This episode of The Analytics Edge is brought to you by NetSpring and showcases John Humphrey, former Head of Data Platform Product and Chief Data Officer at Intuit MailChimp. John joins us to discuss data strategies for Customer Analytics, from continuing investments in Customer 360 to the emergence of warehouse-native data platforms. While enterprises increase their investments in cloud data warehouses for a single source of truth, he foretells of a future with AI-powered Autonomous Customer Analytics that will unlock the possibilities of truly 1:1 personalization.
This episode of The Analytics Edge is brought to you by NetSpring and showcases John Humphrey, former Head of Data Platform Product and Chief Data Officer at Intuit MailChimp. John joins us to discuss data strategies for Customer Analytics, from continuing investments in Customer 360 to the emergence of warehouse-native data platforms.
He delves into the technology gaps that forced a detour from our earlier vision of Customer Analytics. As products became more digital, the data platforms at the time were unable to handle all the event data being produced – leading to the emergence of specialized first-generation product analytics platforms, data silos, and fragmented analytics platforms for product analytics.
John introduces 4 levels of Customer Analytics maturity to help data leaders rationalize earlier investments, and how the cloud data warehouse is enabling warehouse-native strategies and applications to break down those earlier silos, and finally now, deliver on the promise of C360 and Customer Analytics.
John Humphrey is the former Head of Data Platform Product and Chief Data Officer at Intuit MailChimp, with over 2 decades of experience and expertise in data science and data engineering. He was the first data analyst at Goodreads (later acquired by Amazon), helped take LegalZoom public, and has had multiple stints in data leadership roles at Meta and Intuit along the way. John earned a masters in Systems Engineering from the University of Virginia and holds a bachelors in Management Science from Virginia Tech.
Key Quote:
“TBD”- John Humphrey
(Segment 1)
(1:16) Earlier roles in product analytics
(3:30) Shortcomings of 1st-gen product analytics
(9:08) Shortcomings of BI for product analytics
(11:26) Funnels in BI vs. product analytics
(13:04) Customer 360 and the data warehouse today
(15:23) Data streams for complete C360
(17:42) Customer 360 versus a CDP
(20:48) Aligning C360 & CDP strategies with the warehouse
(Segment 2)
(23:53) Benefits of cloud data warehouses
(26:11) Analytics tools for all types of customer data.
(29:38) Maturity model for Customer Analytics
(Segment 3)
(34:02) AI-powered Customer Analytics
(Segment 4)
(37:50) Takeaways
Links
Announcer: [00:00:00] Hello, and welcome to the Analytics Edge, sponsored by NetSpring.
Tom: The Analytics Edge is a podcast about real world stories of innovation. We're here to explore how data driven insights can help you make better business decisions. I'm your host, Thomas Dong, VP of Marketing at NetSpring. And today we have a very special guest, John Humphrey, former head of data platform at Intuit MailChimp, joining us to discuss data strategies for customer analytics.
John, I'm delighted to have you on the show today. Welcome.
John: Thanks for having me, Tom. Super excited to be here and looking forward to the conversation today.
Tom: John, you've blogged and talked a lot about how most organizations you've advised or worked with over the years are all still somewhere in the process of building a customer 360 in their data warehouse with the hopes, of course, of maintaining and working off a single source of truth.
But along the way, as products were becoming increasingly digital, the data platforms that were available at the time couldn't handle all that event data being produced. This led to a fragmentation in the data [00:01:00] stack and point solutions such as product analytics emerged. Unfortunately, those point solutions now hoard large but limited portions of customer data in their own respective silos.
Of course, at the time, they served a critical need and you became one of the biggest champions of product analytics. So can you tell us about the various data roles you've held over the years and how you came to become one of the biggest champions of what we call first generation product analytics?
John: So I've been doing something data related over the course of my career long enough to comfortably predate terms such as data scientist or data engineer, but in practice, I've kind of been a little bit of all of these things over the years.
And, you know, I think what I've kind of seen over the arc of my career is kind of two chapters. Call it chapter one, kind of goes up to about, you know, 2010 ish. And that was really before you saw much in the way of the advent of digital products. most of the products and experiences we delivered to consumers were very much the analog world.
And, you know, yes, some data was brought to the data warehouse. But ultimately, everything we did with the customer, we did sort of offline, if [00:02:00] you will. And then sort of the second chapter is really as we started getting more and more of these digital products, you know, all the sorts of different opportunities and challenges that came with.
And so as I think about my career, I've done everything from Literally being the first data hire at a, you know, series A startup that would go on to later get acquired by Amazon to being the data leader for an organization that went, you know, public. I've worked at some very large enterprises, you know, into it, such as you mentioned, you know, Meta as well.
and, you know, really kind of what made me excited about a lot of these sort of, you know, product analytics point solutions and other such things that you were mentioning a moment ago. is that as we started building these like digital native experiences, a lot of what we wanted to know about the customer was knowable from these sort of streams of events that we have that, you know, took place in the product.
And so, You know, our traditional sort of SQL and BI paradigms just really weren't equipped to handle these things. And so along [00:03:00] came solutions that could actually handle these sorts of data. and so, you know, needless to say, being able to get at such a material part of a story, and, you know, unlike with our sort of analog products, being able to see every little minute detail about how someone uses a product.
Was actually like super, super exciting. And so that's kind of, you know, what sort of drew leaders such as myself to these solutions as a hub. Now we can really start to see in great detail what our customers do and get to know them better.
Tom: Alright, so when did you start realizing the shortcomings of these tools?
John: I've done multiple sort of implementations of these tools. So for example, you know, on the product analytics space, you know, I've implemented amplitude probably at least three or four times over the course of my career. And yes, like, you know, we were sort of talking about a moment ago These solutions are great at fulfilling the promise of I hand you a stream of event data coming from a product and it handles them quite well.
It handles them in ways that our traditional sort of SQL and BI paradigm based tools simply could not handle those things, [00:04:00] especially at the time. The problem that started coming in as you started realizing, you get further and further into these implementations and all the sorts of potential that you expected to realize by implementing such a solution just never quite materializes.
And you start this, you start to look at the patterns as to why this is happening. And what you realize is that number one, the big big thing is you just don't have a complete view. yes, you know, what goes on in the product is a lot of my relationship with the customer, but it's not the entirety of my relationship with the customer.
And so, really all, especially like a product analytics solution knows, is whatever it is that I instrument through front end events in my product. and that puts quite the burden on the teams responsible for instrumentation. Like, if they don't instrument it, I don't know it. But the reality is there's a lot of things I know about a customer, either because it happens outside of my product, you know, such as the conversations that they might have with a customer care agent, [00:05:00] what I might do with the customer from a marketing perspective.
And then there's also times where, yes, I can instrument something as a front end event in a product, but like, that's work. I could have also just come to know that through, say, the back end data that's getting sort of captured by the product as well. And so just as I kept going through these implementations and especially also noticing that like the bigger and more complex the business or the bigger and more complex the customer relationship, the more that these holes in the sort of big picture started to emerge, the more I started to realize that like we were rapidly heading towards a local optimum with respect to I could better and better understand things such as feature usage that exists squarely within the purview of a given domain.
And again, we keep sort of talking about product analytics, but you know, Marketing analytics has similar sort of, you know, silos as well. But the second I want to answer a question about the totality of the customer, and I usually get to that point pretty quickly, the consistent inability to truly answer those [00:06:00] questions time and time again.
was what sort of made me realize that, again, we might be stuck at some sort of local optimal here.
Tom: Yeah, actually, would you maybe have an example here, you know, across these data silos, instrumentation, you talked about the idea of enriching, with other product data, and then you have different platforms looking at different data.
Do you have an example of maybe some metric, that wasn't matching up between your different platforms?
John: Yeah, let's, let's, let's talk about an example that's near and dear to my heart. It's this picky little obscure metric that no one talks about called retention. And so you would think that, hey, if I hand you a product analytics platform, that I would be excited to hear that my PMs are going and looking at retention rate as a function of all manner of different things, in that platform.
But actually, there were a few words that struck terror into the... The hearts of my analysts and myself tend to hear that someone was using our product analytics platform to look at retention rate. Why, you might ask? Because for the case of the product we're working with [00:07:00] here, only about half of the cancellations happen inside of the product.
The other half of the cancellations happen because somebody picked up a phone and called and said, you know, we're done with this product and we want to cancel. And all of those cancellations Happened completely outside of the product experience. No events were ever created. They never flowed into the, the silo, if you will.
And as such, like our retention curves were materially misstated. And that's not to say that it's impossible to pipe those data in, but in practice, like it's a non-trivial amount of work. And the reality is. You know, teams implement their sort of, you know, specific bucket of events that they're supposed to instrument.
They check the box, they move on, and next thing you know, you have these tools that you're a little afraid of people using because they have these material gaps. And, you know, this is hardly like some, you know, inconsequential metric. I mean, retention is pretty existential to a subscription SaaS business.
And so, you know, when you have to start telling people, I know you want to use this, I want you to gather insight from this as well. But don't look at [00:08:00] retention rate. that immediately starts to sort of undermine the credibility, not just of the tool, but really kind of like the investment in this sort of tool and like what you're doing as a whole.
Tom: That sounds like a real, real problem across tools, across teams. Like who, who do you trust? Like what can you, which, which system, really becomes your, your system of record here? and it sounds like a real challenge. you know, obviously. Product analytics served a very useful need. It offered self service to the product teams to get very quick, insights into product usage.
but at the same time, you as a data leader manage a team of analysts who are using SQL and BI tools already. On top of the data warehouse to answer those deeper, broader customer journey, and experience questions. So kind of the immediate thought as well, I already have invested in a BI tool. Why don't we just use a BI tool exclusively for product analytics?
You mentioned that your team's doing that, but Let's deep dive [00:09:00] into this a little bit. What are the limitations of BI being the one, the go to place for product analytic workloads?
John: That's a great question and it kind of comes back to the point of why was anyone like myself excited when these other tools came along given that BI tools and SQL existed long before any of these things.
And the short answer is that none of these tools in the BI and SQL paradigm handle time series events particularly well. What they want to do at their core is collapse these things down to aggregates as quickly as possible. you know, getting to a sum or an average or a max. Like, that's where they want to go.
They don't deal particularly elegantly with time series. And so once you sort of have that challenge, that's fine. I suppose if you already know what it is you want to get from the data, you already know that, okay, the thing I want is, you know, sort of x over y. It's also fine, honestly, if you have that sort of understanding imposed upon you because you have some vanity metric like MAU or DAO that [00:10:00] you have to report just because.
But if you don't know what you're looking for going in, this is a really hard way to make a living in terms of trying to come to some understanding or gain some insight about what the customer is doing in your product. And so, that's really where kind of the SQL and BI sort of paradigm falls down. And in particular, especially if you look until probably about like the late 2010s, your SQL platforms, you know, the databases themselves, really struggled to deal with time series data particularly well.
And once they sort of tapped out, then all the tools that exist sort of downstream of that aren't going to be particularly adept at doing these things as well. Anyone who's ever tried to mimic the sort of flow or Sankey diagrams that you see in many of these product analytics tools in a tool like Tableau knows how painful tools downstream of even something like a database are.
And so, you know, that's been kind of the challenge here is just that. Again, when you don't know what [00:11:00] you're looking for, you need to explore. Visual is kind of how people like to explore things. And, you know, the tools we've had historically just, you know, were completely antithetical to that because, again, they just want to collapse everything to a single row as quickly as possible.
Tom: But the reality is you can create a funnel in Tableau, but what's the difference between a funnel in a product like Amplitude versus a funnel that you're creating in Tableau?
John: Well, the first big difference is the hours I spend creating the funnel in Tableau versus like the minutes I will spend creating it in Amplitude.
And sure, if I want a funnel that's just like a series of sort of like counts in rows, like almost in a like cross tab format. Yeah, I can do that pretty quickly in a tool like Tableau. But, you know, let's be honest, like, we're visual creatures, like, we want to see flow diagrams, we want to see something start wide and get narrow as we move forward.
and building that kind of stuff is hard in Tableau, it takes time. It's also not terribly flexible, which then means when you want to build the [00:12:00] next one, and the next one, and the next one, like, those all take time. And there's nothing that sort of dampens one's intellectual curiosity faster than it takes a long time to answer the next question.
Tweedy said I can answer questions quickly, I can maintain that curiosity, and that curiosity is what carries me through to some novel insight.
Tom: Exactly. Yeah. Analytics at the speed of thought. So, obviously, there's been tremendous fragmentation here. You experienced that at Intuit, where you had Tableau and Amplitude in place.
Now, a big reason for this fragmentation was kind of the technology hadn't caught up yet, to the needs of the business, and we kind of found ourselves or are finding ourselves now trying to stitch together all of this customer data to get a complete view of the customer. Which is something that we had talked about, you know, 15, 20 years ago when everybody was talking about Customer 360, right?
And, you know, there were earlier versions of data warehouses at the time. Cloud data warehouses have obviously emerged. in the last few [00:13:00] years. So, as far as Customer 360 goes, where are we now?
John: I think it's sort of the longitudinal view of Customer 360 that you were alluding to a moment ago, I think is really important.
Because in some ways, over the last decade ish, we actually regressed. In terms of our movement towards that end state, and the reason that we regressed was because of the introduction of a lot of these sort of data sources that are essentially event streams of time series data, you know, back before we could track every click in a product, we just didn't have as many of these.
And so our data warehouses were very, like, conducive to producing this sort of customer 360 view. We then introduced this whole new data source, which. In many ways is a blessing in terms of like all the like insight we have about how our customers interact with our products. It just breaks the sort of prevailing paradigm within our data warehouse, which then makes the 360 view kind of start to fall apart a little bit.
Where we are now is database platforms such as, you know, Snowflake, Databricks, BigQuery. [00:14:00] All are getting pretty good at dealing with event data. And so now it is reasonable to have all of this rich event data that tells me how my customers are interacting with my products in the same place that I have everything else.
And so now, like, kind of at last, I can start to bring together a full picture in the data warehouse and really get back to this promise of sort of a customer 360 view. The key thing is, I have to make sure that all of this other event data actually makes it to the warehouse, that it doesn't just get stuck in these sort of silos that many of our first generation solutions, whether being product analytics, marketing or other places, have built, you know, back at the time out of necessity.
But now that necessity is gone and, you know, getting those data into the warehouse is essentially what allows us to sort of build this 360 degree view and start to, you know, move forward, in many ways, like a more advanced version of what we had, like, a decade ago. Mm hmm.
Tom: So you're talking about obviously, the big [00:15:00] cloud data warehouse is now supporting streaming data. You mentioned product data is one source of event data. As a data leader responsible for customer 360, what are other examples of streaming data that you want to synthesize to build up that view of the entire customer journey?
John: I would say that the big one, especially even as I think about this from like a product perspective, is really anything from a customer contact point of view.
So, the dialogues going on with my call center agents, whether that be over the phone, whether that be via chat, you know, whatever sort of medium they're using, like, in some ways, Those interactions become the non code backed portion of my product experience. It's where my product essentially, the code based version, didn't deliver everything that the customer needed, and so now I must talk to a human to get the rest of the way.
And I need that picture because if I'm really going to understand how my customer interacts with me, like, I can't just, like, ignore this, everything that happens once they get out of the product. [00:16:00] And so that's kind of a big one for me. I think as, as I'm looking forward and I'm thinking about, like, where might You know, some of these new patterns that generative AI is introducing.
Take us. I have a strong suspicion that as conversational UIs become more and more prevalent of a pattern, Like, all of that, it stands to potentially break our very sort of click based, you know, product analytics paradigms, and so being able to introduce those streams as well is super important. But I'd also say just other sort of kind of very core streams, like, you know, any sort of transactional streams that I might have, any sort of, you know, even, it's a very kind of slow and sort of simple stream in a way, but the stream of things that I'm doing from a marketing perspective, like, All the different places where I spend or, you know, have outbound contacts or whatever.
Like all of that, what I really ultimately want to end up with is this kind of single longitudinal view of the customer and all my seller touchpoints with them, whether it be in the product, whether it be through my marketing [00:17:00] partners, whether it be through, you know, sort of customer care or sales. Like really trying to truly understand my relationship with my customer, who they are, what they need is what happens at the intersection of all these things, not just each one kind of sort of in isolation, running sort of parallel to one another.
Tom: Yeah, it's interesting as you talked about, you know, all this click based interactions, instrumentation. I think a lot of our viewers are going to be thinking CDP. I've already invested in these tag management systems and we're talking about customer 360. So maybe it'd be useful at this point if you could share what you view as actually the differences between a Customer 360 strategy and a CDP.
John: I think this is a great question. It's something that probably the last year or so I've spent a lot of time thinking about. And I think the short answer is More and more, they really shouldn't be different. what it really should be about is how do I have a singular, holistic understanding of my customer that I then use to activate that customer and many others like it at scale?[00:18:00]
And that, in many ways, should be the same set of questions and require the same set of data as, say, my analytics use cases. So, for example, if I believe that a key inflection point is when a customer adds their ten... Sort of prospect to my product that a thing should happen that isn't just a significant moment from a, like, lifecycle marketing perspective.
Analytically, that should be a significant moment as well, and I should want to track how often I'm driving people towards that. And all of this should be originating from the same data calculated the same way. Now, the question then becomes, well, why aren't we already there? And here again, the answer is very similar to sort of the discussion we were having a few minutes ago with respect to product analytics, which was that back when these sort of CDP type platforms were coming into existence, The technology, frankly, wasn't good enough to support the use, to support all the use cases on the data warehouse side.
So if you think about for all the important aggregations and things [00:19:00] like that, that you do and put into a CDP, the most important currency at the base of all this is events and streams of them. The second somebody like signs up, you want to send an email. And we just were discussing how data warehouses, especially back in the day, were not particularly good at dealing with events.
And especially back then were very slow, high latency, that kind of thing. Which is all of the sorts of things you simply cannot tolerate in the CDP use case. Well, now we've gotten to a point where the technology underneath the data warehouse is strong enough to support the, like, latency requirements and whatnot that exist on the CDP use case.
And otherwise, you know, like we were talking about, you have the same sort of need for, like, the same data, you know, at the, you know, cut in the same ways. And so, yeah, you really should be trying to bring these things together. One, so that you have a sort of unified view and a single source of truth. Two, so that as you gain these insights about your customer and what makes them tick and how to grow them, it's easier [00:20:00] to operationalize those insights immediately.
Like, you can just put it right in your CDP that probably should be based right off of your data warehouse. And then you can go activate immediately on the same understanding that you're also reporting upon as you look at your sort of growth of the business, you know, sort of day in and day out.
Tom: Yeah, it's interesting you talk about these things as they should be one and the same thing but if we think about it, Customer 360 in a data warehouse is typically the realm of the data engineering. the data side of the house and CDPs have traditionally been sold into the CMO and the marketing organizations. So can you maybe provide a view into how you would advise bringing those two teams together with the underlying technologies also coming together at the same time?
John: Yes, and I think the key is that the reason these things were segregated previously is that the technologies were completely different. you know, the sort of data team, the data warehouse, data engineering folks [00:21:00] are all ultimately, you know, building pipelines in SQL that have probably some, you know, Python or some other sort of wrapper for automation, but that's sort of their currency and that's how they, like, do work.
Versus, you know, these, like, off the shelf solutions that your CMO is buying don't necessarily follow that pattern. They're almost more IT ish in nature, a little bit more black boxy. And again, that was necessary because the only viable solutions follow that pattern. Well, now we've reached a point where there are all sorts of different viable solutions that actually follow more of the data engineering and data warehouse type pattern.
And so now there's no longer this necessity to keep these worlds separate. And so in terms of, so from a technology perspective, that part of the deck has been cleared. These teams can work together. So then it really simply becomes a question of how do they collaborate. And in many ways, that's really not that different.
Then this question of how, like, an [00:22:00] analytics or, you know, data engineering team collaborates with marketing today in terms of some of their, like, reporting and analytics use cases. You just have to sort of follow a lot of those same patterns. And so that, I think, is really kind of the place we've arrived at.
It was like, the only reason these things were in separate worlds was, like, the technology you needed for one set of use cases was not the same that you needed for the other. Now that you've sort of eliminated that barrier, there's really nothing stopping these two worlds from working together, and in fact...
Quicker and more intentionally, you draw these worlds together, the easier other things become, such as, you know, like the sort of reporting and measurement use cases, too.
Tom: Right. So, it sounds like it's actually the cloud data warehouse is the technology that has caught up that is actually enabling these two worlds to come together.
John: Exactly. That is really, you know, if I were to kind of hammer home one point in particular to People listening to this, the thing that has changed over the last, call it, five ish [00:23:00] years is that finally the data warehouse technologies have caught up to the, the, both the velocity and the volume of data that we're dealing with.
And a lot of the problems that we struggled with that forced us to make these compromises and go for these siloed solutions. Those limitations no longer exist. Those compromises are no longer necessary. I love it.
Tom: So let's switch gears a little bit. Let's, let's, let's make the assumption that the cloud data warehouse and the modern data stack is the path forward for any data leader.
And you've mentioned now how the technology and support for things like streaming data has really made, you know, customer 360 and data centralization possible in the cloud data warehouse. What are some other benefits you see with the cloud data warehouse?
John: Yeah, I think there's a number of benefits. And many of them fall broadly in the bucket of keep the boring stuff boring.
and with no disrespect intended towards my CIO counterparts. but much of this, you know, what I have to think [00:24:00] about from a security and compliance perspective. Now I only have to think about that once as opposed to however, you know, as opposed to n times where n is however many silos I have. I can just think about that in the context of my data warehouse and I'm done.
And so there's all sorts of risks and, and sort of problems that I get to sort of, you know, greatly reduce the amount of time I spend thinking about those things because of this. another thing I would point out is that, you know, duplication of data is a real problem. And in some ways, like, the storage cost associated with duplication is, like, the least of my problems.
Like, at the end of the day, the S3 buckets are cheap. What's expensive is the analyst having to explain why the contents of two S3 buckets that should be the same are not, in fact, the same. Why revenue over here and revenue over there are not the same idea. and so, again, to the extent that I can simplify things down and get as close as possible to one version of the truth for each and every truth.
you know, the more I can drive, you know, sort of complexity and risk out of the equation. And these [00:25:00] aren't just analytical questions either, you know, the fact that I have multiple measures of revenue doesn't just mean that, like, I mess up some reporting. It can also mean that I mess up a campaign.
Because if I'm saying, hey, send it to everyone who spent over 500 in the last six months, but I can't agree on what it means to have spent 500 from one data source to the next. Now I'm, you know, maybe not emailing some customers I should or emailing some customers I shouldn't. And all these things are just, you know, the more versions of, you know, your sort of ground truth that you have floating around, the more you introduce these risks that you have to sort of mitigate and then hope for the best.
Tom: Yeah, he had talked earlier, actually, I think in one of the earlier lines of questioning about kind of bringing together multiple streams and really looking at outer product events being one of the key limitations of your first generation product. Alex tools. How do you see the, the cloud data warehouse as kind of the single repository, bringing everything together?
Do we actually have the tools to analyze all of this information, right? Be it [00:26:00] streaming data or all this customer reference data that is typically the world of OLAP.
John: Yeah, I think that's a great question. And I can share with you sort of the mental model that I use. And I think of this in like sort of differing levels of maturity.
You know, there's a sort of base level of maturity, if you will. Which is really kind of these like siloed solutions, whether it be for product analytics, whether it be for understanding where my sort of marketing spend is going, whether it maybe even be the sort of reporting and data behind like my call center and telephony sort of platforms.
what I get with each of these platforms is I have all the data related to that particular domain of the business. And these tools tend to be at least somewhat opinionated analytically about what I need to measure, what metrics I need to have, what reporting I get, etc. And so that's kind of the first level of maturity.
People start to outgrow that pretty quickly because they inevitably want to answer questions that sort of cut across the grain. they have parts of their [00:27:00] experience that are not necessarily getting fed into a specific silo. Oftentimes, for example, backend data from a product never really sort of makes it into a silo.
and, and so, and then there's also the extent to which If I want to measure something differently than my product analytics platform wants to measure it, like I'm kind of, you know, on my own in many cases. And so then that sort of gives way to this kind of second level of maturity that you see a lot of people go to, which is a more sort of data warehouse centric view of the world, which is basically dump all the data into your favorite, you know, cloud database, have a bunch of SQL and reporting layers that sit atop it, and away you go.
And that sort of solves these problems of bringing data from different places. It also gives me a greater degree of customization. With it comes a ton of work, and also, it still doesn't entirely address the problem of, in some cases, I've got data going into these silos, and I don't really have a reasonable way of getting it out.
which then sort of brings to a third level of maturity and I think one that it's like finally [00:28:00] opening up in the last couple of years. as you know, we have improvement both on the sort of database technology side and also as, you know, tools such as, you know, NetSpring, for example, come into existence.
Where now I just sort of make my warehouse source of truth. I write everything native to the warehouse. And then I build up from there. And I think where, you know, existing tools are good is like, you know, you can, you can still do all the same sort of SQL and ETL magic you could do before. With all these, you know, sort of new data sources that you have coming, like, that's sort of like standing, you know, at the, you know, kind of, you know, at the sort of bottom of the hill, you know, watching the avalanche come down on you and thinking to yourself, what am I going to do now?
And that's where I think, you know, for example, again, you know, tools like NetSpring are super, super exciting to me because, you know, you can start to, Give more and more capability to the end user to assemble these things for themselves and, you know, have a better chance of withstanding that avalanche of information and in doing so, unlocking the potential of having that sort of holistic view of [00:29:00] the customer, you know, much like the one we used to have back in the day before, you know, all this sort of, you know, exciting event data from products and whatnot became a thing that we had to deal with.
Tom: Okay. Let me recap, actually, this framework. I think it's very useful, to kind of visualize this in our minds as data leaders here. So level one, was siloed analytics with all these tools that, you know, shadow IT could go off and purchase, like Google Analytics and Amplitude. And whatnot, and be able to track all of my campaigns and product usage, right?
And then the level two sounds like, beginning to move to the data warehouse and having this concept of a warehouse native customer 360 data strategy, where you have all of your data warehouse and then instrumentation tools and your BI tools, sitting on top of it alongside all of these silent analytics tools.
And we're now at a level three maturity for some companies where this movement to build on top of the data warehouse, which is happening a lot [00:30:00] internally at many companies already today, but a new wave of SaaS applications or warehouse native apps has emerged. So warehouse native customer analytics, for example, would be something that we at NetSpring offer, where we compute directly against the data warehouse, looking across all of your.
product and customer data. in your mind, you know, this is where, this is where we're, we're reaching towards, right? Not, not a lot of companies are, are there yet. but for our listeners, what do you see as the unique use cases that are unlocked by level three?
John: Yeah, I think this is in many ways a set of use cases that we already have in our, in our near and dear to our hearts, but we've kind of conceded that we just don't have a good way of dealing with right now.
So if you go back to my earlier example of, like, I don't trust the retention rates that come out of my product analytics solution. Like, that's exactly the sort of use case that now becomes rather trivial to deal with without a ton of, you know, pipelining work [00:31:00] and things of that nature. So stuff like that just becomes table stakes.
But I think some of the other use cases were maybe things we were alluding to a little bit earlier in the conversation, such as I want to understand when a customer Like, when I'm getting an inbound call from a customer, why are they talking, why are they wanting to talk to me? What is it that just happened in the product that might be informative as to the conversation that we're about to have?
and how might I even be able to see patterns of this in the future and be able to mitigate the issue before it ever has to escalate to a phone call? All of that stuff becomes way more approachable. Another set of use cases really become around sort of tying what I did at the marketing stage with what I'm doing in product.
So, for example, if I know something about the ad placement that drove you to sign up, if I know that, for example, it was really talking very heavily, say, about, like, preparing my taxes, then I know where in product to lead you. And I [00:32:00] know kind of what sort of subsequent activations I might, I might want to engage you with.
I might also know how I want to personalize the product. So again, it really just, it really all kind of comes down to these sort of cross cutting use cases where I have to appreciate who the customer is in terms of how I acquired them, what they're trying to do in the product, what they're trying to do outside of the product, and anything I might know about them even beyond their kind of day to day engagement with, with my products and services.
Any use case that sort of relies on data and it sort of spans across those things is exactly the sort of thing that gets unlocked with these sort of, you know, solutions that we're talking about and getting to the sort of third level of maturity. Mm hmm. Mm hmm.
Tom: Yeah, so it actually sounds like you talked about two different types of use cases here that we can unlock.
One is around the analytics. He gave the example around retention, for example, there. And he also touched a little bit on activation, right? How do I personalize that experience and activate directly on the data warehouse? [00:33:00] and so obviously near and dear to everybody's heart right now is this concept of AI.
And I think kind of what ties activation and analytics together is that layer of automation, right? A lot of it today is, you know, involves a human in a loop, right? Somebody still has to approve the insights coming out of the reports and then define the actual workflow as a result. But AI is already in place in many industries for like supply chain optimization to so why don't we end our interview here with a bit of a forward looking, view into the world, you know, when, when AI, and you talked about Gen A a little bit earlier, how does it relate to customer analytics and potentially autonomous customer analytics?
John: Great question. And so for full transparencies, I sort of think about the maturity of Both tools and how people sort of do customer analytics. There's a fourth level that I think about as well, and this one is even like a little bit more forward looking than the sort of level three that we were just talking about a moment ago.
And it is, as you described it very [00:34:00] well, I think the sort of idea of like autonomous customer analytics. it's basically the idea that I can start to use algorithms such as AI to traverse my customer data and tell me what it is that I need to know about my customer, what it is that might be, you know, a meaningful metric, what it is that might be a meaningful attribute upon which to activate, etc.
then basically you can just go in and try a whole bunch of things, and then tell me which ones might work. what I would caution people against is thinking, this is the time to like kind of like skip doing your homework and let AI just come in and solve the problem later. Because no, like AI is much like anything else with data, garbage in, garbage out.
And so getting a good solid foundation of all your different sort of sources of customer data in place and getting it all shored up and putting it together is valuable, not just in the short run because you have a business to run until we sort of hit this AI singularity, if you will, but also in terms [00:35:00] of as these AI based approaches get better, they will benefit from having these data.
So, like, you should not, like, skip level three on the path to level four. you will not regret your investments in level three. But as I think as you kind of look at where we get to in level four, it really is this idea that, you know, once you have your data well assembled, you can very much, like, explore it algorithmically.
And I think one of the things that is exciting about this is if you think about kind of how we do campaigns today and even how we do analytics, it really is very much, you know, kind of a long tail distribution and we focus on the head of the curve. We focus on the big, very prevalent use cases that have enough commas that any, like, one or two of them will move the needle.
I think what's great about these sort of, you know, generative technologies and things where I can kind of just, like, point them in the general direction of data and let them run, is I can now sort of traverse the long tail of distribution as well. I can do a bunch of campaigns that might affect very small populations, and any one of them [00:36:00] individually might not be that meaningful.
But if I do enough of these in aggregate, I get a very significant result. And I can also kind of then, you know, manage having a whole bunch of metrics that might only be or attributes that might only be relevant to a subset of the population. But, you know, again, in aggregate, add up to something meaningful.
And so to me, that's I kind of If I had to sort of guess where things are going now, that would be a big part of my thesis is that, you know, it's all about really getting the data in place so, you know, and well sort of curated so that as we algorithmically grow and get smarter, we can, you know, feed better and better inputs to our algorithms.
And then kind of shifting from a mental model of we must sort of manually touch all the things and so therefore, It's about finding a few big things and instead shifting towards a mindset of about, if it's about that, plus being able to add up a bunch of little things that become something meaningful.
Tom: Oh, that's fantastic. I love these [00:37:00] visionary thoughts that you've shared with us here today. John, you're clearly an exceptional data leader, and an expert in the field. thank you so much, for joining us, on the show
John: today. Thanks so much for having me, Tom. I really enjoyed the conversation.
Tom: All right. A lot of really great ideas and thoughts, from, from John on today's episode. so Just, you know, first key takeaway is, you know, thinking about the example that he shared around inconsistent metrics, that he was seeing across different platforms. Cause I think that cuts really to the heart of the problem that, you know, data leaders need to solve for today, right?
We have to get rid of the fragmentation of tools and the technology has caught up. Right? So with the cloud data warehouse, really emerging in the last, say, five years, it's time to rethink the investments, that, that we've made in the past. And they weren't wrong investments, right? We made investments based on the technology that was available at the [00:38:00] time, but I think it's a good reminder, that we need to continually reevaluate the technologies, that we've invested in.
And he even provided us with a great framework, to think about and to rationalize those investments that we, that we've made, right? So he talked about effectively four levels of maturity, right? Starting with, where many companies have started, with investing in these point solutions, right? You have a business need, let's go out and solve those, obviously very easy to do.
But that is actually what has led to a lot of the fragmentation, in the data stack that we see today. And so Level 2, begins with companies and enterprises beginning to consolidate and centralize that data into a data warehouse, or what he calls the cust or, sorry, the Warehouse Native Customer 360.
Now we've seen, significant investments internally by many enterprises building on top of their data warehouse. An emerging class of SaaS applications now that are warehouse native to [00:39:00] begin with, right? You're not creating, siloed, vertically integrated solutions with your own source. Let's leverage the data warehouse, that already exists.
So companies like NetSpring, for example, we offer a warehouse native solution for product and customer analytics. That is level 3 and is really where the puck is moving in terms of SaaS. And finally, level 4, you know, what we're all aspiring to, but we can't skip level 3. you know, this concept of autonomous customer analytics, that was a very interesting, concept.
And he talked about how, with, you know, the, the use of AI algorithms, to handle that long tail, right. Allows us to get to that ultimate in, you know, one to one personalization. Today, it's very difficult to do, personalization beyond, segments. Right, because you can only define, you know, so much customization, at, you [00:40:00] know, at at a level where you can manage the, the workflows associated with that customization.
But with Jen, I, for example, I can personalize one-to-one and know exactly what's specifically I am interested in looking at, you know, pictorially or, or, or in the words that are gonna resonate, with me. So, level four, ai, it's what, you know, every data leaders out there thinking about. It's where we're going to aspire to be, here in the next few, years.
but let's leverage this powerful framework that, John has shared with us, to rationalize our investments over time. That concludes today's show. Thank you for joining us and feel free to reach out on LinkedIn or Twitter with any questions until next time. Goodbye.