This is a transcript from the AI and the Future of Work podcast episode featuring Dan Turchin interviewing Dave Kellogg, serial CEO, investor, and SaaS metrics expert

Dan Turchin (00:17):
Good morning, good afternoon, or good evening, depending on where you’re listening. Welcome back to AI and the future of work. Thanks again for making this one of the most downloaded podcasts about the future of work. If you enjoy what we do, please light comment and share in your favorite podcast app, and we will keep sharing great conversations. This is your host, Dan Turchin advisor insight finder, the system of intelligence for I O operations and CEO of people rein the AI platform for it and HR employee service. Now we’ve been inundated with 2022 predictions over the past month or so, but you know, one set stood out for me as being both insightful and provocative. I’d say above all the rest as today, we get to discuss not just the future of work, but specifically what’s ahead for the next 12 months and areas as diverse as blockchains security, venture capital, and even causal inference.

Dan Turchin (01:12):
Imagine that Dave Kellogg’s a serial CEO advisor investor and what I call the SAS whisper. He recently joined Paul capital as an executive in residence. Dave’s illustrious career has spanned exec stints at iconic companies like host analytics, Salesforce, smart logic, and even business objects before it was acquired by SAP among other accolades Dave’s Saster talks routinely rank in the top few most watch Dave owns two dubious distinctions in over a hundred episodes that we’ve recorded. Dave’s one of only three repeat guests on this podcast. He’s also the biggest grateful dead fan. I know. And look, I’m just gonna say, I assure you that two aren’t directly correlated, but perhaps loosely correlated without further Aue. It’s really my pleasure to welcome my friend, Dave Kellogg, back to the podcast, Dave, welcome back. Why don’t you share a little bit about what you’ve been doing over the last year or so since we last had you on

Dave Kellogg (02:16):
Super well, thanks for having me, Dan. Great, great to be back. Did not know I was in the verified air of repeat guest. So thanks for having me in the past year I’ve been doing more of the same really advising and consulting board work, agent investing. I think the new thing you mentioned in the past year was signing up as an executive in residence at Walton capital. They’re based in London. So I got to go over there in the fall. Right, right before Omnicron. And it’s been a very fun gig to get started and work with some of the folks at baller, both friends, both old and new. Let’s say

Dan Turchin (02:53):
If I refer to you as the SA whisper you’ve recorded a lot of your talks about SAS and I know one of the predictions in your 2022 Kellblog posts was about some of the metrics that you think are gonna become higher profile for SAS companies this year. Maybe let’s start off with talk a little bit about what’s up ahead in way of SAS metrics.

Dave Kellogg (03:17):
Sure. So I think, you know, for the last couple years, I think I’ve been a big better on net dollar retention, also known as net revenue retention. For the simple reason that I think it’s a far better way to value the install base of a SaaS company than churn. I think churn is too gameable. I think people there’s, I make a little quad of four different churn metrics you can report. And you’re never really sure which one we’re talking about. Whereas NR NDR is far less capable and I think it’s a, just a better way kind of from first principles, a better way of valuing the install base of this task company, which is really what people are trying to do. So that’s probably the biggest change over the last few years. That’s why this last talk, I think it was last year or the year before was churn is dead long live NR because I cuz I really believe that. And it’s, that’s been one of my more accurate predictions, right? You’re seeing more and more about net dollar retention and less and less about churn and less and less about customer lifetime value.

Dan Turchin (04:15):
So you’re a legendary enterprise marketer. A lot of us in enterprise software are struggling to connect with customers. It’s, it’s really hard to get engagement. When our tools became fairly blunt in a, in a work from home world, we don’t have the same opportunities to interact and build rapport, particularly if you’re, let’s say a disrupter as opposed to an incumbent and an incumbent, it’s harder to build relation. Any advice from a marketing perspective for entrepreneurs looking to improve engagement with prospects.

Dave Kellogg (04:49):
Sure. I mean obviously, and I was doing an interim gig running marketing at a 50 million SaaS company when COVID hit part of the things I do. I don’t do them very often is interim assignments. So I got to live firsthand through like relocating the entire events budget over to digital, right. And events in an enterprise software company are often like a good chunk of the budget. So I think obviously COVID forced us to embrace digital anything from content marketing strategies, to SEO and SM improvements to kind of content syndication. Obviously what’s the big one I’m missing? Well obviously to webinars, the whole thing, anything you could do online became basically the older thing you could do and you lost live events, both big and small trade shows, conferences. So that was probably the big reallocation. We try to replicate a lot of those events virtually like a virtual wine tasting and a virtual cooking class.

Dave Kellogg (05:48):
They’re good for small exec events. You can do that still, not quite the same, you know, there’s software packages that like you have private rooms and trade show floors and you know, you can emulate some of that. We, we did a virtual user conference very quickly. A lot of people do that now. You know, O overall, if I tend to pick one piece of advice for, for an entrepreneur, it’s, it’s a book on marketing written by a guy who sells swimming pools and it’s called, they ask you answer. And it was about how he transformed his swimming pool business by just producing good content that answered questions people had, like, what are the trade OS of fiberglass in a concrete pool? Right. and he wrote rather than writing kind of classical marketing content that nobody really wants to read, right. He wrote genuine, like I make concrete pools living. I’ll tell you the trade offs, if you are right for a concrete pool by this, if you’re not, you should look at fiberglass. and he literally built a big business by simply producing more and more content that tapped into the questions people were asking and answering them. So particularly for an early stage startup, as you’re trying to figure out your content strategy, why not just, I make the world’s best back for your category. Right. and put it online

Dan Turchin (07:03):
Two years into the pandemic. Would you say that there are any technologies that have enabled remote work and even potentially remote engagement from a marketing perspective that are now mainstream that have been introduced and mainstreamed on an accelerated, accelerated rate because of the pandemic, but that a couple years ago just didn’t exist.

Dave Kellogg (07:27):
I’m trying to think. I can think of a lot of examples of acceleration, obviously video conferencing and such. I can’t think of entirely new things, right? I mean, slack, video conferencing, these things are all part of our day to day life. If you know, hop in certainly for doing live events online, that would be a, a new one kind of driven by the pandemic. But I maybe drawing a blank on other, at least marketing technologies. To me, it was all about acceleration, right? I mean, the this is more broadly, but I still love the little joke mean that said what drove digital transformation in your company, right. CIO, CEO, COVID 19. and I think COVID 19 drove digital transformation of entire business from supply chain to marketing and everything in between. So but I can’t think I’m sure I’ll think of five ’em we’re done with the podcast, Dan, but I, but I can’t think of anyone’s view. It was all about accelerating stuff that was out there.

Dan Turchin (08:18):
Yeah. Virtual event platforms like hop in are very easy to use and certainly cost effective versus live events for vendors. But I found that engagement has waned over the last say 18 months. First of all, do you agree with that? And second of all, will there ever be a time when a virtual event will be as as compelling or, you know, as, as effective as a live event?

Dave Kellogg (08:44):
So, so look, I’m, I’m a, you know, longtime enterprise marketing person. So I like live events. I think live events are particularly important for startups. Like if you’re selling to a vertical and they have a association or you’re selling the CFOs and there’s a CFO club, right. Those kind of early opportunities to meet in small venues, the right contact is, is incredibly important. So I’m a kind of small live events person for startups big events, trade shows, conferences, people like those people go to those, you know, I went to SA that was great. You know, he spent $300,000 testing everyone. Right. So, so to get in disaster, you needed to get an antigen test before they let you in there. It was the first time since COVID hit that, I was at an event where I felt like it was all PreOn. I could kinda let my guard down.

Dave Kellogg (09:29):
Right. Cause I was vaccinated, everyone was tested and it was fun. And I have to say it was great to do that. By the way, the event was also largely outdoors. Right. It was going indoor outdoor and venue. So I think, you know, will events come back? Yes, I do believe they will. Has anything replaced them with the engagement level? No, I think look, the biggest learning I took from all this Dan was don’t pave cow paths, right. That was a popular expression back bubble 1.0, right where they’re saying, Hey, we have this new technology. Let’s not just pave over the cow path cuz the cow path was rent for the cows. Maybe we should put the road somewhere else. And I’ve always liked that metaphor. And I think the way to do digital events wrong is to pave a cow path, keep the same se se session structure, keep the same day S sure.

Dave Kellogg (10:16):
Right. You, you could take the agenda for the online event for the physical event and move it online without changing. And I think that’s a big mistake. So, so I think the real question is how do you reinvent the event for the digital platform? And I’ll give you a simple internal operational example. I was running QBR and for me at QBR used to be a three day, two, two or three day live meeting, big dinner, everyone just grinding it out. Part of it was an endurance test, right? Like get there at eight in the morning. And if you’re fading by six, you’re weak, you know, and you know, , you know, back in the day we go on eight. and you push it really hard, but it’ll force me to change that. By the way I ran him in, like I ran a full QBR in two or three days, three hour sessions. I literally cut the meeting back by well, more than half. And I thought it was just as productive, maybe more productive. So I think that was to me, the opportunity on this stuff, reinvent, don’t just pave over. and I think the more you pave, the more engagement’s good as suffer, the more it’s not gonna work.

Dan Turchin (11:21):
So you’ve teamed up with balder, a venture firm based in London. But prior to that, you spent most of your career in and around the venture ecosystem in Silicon valley. What should entrepreneurs know about the difference in the cultures of venture outside of outside of the United States or even outside of Silicon valley?

Dave Kellogg (11:46):
Well, that’s a big question. I think, you know, let me contrast a day because I’ve always been a bit of a, you know, I grew up in York city not New York city, New York city suburbs. And I came out here to go to college in California. But I’ve always been a city person and I’ve always been more of an east coast person. I was kind of a, a reluctant Silicon valley resident. Like I was here for the work cause cuz that cuz you had to be here for the work back in those days. So I, when I had a chance, like I worked for Silicon valley companies in New York city for a while, a anytime I was given the chance to go somewhere else and work for Silicon valley company, I did including five years in Paris in Paris at business object.

Dave Kellogg (12:23):
So my first five years at business objects were in Paris. And at the time I can say the venture scene was totally different in Europe than here. That the first there just wasn’t my much of it. Second, the terms were horrible, big dilution, bad terms, participating preferences, all those stuff you could think of was just bad was, was the only way you could get money over there. Even the mental was different entrepreneurs wanted to have a large piece of a small pie. Like when I was looking at France, a typical French entrepreneur would literally rather own 80% of a 20 million business right than 5% of you know, 200 million business. And the last thing, and this was a big deal in France. I can’t say it was true across Europe, but I’ve always viewed failure in Silicon valley as kind of the red badge of courage.

Dave Kellogg (13:10):
If you remember that book, you probably read in high school, right? Hey it’s okay. I have a red badge of cars. I was at a failed startup. It’s been good for me, character building, arguably and in Europe, that was not true. And that, to me there, there were two big differences between Silicon valley and Europe. In my experience, one was failure was a stigma, not a kind of, you know, a strike or a badge of courage, right? Time served. the other was the two, what were they? It was the stigma of failure and the lack of a regenerative ecosystem, meaning in English that everybody who made money, business objects left France and every who makes money in Silicon valley becomes a venture. Capitalist does another startup becomes board member becomes an advisor. And that was a thing that I watched as I moved outta Paris around year 2000.

Dave Kellogg (14:08):
It was just sad because we had a lot of people in Europe. We had like, who knew how to build a European business that attacked the us market and succeeded. And it all kind of dissipated people had good careers, some retired, some moved into venture capital, but not in France. Right. and that’s changed now. I can say that the European VC has now started to catch on big time. And we’re now in that second generation where you have repeat founders where you have founders who’ve been successful, starting funds becoming in advisors, becoming Scouts, becoming angels. So that’s like super exciting to watch because cuz that to me is almost like the second stage booster of the rocket. So I’ve always loved Europe and working over there. So when I got in touch with, I was happy to do this and the biggest thing I was happy to see was wow. It now regenerates cuz cuz it didn’t used to

Dan Turchin (14:57):
Are the soaring valuations from Silicon valley permeating the rest of the world or is more of the conservative approach in Europe permeating into Silicon valley?

Dave Kellogg (15:10):
I think more the former than the latter. I don’t know. I mean you could, I don’t really know how global the VC market, you know, was five years ago. Like I don’t think if you were in London, you could raise money from a Sando road firm that easily. Now of course they’re all starting London offices. So, so a lot of them have dropped down offices in, in London or other European cities. So now you can argue that the market for startup capital is more globalized, but I always thought there was kind of an arbitrage play like 15 years ago. I was like, Hey, I should start companies in Europe, move ’em to the us. and do the arbitrage play between the two different, I think that AR plays a lot less now. I’m not an expert on this one cause I have it with that many deals. I do more operational advising to be Frank, but, but everything I read in here says European evaluations, a crap. They’re probably not the same level as us, but they’re definitely come up and I don’t think that our opportunity works anymore.

Dan Turchin (16:06):
So by your own admission, you got about 70% hit rate for your 20, 21 per day, which I think is pretty good, solidly better than a passing grade. and those are some pretty pretty ambitious predictions. So now we look at your 2022 predictions and one of the sections that thought was thought provoking was your section on on web three. And one of the sub themes was I’m gonna paraphrase here, but blockchains are the enterprise data store that nobody needs. talk a little bit about your thoughts about blockchains and use cases in the enterprise.

Dave Kellogg (16:47):
Sure. So, so, so, so first that, that 70% was on my admittedly self charitable scorings scale, but yeah, it was a pretty good year for, for my predictions. Web three and blockchain let’s go straight to blockchain. I I’ve kind of always felt this way about blockchain. And as a reminder, Dan, I’m a database guy from way back, like I was using relational databases in 1983 at Lawrence Berkeley lab was an undergraduate at Berkeley. And my first job was tech support for database in 1985. So I’ve been using databases for a long time. After leaving that job, I worked at an object database company then went to business objects where we sold kind of a light multidimensional database. I ran mark logic, which is an XML database. I was on the board of asked data, which is one of the original, big data databases. I could probably find some other database stuff I’ve done. but I’m a database person. And I’ve worked a lot with both general purpose and special purpose databases. And my, my overwhelming theory is special purpose databases are a good thing, but people always seem to wanna overly them, like for example, the multidimensional database.

Dave Kellogg (17:57):
So one way I characterize databases is, is what’s their data model. So a hierarchical database, it’s a hierarchy, a network database, it’s a network and a relational database. It’s a table in a object database. It was typically a C plus plus class structure, right? That, that was the kind of, what did it model an XML database. It’s an XML document in a graph database. It’s a graph and a time series database. It’s a time series, right? Every database has this notion of kind of fundamental modeling unit. And then you take stuff that isn’t that and you try and map it in, right? Like you could build a table on XML and I could store know the employee in the department table and XML documents, good luck running a join. You, you gotta do the world’s slowest join. Right. Cause it doesn’t understand the data structure, joint optimizer.

Dave Kellogg (18:38):
So I think as somebody who’s been into specialized databases, as well as general purpose databases, whenever I see a special purpose database, I say, what was it built for? And the are it’s really good at that. Right? Like I always say, if programmers got to pick what kind of databases we’d use, we’d all be using object databases. Right? Cause they have no impedance mismatch between the programming language and the database. Right. so when you look, when I first looked at blockchain and I read the Satoshi Bitcoin paper and started learning about this stuff, I was like, oh, this, for this use case, this makes a ton of sense. And then what happens in Silicon valley too often, in my opinion, is people get excited. Sand hill road gets excited. Technologists gets excited and they start saying, look at all the other things we could do with this and look a big part of Silicon valley is it’s exactly that bridging process.

Dave Kellogg (19:32):
You know, Facebook bridging from college students to, to university students, to humans, to the whole world, right? This notion of systematic build is a very important part of what makes Silicon valley great, but it gets over applied sometimes. Cause people look at blockchain and go, oh, it’s got a hundred enterprise uses. And I’ve always been like, why is the enterprise need blockchain? Why do I need a public slow database? and I don’t that that’s not asset or base. I just don’t understand. And the answer is if you wanna build Bitcoin, it’s a great use case, but I’ve always felt like other use cases were to use the famous quote technology and search of a business problem, right? They’d say, oh, I have this really cool technology. And blockchain is cool. And then say, let’s go apply it to other problems. And they kind of search for problems to go solve.

Dave Kellogg (20:25):
But I just think that’s the wrong way to do things right. As, as a somebody who wants to advise companies and technology, tell me about your problem, go find a technology built to solve that problem. and frankly, the further your use case gets from why they built it, the less you should buy it. right. Like tell me what you were thinking. It’s always a first question. Freddy founders, where were you working? And what were you thinking when you invented this thing? Cause I’m betting that if you’re good, that it’s gonna be really good at that thing. And then we need to see how well it generalizes. And I just think blockchain got so much hype around it that people just started to try and, you know, kind of pushing on a string, just like, oh, use blockchain for this, but blockchain for that.

Dave Kellogg (21:07):
and if you look at blockchain, it’s, it’s a, it’s a public database, all data is public. I mean, you could build a private blockchain, but by default it’s public, it’s deliberately slow. Right? I I’m building a database that has basically a list structure that runs slow on purpose. Like I’m slowing it down through proof of work and proof of work. The best metaphor I heard from proof of work the other day was when airlines fly empty planes to hold their gate slots. , it’s really that proof of work. So it’s incredibly, it’s not just slower inefficient. It’s deliberately wasteful right. To slow down the system. and it’s, it’s, it’s imutable, which is probably its biggest strength, right? That that’s awesome. But the consensus, the consensus algorithms behind it are quite wasteful. So I just don’t think, I mean, enterprises are almost the exact opposite of that use case, right?

Dave Kellogg (22:02):
We have our own databases, we have our own security infrastructure. We have our own it, we, we are a central thing. Like we have a CEO, we have a CIO. I think my enterprise blockchain argument is I have trouble finding low trust enterprise use cases. Like tell me the problem again, that, you know, like relational da, well, all databases were basically originally built to do transactions, right? Databases are really good at transactions, right? I mean, those, the democratic benchmark was a, a banking checking transfer, right? I mean, that’s what we do here. So if you’re gonna tell me you’re gonna do that better and make my banking get slower and distributed and decentralized, I don’t understand it. I think, look, I do believe there are some enterprise use cases. If you look really hard, you can find something other than PLU, by the way, most of the time, if you just type enterprise blockchain, the terrible Testament to the state of search today, but it’s P I was on a slide this this morning, it was just jump. And it’s, it’s actually quite hard to find intelligent writing about enterprise blockchain use cases. I’m sure they exist, but to me, it’s the pattern setting off in my mind is cool technology. Everyone got excited and one’s trying to apply it to do things that was never really built for.

Dan Turchin (23:14):
So you pulled kind of an, emperor’s got no close on on web three, which I I’ve been thinking about a lot. And you said basically kind of the, you know, the whole thesis around decentralization inevitably needs to, in all order for the vision of decentralization to be realized you’re essentially gonna need these centralized platforms to harden web three. It’s a little bit controversial in Silicon valley. Talk about what why you think centralization is gonna essentially be required for developers to, or, or for, for users to start trusting the vision of a decentralized web.

Dave Kellogg (23:54):
So I think a couple things, I mean, first my favorite part of that part of the post is the links. I have some links to what I think are really awesome articles. Tim O Riley’s post probably the most balanced of the lot is fantastic. I’ve gritty boots tweet. That’s pretty funny that I’m looking to after the initial publication of the article, I found another web three source by Moy modeling house. But called first impressions of web three. I’ve now added that and linked that in it is a phenomenal work. So the first thing I’d say, as you’re interested in web three, I spend a lot of time finding those links, go read the stuff at those links, consider it a curated link collection of good stuff to read on web three. Now to the question of centralization, you know, what basic prediction is, web three is gonna deliver a lot of cool stuff, but it probably will not deliver decentralization that, that, but like so many things to technology, it will accidentally deliver a lot of other cool things.

Dave Kellogg (24:50):
but the actual thing we’re setting out to achieve if you believe that is the decentralization, I’m not sure it happens. And here’s why one of my little theories is the lower you are in the stack. The more the world wants standards, right? That, thatI want there to be one standard for how to read a CD rom or back in the day, a tape, right. I want there to be one standard for oh, I don’t know for USB, well, there’s multiple USB standards, but like to be able to buy, have a standard USB port and a standard USB device. I don’t wanna, I don’t even wanna make decisions about those things as is a buyer, right? So, so to me, the lower level you are on the stack database is being a great example, operating systems, being a good example, right?

Dave Kellogg (25:34):
The world doesn’t like entropy at low layers of the stack, the world likes standardization CIOs like standardization at the low level of the stack, right. I once met the CIO back in the day of GE and he basically, and not so many words said, my job is standardization of infrastructure. And it was right. Cause cuz you’re trying, there’s so much by default, right? So much stuff can creep into your stack and make it so chaotic that as a CIO, you’re basically defender of the stack and the lower and the stack, the more you need defense, right? You need to say, no we’re standard on this version of right at athletics or we’re gonna use windows or we’re gonna use whatever, but it’s gonna be very important that we standardize. So I think, I think that’s, what’s hurting web three, which is, it’s a bunch, it does a really good job of decentralizing kind of the internal operations of the database.

Dave Kellogg (26:25):
But if you want to call this database as much, see Monica has pointed out in his article, you’re gonna call it through some APIs. And now those APIs become, and by the way, the world wants standardization at that layer, right? You, you want programmers to know how to use those APIs. So I just think there are these kind of irresistible forces that drive us to standards. The best example I saw was if you read the morning house, first impressions of web three posts, he built two dApps, one called autonomous art where anyone can commit in an Ft. The other one called first derivative, where he basically creates a kind of tracking stock for a popular NFT and open C bandit. So if I go on wearable, I bought one. If I go on wearable, I can see it. Whereas open C has said, no, that’s a copy because it’s basically the same NFT, so we’re gonna block it.

Dave Kellogg (27:19):
So it’s still on the Ethereum blockchain. If, if I went into an Ethereum block Explorer, I could find my NFT, my first derivative. But if you’re gonna access the Ethereum blockchain, the way most people do, you’re gonna go through open C and open C is blocking it at their API. It doesn’t exist for all practical purposes. So, so here we are, we’re centralized again. And by the way, all these curated collections and by the way, all the scamming on discord, I mean, one of the more disturbing things about web three is the number of people who like my apes are all gone. There’s a famous tweet. where, where literally someone DMS you, something, you connect your wallet to it and your apes are all gone. And by the way, each ape today sells for a minimum of $300,000. So if somebody’s losing apes, you’re talking about a lot of money.

Dave Kellogg (28:08):
So and you see over and over, I just tweeted this morning, a woman that had been scammed, trying to buy, I think they’re called cyber girls. And it happens a lot. So we want, and to some level yeah, welcome to the wild west and be careful who you accepted DM from. Right. and the, how we want open sea security team, a meta mass security team, trying to protect us from those people. What is that centralization? Right. So I think, I think the dream of decentralization is actually not what people want. I think people like NFTs. I actually like Ts, so it’s kind of a cool idea to create scarcity and collectibility I think Ts are kind of cool. I think there’s a lot of crap Ts out there, but Ts are cool, but do I really?

Dave Kellogg (28:53):
But I buy ’em all through open sea right. Or wearable. Right. And there’s our middle, my favorite it’s one of my favorite tweets was Sharma founder of sky flow. You may know him. He’s like, if you don’t need intermediaries, why is Coinbase worth 50 billion of dollars again? so, so it’s like, yeah, we do need the intermediaries. And yeah, I have an NFT that’s on the blockchain, but it, for all practical purposes, it doesn’t exist anymore. Cause it see doesn’t like it. So, and I’m okay with that, to be honest, cuz It’s not gonna make me, you know, not wanna buy on open seat. So I just think it’s ultimately, if you wanna Tovo at these services, you’re gonna go through some API and the argument is, look, here’s the, let me do the counter argument. The counter argument is yeah, but the blockchain is public.

Dave Kellogg (29:35):
Somebody else could build one that shows your NFT, Dave, that open C is choosing to hide and that’s true. But in Silicon valley law of increasing returns, E E even just in consumer marketing in general. Yes. I suppose it’s theoretically possible that the, but this way it’s not the theoretically possible disrupt Facebook, right? Their data is in their databases. They own it. Right. I guess you could disrupt them, but you can’t take, they own their data. Never getting the data. Whereas the difference here would be the data is public. That’s a public blockchain. Anyone can read it. And you, you could see it. So, so there are some differences, but I think they’re practically speaking moot frankly.

Dan Turchin (30:16):
So this creates a ton of cognitive dissonance in the web three community. And because we of our predictions, let’s say a year from now, we’re sitting here and we’re having another version of this conversation and are, you know, are the, are the web three pioneers gonna mute me when it becomes clear that there’s consolidation around a small number of centralized platforms, your open sea and wearable examples are great ones has the community, is it, is that gonna reject the antibody?

Dave Kellogg (30:44):
Yeah. I don’t know. I don’t have a prediction of what they do about all this. I’m sure. Look, my basic belief is all this stuff gets overhyped web 2.0 was overhyped. We wanna go back in time, by the way. I also link to Tim Riley’s original web 2.0 paper, fantastic paper to go reread right now, get a feeling for what it felt like before this same sense of hype and confusion. And what’s real and what’s not. Cause there are a lot of people were anti web 2.0, right. and no, the notion of my, my favorite definition of web 1.0 versus two point was read only versus re Wright, right? There’s a blog called the rewrite web. and it was kind of a technical way of looking at it, but that’s what social media was, right. It was a read, write web.

Dave Kellogg (31:30):
I could tweet, I could blog. I could do a lot of things without having to kind of run my own web server to go build a blog. I could just use a service. I could just use Twitter to do micro blogging as, as it was initially home. So I think it’s good to go read that paper to get historical perspective. And I think if that cycle repeats itself, what happens is there’s basically some of it ends up being empty hype. I have a feeling decentralization, it might end up more empty hype than reality, but a lot of good things will come of it. A lot of good apps came out of web 2.0 I just don’t think it’s gonna solve. Look. It, it all depends what you’re looking for web three to do. and because that, that’s the other point, right?

Dave Kellogg (32:17):
If you want to create this decentralized idyllic world, that it probably won’t do that. but, and then one of the reasons I go use all these things, Dan is to get a reminder of that. Discord is literally terrifying. I’m, I’m afraid to press any link in discord. Like the first thing they tell you to do in discord is disabled DMS. Like if you want to keep your NFTs, you know, disabled DMS. So I guess it’s great that it’s decentralized, right. Code is law. There is no intermediary, but holy, holy cow, maybe intermediary, aren’t such a bad thing. When, when you lose all your apes, like if, if, if by bank of America account just showed up with zero balance and it, I could call B of a and start legal proceedings to get my stuff back. In fact, one of the more cold hearted responses to the, I lost my ape sky.

Dave Kellogg (33:02):
I dunno if you saw this one on Twitter, is somebody saying, why does your profile pick have an ape? You no longer own oh man. That’s cold. so I don’t know. I don’t think in the end, I don’t think there’s as much. I don’t think people want you. I think people like cryptocurrencies, I think people like Ts and I see a bold future or a bright future for those things. But I think the it’s kind of, you know, what it might be like Seman web. Remember when web was happening the next level beyond that was semantic web with the original web invent. Can’t remember the name right now. Tim burner. Lee. Yeah. Thank you. Sir, Tim burn Lee, I Lee pushing the semantic web vision Seman web didn’t happen. RDF didn’t happen Al didn’t happen ontology web. And it happened a little bit, but basically macro, it didn’t happen, but we got Aite web. That’s kind of what I think will happen here. We’re gonna get decentralization that they want, just like we won’t get Savannah web, but we are gonna get a lot of pool stuff including cryptocurrencies NFTs. I’m sure I’m getting some stuff out, but that stuff’s gonna happen.

Dan Turchin (34:14):
One of the things we talk about about a lot on this show is building public truck us in AI, and specifically who owns your data. And when AI’s making decisions based on your data, what rights do you have to understand how those decisions are made? And that that’s what intrigued me about one of your predictions about moving beyond or to increase trust in AI again, paraphrasing and moving to a world where we expect more, what you call causal inference. So with respect to, you know, 11, the 11 months left in 2022, it seems so bold that something as foundational as causal inference will start to permeate or infiltrate how AI based decisions are made. It, it, do you really think that we’re gonna make that much progress in, in 11 months?

Dave Kellogg (35:08):
Sure. And I’m not even sure these are only AI based decisions. These can actually be human decisions as well. So let me go back for a minute here to one of my favorite lines is if, again, show my age, but in the graduate 1970s, something movie with Dustin Hoffman one of the dads grabs him at the swimming pool and goes, plastics. I have one word for you, plastics. And the future is gonna be plastics pretty ironic. Now, when you think about the situ I guess he was right at blood level judging by the oceans. But the in any case I’ve always liked that scene where you, you, an older person goes, a younger person, says I have one word for you. And for at least the last 10 years, and I started to say long time ago, it was data science.

Dave Kellogg (35:50):
I was, and here’s the thing about this prediction. My meta prediction is for the next five years, it will be one causal for will be one of my predictions, just like data science was, if you look at my predictions ever since I’ve been making them, one of them has always had something to do with data science. In fact, this year one still does which is BA data. Science is the new MBA is, is the prediction this year. but I do feel like causal inference is the new plastic using my metaphor. And I think it’s very, very early. I don’t think it’s gonna change a lot in 2022 to be honest, but it’s one of these predictions that I’m predicting I’m gonna make every year for the next five years, 10 years, and causal inference is gonna enter the mainstream vocabulary. One of the things I try to do in my predictions today as well is try to make some that are relatively obvious extrapolations of current trends.

Dave Kellogg (36:38):
Usually the last one is kind of the the not always, but often the last one is kind of the craziest one or the most visionary one. And therefore it’s, in my mind, it’s actually on something of a different timeframe. But I do believe if I can dig into the prediction itself that, you know, I’ll just pick the example I use in, but I was at a meeting one time and somebody said, you know, while I’ve been analyzing our churn and, you know, I, 50% of the customers who churned of the big customers who churned in the last year had filed more than five severity, one calls with tech support and then followed by basically God. We need to do something about reducing the number of severity, one calls. and you hear that all the time.

Dave Kellogg (37:24):
And that statement will go unchallenged. In most business meetings, people will be like, you know, righteous. Yeah, let’s go do that word . and it’s like, wait a minute. Like, no, I mean, what else do those companies have in common, other than they filed five, it support calls. And by the way, this is an old blog post I did called you. Can’t analyze churn by analyzing churn. Which is, if all you’re looking at is the people who churned, you’re never gonna be able to differentiate what separated those who churned from those that didn’t the whole, the whole premise is analytically flawed to go look for common patterns in those who churn. When the question you’re trying to answer is the other question, which is what differentiate those who turn from didn’t. And then the even deeper question, which is, okay, what does separate those two groups? Was it causal? Is it just correlation, right? Does the windmill turn because of the wind or does the wind blow because the windmills turn it, the silly example that we all intuitively know the answer do. Right. but still, if you were from another planet and you landed on earth and saw a windmill turning, you would have to think for a minute about what’s causing what’s effect.

Dan Turchin (38:31):
So I gotta get it to a stopping point, but not without asking one drill down on that, how do we know when we’re there? What does it look like when we’re making decisions where we’ve fully diagnosed the, you know, the root cause such an elegant prediction and yet so hard to measure?

Dave Kellogg (38:49):
Yeah. So I think, I mean, to me, the measures gonna be pretty simple, frankly, when, when your company has a causal inference team, I think that’ll be the first sign. Just like when your company had a data science team, like, but, you know, we had analysts long before we had data science teams, they were just called analysts and maybe they worked at finance, or maybe, maybe there was an ops guy and an ops team or an ops. But the manifestation will be, and look, I’ve got an inside track here, Dan, my, my son is a PhD in behavior economics and works on the Colin team at Google . So this is, this is where I got the idea. And I’ve started to learn about Colin, for instance, just like you said, the first thing we’re all taught is correla. Correlation is not causation.

Dave Kellogg (39:33):
Right? Everybody knows that, but what do we actually do with that knowledge? Nothing so, so we all know it. We all believe it. And then we go to a business meeting and we say, all the customers who churn have blue eyes, let’s not go get any more blue eye customers. Right. and it’s like, we’re not applying and interpreting the data. So an even data science teams could fall affected with this because they’re, most of the tools we have to air modeling tools and the models are effectively some sort of regression. So, so, and they’re just saying what things move together, right? They’re not saying what moves, what, and the difference of the fundamental. So, so to answer your question, there is an easy answer to your question, which is when you have a causal inference team. and basically when the data scientists and MBA analysts gets stuck and they say, uhoh, we need to causal guys then, or gals, then you’ll know that, that we need to causal team and we’ll go in there and work with them.

Dave Kellogg (40:34):
That’ll be the easy way to detect it. Other thing hopefully you’ll start to see in meetings is that people are gonna be more attentive to the fact that these two things are correlated, but potentially not causal. and just think awareness, right? Like just, how did data science happen? This to me, it’s just gonna be a replay on data science, which by the way, great irony that itself is kind of inferring. Cause invalidly, so this is the first thing my son saw when he read the post. He’s like, you do see the irony here. It’s like, yes, I see it. And I was wondering how fast you’d see it. Cause I said, if this basically repeats like data science did, this is what’s gonna happen, which is on what basis would you assume it would repeat? And I don’t really have one, but let’s just say without getting too much in the weeds, if it does repeat in a similar fashion to data science, you’re gonna first hear about it here.

Dave Kellogg (41:23):
It’s possible. People are gonna start hiring PhD level people to do it. They’re gonna start getting big insights. We’ll hear those stories. People will start majoring it as a degree, it’ll become a job, a data scientist, literally wasn’t a job 15 years ago, right? Like so that’s what I’m thinking is gonna happen. And that’s why this is a long term prediction that now that we’ve gotten so good at basically regression and correlation models, people are gonna start to say, what’s actually causing what. And when you know you’re doing that is basically, well, one of two ways you’re running a randomized control trial, right? You have to run an experiment. If you really wanna do causality, you need an experiment. That’s the easiest way to do it. And that’s what pharmaceutical does with, you know, drug trial and stuff, a double blind randomized control trial.

Dave Kellogg (42:16):
That’s an experiment you run to determine cause this year’s to give you an idea of how far out this may be. This year’s Nobel prize was given in economic, that it wasn’t economics. I can’t remember, but the Nobel prize in something, I think it was economics. The prize in economics was given to people who figured out how to basically get causal inference off observational data. I E without running an experiment, because there are issues in public policy where you can’t run an experiment or morally ethically, practically, you just can’t run an experiment. So they figured out a way, if you have enough data to just use observational data, to get valid causality inferred, basically. So, so is this stuff kind of big and hard? Yes. Experiments. Aren’t that hard by the way, we can run experiments now, thank goodness in today’s internet there’s we can run lot and Google and all the big tech guys run lots of experiments, right?

Dave Kellogg (43:15):
We’ll give you this and not give it to you. So I have a control group. We can watch a thousand users and another proceed differently where one had the feature and one didn’t that If I hold everything else constant and have a kind of random selection for who goes in which group that that’s a pretty good experiment to tell me something. The really cool thing is that you can also do it when you can’t run the experiment. So long story short, I’m a kind of a new at this theory, as you can probably tell, or at this area, and I’m trying to learn more about it. So why I put it into predictions? Cause I’m excited about it for the simple reason that we talked about. Everybody knows that correlation and causation, that correlation does not imply causation, but we say that and then move on and act as if it does. and I think over the next 10, 10 or 15 years, that’s going to change. and I think that’s super exciting. And that’s why I’m saying causal inference is kinda the new plastics.

Dan Turchin (44:06):
I think we have to have a true expert. Come on this show in the form of your son to talk us, talk to us about applications of true rigorous causal inference and you build, build on some of these themes.

Dave Kellogg (44:20):
Sure. Be happy, happy to connect you to him or his teammates. Yeah. It’s a super interesting area. Right. I mean, as soon as I heard it, I was interested in it. I was like, what do you do Google again? And like, oh, wow, that’s really cool. I mean, it was like data science soon as I heard it, I’m like, oh, that’s cool.

Dan Turchin (44:36):
Very German. Yeah, very germane. It’s the future of data science future of enterprise data. It’s the future of consumer data? Love it. Well, Dave, we’ve done way over time as we always do, but this was too good to cut short. And I’m not gonna let you off the hot seat answering one other question. And for some reason, I don’t believe this appeared in your in, in your set of predictions. If 2021 were a grateful dead song, what would it have been? And more importantly, if 2022 were a dead song, what would it be when we’re looking back on it in 2023?

Dave Kellogg (45:14):
So this is exceedingly difficult question, but I’m gonna go with Stella blue for 2021 all the years gone by, they melted into a dream, which is kind of what, what 2021 felt like 20, 22. I’m gonna be optimistic and say, trucking. I hope it’s the year we get back trucking off and you know, start to move forward outta the pandemic. Cause it’s, it’s now getting, you know, I have a, a friend who’s got a, you know, four year old kid and it’s like half, half their life has been in the pandemic mode. Right. More than that it started to be a big chunk of, of people’s lives. Right. and it’s I hope we can get backtracking on. So I’m gonna go with still a blue for 21 and chucking for 22,

Dan Turchin (45:57):
Dave, as long as it’s not monkey in the engineer for 2022, I’ll be happy. good stuff. Well Dave is always just a pleasure speaking with you. I think we only covered probably about 5% of what’s in your predictions, but hopefully we’ll, we have more time to discuss some of the others and more important when have you back next year. We’ll we’ll be able to rate rate how, how, how you did.

Dave Kellogg (46:20):
Okay. Good. Good. Well, it’s great talking to Dan. So it

Dan Turchin (46:22):
Pleasure Dave Kellogg talking to us about his predictions for 2022, I’ll make a point of linking to all of the all of the posts that Dave referenced, not just the predictions go out and read, go out and follow Dave he’s at Kellblog on the Twitter. You’ll be hearing more from him very soon. This is your host of AI in the future of work, Dan Turin signing off for this week. But we’re back next week with another fascinating guest.