About the talk
The future of software products is private, personalized, and on-demand. In this talk, Gabe will cover the lessons learned from the launch of private personalization startup Canopy's first consumer app, Tonic and how they can be applied to making future-proof decisions when designing products in 2021 and beyond.
Gabe is a Cuban designer based in New York City, currently serving as the Head of Design at Patreon, where he’s building a new recommendation architecture for a better internet. Prior to that, he managed a design and research team at Alphabet's Jigsaw, an incubator focused on building technology-agnostic products for at-risk communities, Google Daydream and Facebook, where he led design on a variety of products including Pages, Photos, and Virtual Reality. Gabe is energized by exploring new mediums and industries — from social good to emerging technologies and smart hardware — to learn how design and technology can be leveraged to enact positive behavior change.View the profile
My name is Rodger. I am really excited to be here today. I'm a designer and I'm currently working on a tree house like to tell you a bit about myself. I said the last 10 years working with some of the biggest tech companies out there in 2012. I join a small startup in automatic where we worked on the future of. Can I take cars? We built a device that I gave you such a peek into a car's computer and Empower them to use the data to make driving more safe, fuel efficient and fun.
After that, I spent four years at Facebook, where we explore the future of social storytelling, and media, in virtual reality. I did not Google. I spent 2 years working on a team that focus on Building Technology for social good, especially tools for fighting and its information and censorship. For today stock. I'd like to try to take us to a simpler time. A time when I didn't own a single face mask and I didn't think twice about going outside. The year was 2019 and a company called canopy
launch a product called tonic. Spell canopy with the small startup based in dumbo New York City. If you don't know Dumbo, is that place in New York where every tourist take an Instagram photo? Anyway, these Founders were responsible for the feature in Spotify called discover weekly, which automatically generates playlist that recommends you new artist based on your listening activity. If you've ever used to cover weekly before, you've probably been delighted by his recommendations, they feel uniquely curated and personal to you.
Spotify canopies co-founders were struck by a problem. They wanted to solve. The problem is that people are growing increasingly distrustful of Technology. Privacy leaks in a central threats to democracy cause people to think twice about downloading your app. And screen time is higher than ever, and people find themselves addicted and feeling lousy about using ecology. So we have four cells. Like, how do we get here? Like what happened in the last 10-15 years? That we now feel this way about using technology. Well,
we think that it all started out with the web to for now about 15 years ago. And for those of you too young, to remember the Web 2.0, what's the stadium that the internet for people to create and share their own content with others? This paved, the way to online communities for the day of a lot of creativity and self-expression. That's great. This friend was accelerated by the introduction of the iPhone and the rights of smartphones with the Web 2.0 and our hands, we could not post anything from anywhere. This led to an exponential growth in the
amount of user-generated content. Sovaro moments were created and the internet. As we know it start to take shape. Then these platforms found themselves with more content than any person could ever consume. So naturally companies turn to personalization algorithms to show, only the interesting bits do the users at the time. The only way that you had to do this was by collecting a bunch of personal data and uploading it to their servers, to train their algorithms. And this was the beginning of the cloud. Then a business model was developed by
companies package and resold that data. So both so advertisers could better targets across platforms. I need to measure the success of these models companies begin begin optimizing for engagement. The more people interact with the products. The more valuable. They thought they were So the cloud business models and success metrics. Let's do a misalignment of incentives between product makers and their users, as a result, every website or app. Now ask for your email, your phone number, or give me your credit card in exchange of a marginally
better user experience. A company that uses data to create an assumption of who we are on the web and traffic it without our explicit content, to top it all. We can't change or Escape that assumption. It's as though, every time we pop into a local Bodega, they check and review or purchase history. This has become the price of admission for the internet. Not when we started kind of bee, we wanted to make a different kind of software. We wanted to make software that was Private by default wanted to provide the life of discoveries like
this cover weekly, but this time want to do it privately without collecting personal data. We also wanted to create software that gave users unprecedented control over their experience. I finally, you wanted to make software that could nurture at Mindful and sustainable consumption. It's our first goal was to create a consumer experience that showcase canopies unique approach to building products. So we assembled a small cross-functional team to do just that product, engineering and editorial to craft our first consumer
product. We knew the team could do audio recommendations really well, so, we tried stepping outside of comfort zone. I built a product that recommended you things to read, and we call that product tonic. Atomic was a really simple app everyday, the app combine machine learning and human editors to scour, the web and find the most interesting articles to read Trader, Joe's for you. What a tonic to equip you to tell the best stories at a dinner party and the app activity and fine-tune the recommendations as you go
now that the cat is there no accounts and there did you don't have to login everything lives on the device and you have full control of your data? So I want to watch you guys through three other user experience challenges that we encountered as a rebuilding this following principles. The first is how to explain to folks how this works. Privately recommends content without requiring any user data to be right by the server. Talk to do, this will use a technique called differential privacy.
Now differential privacy is a system for sharing information about a dataset by describing the patterns of a group but withholding information about the individuals that I don't know what that means either. But basically, this is what we did. We look at the activity in your app. So what you choose to tap on how long you spend on the nautical Etc that we create a profile of your interest. Can we add noise to the profile and make it mathematically impossible to decipher Who You Are? I will send that to the server where we collect New pieces of content
recommended to you. That was like the ones that we think you're really going to love. And then we download those items into your phone acts as a filter to show only the relevant bits. To you, when you open the app? So the person's identity is encrypted on the phone and the server never knows who they're recommending counting to. They just know why they like this is important because at any point, the user can walk away and remove their identity from the equation and leave no Footprints behind. So we wanted to explain this process in the app. When you first download
it. We use diagrams and as fractions to explain what's happening behind the scenes at user research staff and we didn't want to rely exclusively in our information to decide whether the solution was the right call. So we start a beta program and gather feedback from over a thousand people. We send out user surveys and conducted regular user interviews. Another way We Gather feedback was by adding a feat that button right at the app. So anytime you could
tap that button on the top left. I think it's the message. I will go straight into a slack channel. So that was a really helpful too old to get on filter access to what users we're really thinking. But was working, what wasn't even feature request and everything was done anonymously. Go through all this feedback. We got, there are three key pieces of inside, one people don't read. So when you download the last thing, you want to do is read a paragraph of taxes. Are you the app works? Instead? We found the most successful approach to spending enough mechanics is to show
how it works better than The second is that trust is lost not earned. No matter how deep you went on explaining differential privacy. We found folks who have little patience to Think Through how it work in the background. Instead. They were willing to take out to take a leap of faith until their trust was broken by the app. And the final one is that people like people and we found that the most successful approach was not showing a diagram of the process and showing faces and people conveying a Feeling. So we went from this flow
when you first downloaded the app to this film will use concise language and culture imagery to convey a simple Valley prop, and desire feeling we wanted to achieve by using a wrap. What's on boarded users? Could start using the app, but remember there was no need for users to create an account. So our next challenge was, how do we know why you're going to like the first time you open the app? We have no idea who you are. Like other apps will dump you into a float for you to choose categories in interest
work. Instead we showed you search the sample of the content. They could finally have the app. They could read the article right there. The first task was to select five articles that they wanted to read. But adding it to the bookmarks. This help us accomplish two goals and one assist people to bookmarking, which is the main option to be performing were using tonic and two. We avoided the ghost town problem and make you some pictures that you should always have something to read even on day one. So based on their selections, we assign users to one
of 11 archetypes which would give us a starting point for recommendations. These archetypes were based on the zodiac signs. The kinds of us were like, none of them are wrong. That you can see yourselves as part of all of them. And this example, I am imaginative meditative and blues for phone that people really enjoy this this opportunity for the app to show them. And then for them to feel seen by the product. So how do you teach people the basics about your app will first human language and imagery? Can you take advantage of your first impression to teach
the basic mechanics and then you give users immediate rewards for their investment. Okay, the Second Challenge was how to create this idea of mine for consumption and avoid making you feel overwhelmed. When you first entered, the app Wilson, wanted to avoid the feeling of full moon. Like, always have constant for you to save time whenever you were ready. Your goal was to give people a reason to come back to the app without relying on engagement mining as both tactics.
So we decided to apply the old adage of using quality over quantity. So we created a constraint constraints and instead of juice rolling through an infinite feed of content like every other have out there. We bought the best content in 25 recommendations everyday. So this made the actual fresh and gave you a reason to come back every day without having to rely on growth hacks or annoying notifications. Each bundle is uniquely tailored to you based on what the app now is that you can love there in a minute. There are
four articles. We know you've been alone for sure, base of the art of the interest that we gather from your activity. But there's also one wild card and that is the best from a topic that you may not be typical exposed to. And that's how we try to delight and surprise people who are recommendations by first thing you out of your bubble and introducing you to be familiar with or a publication. You might not know. So now that we know the apps mechanic, let's talk about the app structure. So we knew we wanted to have a Home
tab where you can see your daily recommendations and take the ones that you like to read. If you don't have time to read that article, you can bookmark it for later. And every article you've ever read or rated gets saved into an activity tab. So that you have a clear sense of what the app things you like. So in other words, we have a temporal structure for the app activity. Is everything that happened in the past home is everything that you can read in the present and the cars is where your store. When do you want to read the future? Just look at home.
We wanted to make the Outfield simple and intuitive as we design the Home tab, to show you everything you needed to see at a glance. Purse each recommendation is displayed as a beautiful card for the content can shine and feel like something worthy of your time. When you read a combination, it moves away from the recommended stack heading to the last open section so you can easily pick up where you left off. Allen below. It, there's the rest of the articles that you saved for later. So like this the Home tab is showing you. A quick list of everything. You can read every time you open the
app. If you want to explore all your bookmarks, you can tap an icon in the top-right and if you want to understand your recommendations activity, tab can be accessed from the top left. So here you can see the side where the car is at this way on the top and you can decide whether you want to read one of your recommendations or one of your own open tabs. If you click the card, you enter a customer service browser that we design that I get was private and there was no trace of your identity as you browse the internet. After you're done, reading the
article with a simple gesture. You can dismiss it and it moves from your recommended stack into the last open section. About if you want to save the next car for later, you just stopped at bookworks Icon. I didn't move to your own open Bookmarks section. So how do you introduce my phone consumption into your app? You lean on quality over quantity, representing content to users. You find Opportunities to introduce surprise at the light and then you keep it navigation. Really simple and intuitive meaning users where they are.
Okay, so the last few extra dollars, I like to talk about today is how to give control over their recommendations to users. So we've all been there when YouTube out of place. A video that we don't really like or the next song in the playlist. Just kind of killed the vibe. We wanted to avoid that one of the, the control features to be at class at 1st Class citizen in the oven. The feeling that it was something a kid, getting a mixtape from your buddy, you know, some of the songs in that mixtape maybe bangers others may not. So we wanted to go back and forth between the user
receiving that recommendation and the system provided those recommendations. First, we looked at how companies implemented control features across different products music, Tik Tok for video and we noticed a few other key takeaways here. So you just sent to have either a binary option like a thumbs-up or thumbs-down or favoring something Harding. Something. Just give me like two limiting. Imitation is just okay, and the users has no way of sending that signal to the system. The other option is to
use the rating Spectrum, like a five-star rating system. Will you see there's a lot of Yelp and other reviews? Put the brush a little confusing, like our two stars awful or kind of good or, you know, like how do you know where the bar is that? I got the third inside of a father's that negative feedback is really high expectations. When you say you don't like something if you see something else like it all the sudden you start to lose trust in the recommendation engine. So we decided
to implement a spectrum of positive feedback. That looks something like this. We have four degrees of positive feedback on the right and one option to dismiss the recommendation altogether if it just didn't hit the mark. So we went from all right. Good to great love and will use this imagery to represent. A prototype this version where you can long-press on any article that you see and then this, you I would come up and you were still alive, the lady that you were given. I got we tested as we are users and found that this warning was still a little bit
ambiguous, but the attraction feel really good. That night I went home and I was playing Zelda breath of the wild which is an amazing game and I noticed detail. And it's every time the game has used a heart / 4 quadrants to indicate your health. It is a really simple way of showing your health in the game at that moment. My mind was focus on rating systems. And how do we create a positive rating system when we refer to as the Zelda hearts? And that you, I
just indicates how much you love something. So you have on the left, if you don't love it at all. You can dismiss it. And then from there, you have different degrees of love and it seems pretty simple, but it took us a while to get there or something like this. Anytime you are seeing an article you've read, we show a heart indicating how much we think you enjoyed it. At any point. You can correct that rating and teach the algorithm what you like does long press on the article and drag your thumb to the railing. Then the heart update to reflect your desired
rating or feels like they have control over the recommendations are there getting and they are an active participants. They can send a signal back to the system. I supposed to know how you get recommendations on Netflix or Facebook or Twitter. Where is just kind of a black box? Three key takeaways yet. Having an activity feed with full transparency on what the algorithm thinks you. Like. It's something users, really appreciate. And once you have the visibility for that, they can correct the recommendations to make sure they're just right. And then we found that positive feedback
is much more effective at gauging so much interest in something. So that's how we look at how we apply design principles to create an experience. That was Private by default controllable in my phone. The app was really well received by our users and the Press here, a few of the App Store reviews that we received and her average rating was 4.8. That we took all these principles and we open doors them. We called them sensible design, our hope is that the entire design Community can
piggyback of this starting point and see if they could apply similar Concepts into the products are working on. Does it mention before we launched the product in 2019 shortly after its launch? Kind of bee was acquired by CNN to leverage the team. That the ecology design principles towards a larger kind of news pet. Earlier this year is a company that uses is also an underdog like patreon tries to use a new business model for creators out there to fun creativity and art
on the internet as you move away from where we are now, which is very much at base into something. That is more direct with his its fence. Patreon exist, because when creators are paid, they can create more amazing things, things that Inspire us. Teach us Challenger. Things that make us laugh. Really easy for creators to get paid really old idea Mozart Shakespeare DaVinci. They all have patrons mostly Aristocrats who paid them to create so they could enjoy their works and brag to their friends about how cruel they are for supporting creators creators of every time. Podcasters
YouTubers musicians Riders, allow their fans to become patrons for members patrons at the monthly subscription style payment for the level of membership. They want, like 45 bucks a month to get early, access to content, 10 bucks, a month extra videos, 20 bucks a month behind the scenes, stuff. You name it. This provides us with a sustainable income while retaining Creative control and allows fans to connect with them on a whole new level of freedom is. I can't tell you how important
level of freedom is to run their business, their way. They're doing what they love and they're being paid to do it by the people who love their work. Most of your professional Creator. Start a page, you give your fans the opportunity to become patient. I can help you, do the best work of your life. So cord to this idea, is that where this inflection point, where? The way we understand the internet and what happens out there is changing recommendation systems. And what what they require from us as
users are changing and the way we create content that way, we are rewarded for that content is changing. So the reason why I can't come here and chat about it is to see how design can play a role in that, how we can use design to validate some of these ideas. Thought I would be remiss not to mention. That was. So if you want to make similar experiences on a larger scale, please reach out to other that. Thank you. That's all. Even up for Gabriel, K School. We had a few questions here.
So Taylor would like to know, can you elaborate on how the select five stories onboarding solve the ghost town problem? Any other ideas? You explore to solve the ghost town, every time you start a new time when you create a Facebook account, you have. No friends. So its, if you have no friends on Facebook, you're pretty bad experience. And products that are like, kick-started. Our thinking was that by presenting people with articles, right away. And having them going to create a stash of content that they could consume, they would never see
an empty feed that they haven't kind of empty themselves. But I was going to say is our kind of approach to kick-starting you and I'm getting you in the in the process of using the app. Awesome. Justin wants to know how did you address the quote? I need to see it to believe it quote and quote crowd, that knows. Well camouflaged data can always be monetized and can and has an origin source. Yeah, there's a lot of cat skepticism on privacy because we've just been burned for so
long. One thing we did is that, you know, we made all this. Information readily available by car where written in human language. You could read, you need to be a lawyer to understand it there. And so if people want to learn about it, hopefully you understand it. And then we built going to say Gates where, you know, we wouldn't be able to share people. Even. If you wanted to live, like our CEO left and a new one would come in and had a different vision for the product. We built in
safeguards on the product, so that that wouldn't be possible. So, our promise to the user was not that we, your pinky, promise that we're not going to sell today. I promise, is that we can't do that. How would you recommend people starting with the conversation about flipping their model to private by default? Yeah, I think it depends on your organization. I think there is there are alternatives to achieve in the same outcomes. A lot of consumer products are traits of
attention, or engagement and much healthier metric to follow. So if you're part of an organization that is following engagement as as the demetric to buy wish you define product recess for not have a star, they're like Define tweak the way you define success within your, your product. And then I just provide alternatives to what we all take for granted. And they're already many products out there. That apply similar principles that are equally if not more
delightful than the alternative. Is that we all use Every time you use it as just come outside for me on your data and they're already Alternatives out there. There's a brave browser, there's doc.gov and they're getting better and better and better as you try to proposal. Awesome. Thomas asked, how did you decide to put a default automatic rating on articles users had read, rather than having no rating until the user Reddit. So every app that recommends content out
there, Justice, they all do it. You do a Facebook, Twitter, or YouTube? They all do it. They'll assign the rating to everything. You scroll past. The difference is that they don't show it to you. They all have a model of like this content looks at her way and has this degree of affinity to the user's interests. That is how you build a recommendation engine. I think our Innovation that was just showing people in the way, they could understand what we think they like, and then provided the tools to correct that which is not something you can do
easily and other platforms today. Okay. Awesome. Anna asked, how did you discover the phenomenon where negative reviews create unrealistic recommendation? Expectations. User interviews have anything fancy? Try to find people that were in a representative of a broad enough user bases and we try to extrapolate scenes and we noticed that very consistently when someone that had an encounter, something negative removed from their argue, with make identity. They felt really strongly
about it when the app was not able to meet that expectation. Then that starts with a road to trust, our goal was above all to keep us. So we just Did a lot of interviews as and kind of user information from that sounds like that was a very core principles to it. Well, that's all the questions. Let's give it up for a Gabriel again. Thank you.
Buy this talk
Buy this video
Our other topics
With ConferenceCast.tv, you get access to our library of the world's best conference talks.