice. Very nice. Now if people want to get in touch with you, Michael, how can they reach out?
M. Greenberg: 36:47 Oh, well I would love to talk to anyone and, and answer all your questions personally. Um, my email is Michael @ retina retina in your eye, retina.ai and uh, they can also just go to our website, retina.ai and again, I’d be happy to personally answer any of your questions.
Allison Hartsoe: 37:04 Excellent. Michael, it’s always a pleasure to talk to you. I just, I love the way that your mind works and how our conversation just ranges over the depths of the AI space with like a nice nods to autos and refrigerators and any other number of topics.
M. Greenberg: 37:21 Well, I’ll say this is a true pleasure. It’s always wonderful to talk to you, and I look forward to our next chat.
Allison Hartsoe: 37:26 Thank you. As always, everyone links to things that we discuss are at ambition data.com/podcast remember when you use your data effectively, you can build customer equity. It’s not magic. It’s just a very specific journey that you can follow to get results.
Allison Hartsoe: 37:44 Thank you for joining today’s show. This is your host Allison Hartsoe and I have two gifts for you. First, I’ve written a guide for the customer centric Cmo, which contains some of the best ideas from this podcast and you can receive it right now. Simply text, ambition data, one word, two three one 909 six and after you get that white paper, you’ll have the option for the second gift, which is to receive the signal once a month. I put together a list of three to five things I’ve seen that represent customer equity signal, not noise, and believe me, there’s a lot of noise out there. Things I include could be smart tools. I’ve run across articles. I’ve shared cool statistics or people and companies I think are making amazing progress as they build customer equity. I hope you enjoy the CMO guide and the signal. See you next week on the customer equity accelerator.