Unscaled 5: Just the Right Amount of Personalization

Is personalization really the golden ticket that some product creators think it is? In the tech industry, it’s a widely-held opinion that personalization is the answer to everything, and that successful products must be custom-tailored to meet the unique needs of each user. But it’s difficult to think of really successful products that are hyper-personalized. The iPhone, for example, is more or less the same phone for every user. So is personalization actually important? And if so, to what degree?

Show notes

(Full transcript at bottom)

Hemant Taneja’s book Unscaled: How AI and a New Generation of Upstarts are Creating the Economy of the Future (referral fees will be donated to charity)

Hemant Taneja, managing director at General Catalyst

Ronda Scott, marketing partner at General Catalyst


Is personalization just a Silicon Valley fetish? (2:11)

Livongo provides chronic care management solutions for type 2 diabetes patients (2:57)

Concept of personalized medicine (4:20)

How to create boundaries for personalized groups (5:21)

Optimizing for a demand-oriented product (6:10)

There aren’t 7 billion types of diabetes (6:50)

Comparison the music industry (8:30)

“What is the right amount of personalization?” (9:35)

Don’t underestimate the power of delight (10:14)

Flocking to certain groups of people at the expense of others (11:07)

What do users actually want? (12:56)

Shift from a supply-based economics mindset of products to a demand-based one (13:44)

Products with the right amount of personalization (14:49)

Strava, an app for runners and cyclists to track their progress (15:07)

Sectors where unscaling has been happening for a while, e.g. online shopping and video (16:30)

Digit, an app that helps users save money (16:47)

Wealthfront, an app for financial planning and investing (16:48)


We want to hear from you

Please send us your comments, suggested topics, and listener questions for future All Turtles Podcast episodes. Season 2 is coming soon!

Email: hello@all-turtles.com

Twitter: @allturtlesco with hashtag #askAT

Voicemail: +1 (310) 571-8448 Leave us a message and we’ll read your voicemail on a future episode.

For more from All Turtles, follow us on Twitter, and subscribe to our newsletter on our website.



Phil Libin: Welcome back to the Unscaled Series from the All Turtles Podcast. I’m Phil Libin, and throughout the series I’ve been talking with Ronda Scott and Hemant Taneja from General Catalyst about how meaning of the term “scale” has evolved. Companies used to have to get big before they could have impact, but now they can just focus on making an impact before they worry about getting big. That shift has had a number of implications for the way startups and entrepreneurs should operate. In our fifth episode, we are going to talk about personalization and whether or not personalization is really the golden ticket that people seem to think that it is. I don’t actually think so but lets see what Hemant and Ronda have to say.

Phil Libin: Alright, welcome back Hemant and Ronda.

Ronda Scott: Thanks, it’s good to be here.

Hemant Taneja: Thanks Phil.

Phil Libin: The first half of the series, the first four episodes, we, I think lay a pretty good arc about what’s different about making new products, new start-ups, new companies today. Based on Hemant’s book, Unscaled, and the central hypothesis is the way the world used to be set up is you had to get really big first, then you could have impact. And now it’s kind of flipped on its head, it’s the other way around.

Phil Libin: First, you have impact and then that’s what lets you scale. And today I want to talk about personalization. In an early podcast segment of the normal All Turtles Podcast, we had the segment that we called “Maybe it’s Kind of Bullshit,” where we talked about things that maybe are bullshit, we’re not exactly sure. And the first one we did was about personalization and I kind of said, “Well maybe personalization is sort of bullshit,” in the sense that the, you hear personalization in the tech industry as kind of the answer to everything. Like every product is gonna be better because we’re gonna make it uniquely special to the needs of the customer. And it’s really hard to find actual examples of really successful products that are like hyper personalized.

Phil Libin: Most of the really successful products, like the iPhone, they don’t not really all that personalized. It’s more or less the same phone for everywhere and for every market. You hear this a lot about healthcare, about marketing, about advertising and it was harder and harder to find examples where specific personalization for the actual person is what made it a successful product or big difference. And so I’m wondering if this whole thing is just kind of a fetish. A silicone valley fetish for personalization. What do you think?

Hemant Taneja: Well I don’t want to start this session with a disagreement with you but I do disagree.

Phil Libin: That is the best way to start the session.

Hemant Taneja: I don’t think it’s bullshit, I think the question was what does personalization mean? The idea isn’t that every single person needs something different in every single category. So maybe when we talk about shoes our feet are all different and so we immediately think, well everybody’s got to have a custom shoe that just looks different from everybody else.

Phil Libin: Right.

Hemant Taneja: But that doesn’t apply in

Phil Libin: Had it been like five hundred start-ups that have done that, right?

Hemant Taneja: Correct. But that doesn’t apply when you think about how unscaling is actually taking on personalization. Take the example of Livongo, Livango is company we’ve talked about it focuses on providing technology based chronic care management solution to folks that have type two diabetes. And in the older, or today, the way we have always treated diabetes is as of this protocol that you go see your PCP two to four times a year, primary care physician. And he’ll checks your blood glucose and if it’s generally looking okay, he’ll assume you’re fine. And that’s the same protocol that you push everybody through.

Phil Libin: Yeah

Hemant Taneja: What Livongo is doing, is it actually captures data about you every time you check your blood glucose. And it mashes it up with other data around you and get a deeper understanding of you. And studies that are not looking deeper at the cohorts of type two diabetes would probably say there’s five to six different kinds of diabetes that have been lumped together as type two diabetes. I bet by the time we’re done, we actually have more comprehensive data, there’s maybe dozens of different kinds of metabolic diseases that we lump together as type two diabetes.

Hemant Taneja: Well that’s personalization because then your context, we’re understanding exactly what’s wrong with your metabolic pathway. Now there might be hundreds or millions of people like you that have the same issue and they, they did the exact same solution, it’s the best they did for them.

Phil Libin: Right.

Hemant Taneja: But it’s not category, it’s groups. So it’s categories

Phil Libin: So we kind of push on that a little bit. So this idea for, you know, you hear a lot about personalized medicine. About how in the future or, you know, in the current day, there’s not going to be eight prescription drug. There’s going to be a drug just for you. It’s gonna like understand exactly your unique snowflake needs and capabilities and a treatment that’s just for one person. And that’s always felt a little bit fishy to me or maybe it works for some diseases but not others. And certainly for like metabolic diseases like diabetes, it’s always seemed like there’s not seven billion different types of diabetes. Yeah you can combine all sorts of data about each individual and give behavior recommendations, but you’re never going to recommend like your behavior recommendation from your customized AI diabetes pap, it’s never gonna be like eat another donut. It’s always gonna be more or less the same stuff. So how personalized is it or is it just more or less the same thing for large groups of people and we’re just trying to figure out where those natural boundaries that define those groups?

Hemant Taneja: I think understanding the boundaries of groups is really what this is about.

Phil Libin: Right.

Hemant Taneja: Traditionally, we have tried to create those boundaries with intuition. So, oh type two diabetes is a general pattern that people have and as more people, more people started having those generally similar metabolic conditions, we lumped them all to the same clinical protocol

Phil Libin: And I think that’s the super interesting point that, that you make in the book a little bit. Right? It’s this idea that you have to group people together, right? You have to make these clusters and in the past, the clusters were made for the purposes of being able to efficiently produce some product. They were like the clusters that made sense to the company, to the producer. Not necessarily the cluster that made sense to the customer.

Hemant Taneja: That’s right. So it was, it was personalization was a supply oriented phenomenon. Which was, when we think personalization, how do we make something that’s different for everybody is what we’re talking about. When you make it a demand oriented concept, it’s about well what’s best for you? Almost doesn’t matter if seven billion people have exactly the same need or everybody has a different need out of those seven billion people. And how do we do product development and servicing customers with that demands centricity? I think that’s the essence upon scaling that we are going to try to explain. And that will automatically generate patterns around what these groups naturally are, because data and feedback loops tell you. In the case of diabetes, here’s the kind of metabolic conditions you actually have and the kind of digital therapeutic that’s gonna work for you and automatically figure that out for you. As opposed to having a supply centric…

Phil Libin: So it’s like, there’s not one or two types of diabetes. But there’s also not seven billion types. Maybe there’s twenty types.

Hemant Taneja: First of all, I think that’s what we’re going to find. But all of this doesn’t matter. That’s the whole point. Which I think with data and iteration, you will figure out how many there are. But more importantly, will understand you as an individual and figure out how best to treat you and not even have the pressure of figuring out how to lump into this. Cause the way you create these products is fundamentally changing. It’s not supply centric, as I say.

Ronda Scott: On the flip side, when it’s not a healthcare, it’s we’re talking about personalization outside of healthcare. You’re still not looking at it for an audience of one, right? Because we’ve put contend commerce online, you can create a product that, you know, maybe back in the day when you were selling it through malls or something like that or catalogs you might find, you know, hundreds of people who would be interested in that very unique product. But today, you can actively and easily find those groups online. So, I don’t think personalization is down until you’re creating a widget that’s unique for every single individual that might buy it. It’s you’re able to also on the other side of, on the other side of the equation find the exact audience that you need.

Hemant Taneja: And how do you draw those lines to optimize for maximum impact? You know, maximum delight, maximum satisfaction among the people who are, who are using it, right? So it isn’t, it isn’t individual, unique snowflakes, it’s I guess naturally occurring groups but groups that are defined based on what’s actually best for the world not what’s possible for the company.

Ronda Scott: I guess I’m not really thinking about it as, like the moral play there. But I’m thinking more of like, you know, a band puts out an album, right? And it used to be, you needed to rely on getting on the right radio station, you needed someone pick up that rolling stone, right? And find the picture and be compelled to go out and buy your album. And now you can find cohorts of people who are going to be predisposed to listening to your grunge death metal or whatever it is. And you know you can bring delight to that group of people because you’re able to, you’re able to discover them and find them and get to them in ways that you were never able to before.

Hemant Taneja: Yeah. It’s a customer centric point of view, which is how best to serve every single customer. And these natural groupings emerge. Just think about applying machine learning to data stats. It helps you identify natural learnings. That applies in how to think about drug discovery targets. In the case of how pharma is done, that applies to how you teach education and how somebody does personalized learning online. It will apply to every category and those natural groupings, identifying those and building products towards doing the best for each of those groupings. I think that’s really the right way to think about personalization in the unscaled era.

Phil Libin: I think a mistake that a lot of entrepreneurs make, a lot of product designers make, is they think that the best way to actually figure out what’s the right amount of personalization is to ask people. Is to like ask customers what they want. They kind of confuse personalization with preferences. And if personalization may or may not be bullshit, I’m kind of coming around to your way of thinking about drawing those lines correctly is important. But this idea that people actually know what they want, they can express the and they can customize their experience, you know, implicitly or explicitly to be the way they want it to be, that always seems far fetched to me.

Ronda Scott: They can but you lose the delight, right? I can go and customize the Audi that I want to buy right? And I down to, probably twenty different variables before I order. But it’s the delight of actually getting in the car and finding something that was like wow, that was really well designed. That’s exactly where my hand naturally hits, you know, that’s exactly the way the button and the UI should be. People have preferences and they’re valid but if everything’s built to spec, then where’s the delight?

Phil Libin: Yeah. And I guess for some products, it’s about the delight. For other products, people are just completely unqualified to actually say what’s good for them. Life or diabetes care, you wouldn’t necessarily expect people’s preferences to matter nearly as much as what is the reality of their

Ronda Scott: Behaviors.

Phil Libin: Yeah, behaviors, their metabolic condition and so on. So how does this work when we’ve gotten a world where we can identify groups of people that we make a product for this particular group of people, we draw the line in the particular way, we can have a high amp product that really serves those people well. How do we deal with everyone just flocking to the most desirable groups of people at the expense of the other ones?

Phil Libin: We talked about, for example, some of the student loan companies in previous episodes. Well now you can segment students before you couldn’t. You were, you were a graduate, everyone went into the same pool, everyone kind of paid off their student debts. Now you can use big data and social media, whatever, to figure out who are the best credit risks for students because they’ve got the most earning potential or something and you can give them fantastically good phen tech products to make their student debt repayments really good and its good for the students and its good for the companies and its a great business. Because you’ve sort of concentrated on this, on this specific group. But you’re kind of screwing everyone else, you kind of neglecting the groups that aren’t quite as, as profitable to work on. How do we deal with that?

Hemant Taneja: I have a belief that entrepreneurs will obliterate these, sort of, scaled ways of serving large populations. So when somebody goes and creates a student loan product that’s best for students coming out of great universities that are most likely to have great credit in the future than this lower risk product. I think somebody who relates to the other problem, this other cohort is not being served well, We’ll build another unscaled company. And so all of sudden what’s going to happen is, you’re going to go from this financial services infrastructure, where you’ve got these large banks that tried to serve everybody, qualifies poorly cause nobody’s really happy, into a bunch of these segments or groups where the products are designed specifically for them. And when you think about that in the context of insurance, it’s going to blow that industry up. You’re going to see all kinds of new insurance companies, we’re already seeing, that we’re seeing a lot of companies that are building next generation insurance products specific to these demographics or groups.

Phil Libin: So, what do users actually want? What do customers want? What, kind of, sets their preferences and what are we conditioning them towards?

Hemant Taneja: Yeah, I think that’s a great question. If you, if you think about the model T ford era, it was all driven by supply economics.

Phil Libin: Right.

Hemant Taneja: So, the reason Henry Ford famously said, you can have any color Model T you want, as long as it’s black, is because they had total supply chain to go build that cost effectively. And it was not cost effective to go create other colors. We got better and better at that, we learned how to create customization that mimicked a choice and came, got consumers as close to what they actually wanted as possible. But it wasn’t until the internet era, that we truly shifted from a supply economics based mindset of products to demand based economics, demand based mind set of products. But what do I mean by that? Well when you were building software products, whether it was games or applications, the cost, supply costs were essentially zero.

Hemant Taneja: And so all of a sudden those companies that were a digital native selling software and software based services, started learning from data from customers and building all kinds of per mentations for customers and its a potification of products become the norm and the digital era. Then as… platforms like Flextronic, FedEx and others that allows to bring that same mindset to the physical products, we have now brought that unscaling mindset to virtually everything. Including retail, including electronics manufacturing and so on. So I think we have gone from this supply oriented view of how to create choice for customers to a demand oriented view of to create choice for customers because of the existence of this platforms.

Phil Libin: Do you have any examples of products or services that you use, that you think are high quality, that are important, that are that ways because they’ve got just the right amount of personalization. Because they’ve kind of included you in just the right group of people.

Ronda Scott: Well I think, so, one comes to mind is, is Strava and especially the Strava integration with Trainer Road.

Phil Libin: Okay

Ronda Scott: So they use a bunch of data about you, the power you can put out on a bike, your heart rate, the distance you can go, how you climb and compare and contrast that with a bunch of other people that come up with personalized training and it’s a fantastic experiences.

Ronda Scott: They realized in order to baseline, you have to do what’s called a FTP test in Strava, right? And they realized using data from thousands and thousands of cyclists that they weren’t able to actually get, you know, they weren’t able to get the right FTP’s. They weren’t zeroing in on what FTP, and so FTP basically your threshold, functional threshold. And the bottom line is, they aggregated a bunch of data from cyclists all over to kind of really hone in on batches of cyclists. I am not a unique cyclist, I am a casual cyclist. But compare, using my personal data, my performance against these large data sets, they’re able to come up with a really effective training program for me. There’s some coaching that goes on in the training program. You’re able to actually kind of insert delight into a very painful rides via that way. And they think they actually do a really good job addressing different types of cohorts.

Hemant Taneja: And in terms of Strava example, I would say you would find working examples of effective groups in sectors where unscaling has been happening for a while. So if you look at books and Amazon, that is an example. If you look at YouTube you and get sucked into their sequencing of what they’re feeding you, that is designed, there’s a reason why people keep clicking and listening over and over again. I think if you think about how [inaudible 00:16:44] can get applied into financial health. So if you look at companies like Digit or Wealthfront, they are identifying you by groups saying, “Oh here’s a cohortive consumer that needs to learn how to save” and Digit goes after helping to manage their financial health. Wealthfront find folks that have hit their disposable income, they’re trying to figure out how to get the same kinds of opportunities that institutional investors have. And lets go mimic that for them.

Hemant Taneja: So you’re starting to see that happen now. That is the essence of why these sectors are not changing but at scale where we can relate to the examples on a daily basis is media, it’s in retail. You don’t know what you buy half the stuff on Amazon that’s presented to you. They have that grouping for them. So I think is happening at scale.

Phil Libin: Probably lots of examples in, in particular food and beverage that have kind of followed this trend. Where, you know, even like they used to be just Coke and Diet Coke.

Ronda Scott: Right.

Phil Libin: And now there’s like ten different Diet Cokes. But not, but Coca Cola also had, remember the thing we had at [inaudible 00:17:45] we had that machine where you could make your own. Seven billion combinations of different types of Coke that you can make.

Ronda Scott: We were supposed to get it. We never got it.

Phil Libin: You never got it but whatever. But those things, those things that it turns out no one wants. Turns out that no one actually wants to buy [inaudible 00:17:55] in their own soda.

Ronda Scott: No. You want to do it once or twice and then you’re done with it.

Phil Libin: But they do want, but they do, but they do prefer to have a choice of the eight different types.

Ronda Scott: Yeah. So I think the thing that’s easier now is because you can, you can think of a type of whether it’s coconut, like coconut water based or cambucha, you can think of a type of, a new type of beverage, right? And you can take it to market and you can find the people that you want, that, that you know, you could find your target audience for this particular kind of beverage. And unfortunately, Coke and Pepsi are still in there, what other, how other ways can we slice our brand and create something new and a new type of Coke, a new type of Pepsi. We’ll add some cherry to it, we’ll add some vanilla to it, whatever. They can’t innovate beyond that so a smaller person who’s saying, “Hey, we’re cambucha.” Seems like it’s taking off, people are doing shots of vinegar and the palate has changed and people are away from sweet and towards this. You can spin off very successful line and then acquired by one of those large companies because they’re not able to innovate on their own.

Phil Libin: Yeah and because of the unscaled economy, not have to figure out your own bottling plants and distribution and all that stuff. Just rent, rent those parts of the platforms as necessary.

Ronda Scott: Right. And those large companies are all renting parts of their platforms.

Phil Libin: Right

Ronda Scott: Some of them are taking interest in the companies that use their platform, some of them are not. But that is the way new things are coming to market. And, you know, you find your rabid fans, your rabid twenty-five thousand fans who want a grape flavored cambucha, not hard to find them.

Phil Libin: So an interesting common thread that I think runs throughout this discussion, is this idea that we’re shifting from, I think what you called in the book the atomic unit of supply, to the atomic unit of demand. This idea that categorizations were more important in the past based on what the company could actually profitably manufacture, distribute, whatever and now what’s important is figuring out what is the highest impact that you can have. And drawing the lines that way. It’s probably a mistake to think that this kind of personalization, this atomic unit of demand, results in there being seven billion versions of anything. Because there really aren’t.

Phil Libin: But however many versions there are, they are defined based on what is the highest purpose? What is the best possible version of this product for those people. And platforms have been created to let companies build just those products. And that exposes a lot of potential dangers, tying back to some of our previous episodes where we talked the minimal virtualist product. They [inaudible 00:20:23] How do we know that this kind of targeting, even though it may be surveying one group well, isn’t leaving other groups behind. And whose job is it to do that? Obviously, it’d be better if that was the job of the company’s entrepreneurs but it sometimes isn’t and sometimes the authority’s, the government has to get involved and that’s what we will talk about next time. What is the role of laws and governments and regulations in making this view unscaled world a better place to live.

Phil Libin: Thanks Ronda and Hemant. See you guys soon. That was a lot of fun.

Ronda Scott: Thanks