Imagine a major intersection where all the innovation taking place in data analytics and all the advances in hardware meet. It would look a lot like Monica Rogati’s job at Jawbone. As VP of Data, she built a world-class team of scientists and engineers who pushed on the boundaries of wearables, data and the Internet of Things. Today, she spends her time advising multiple companies that want to make the most of their data.
If anyone’s native to the field, it’s her. But that doesn’t obscure the fact that many companies are just learning how to harness data to build more compelling products. Enrollment in machine learning classes and the like may be on the ascent, but shaping that knowledge into simple, elegant solutions for the masses is beyond the scope of most degrees. Of course, none of this keeps consumers from demanding more sophisticated data products than ever before.
“About 10 years ago, a video went viral on YouTube showing a toddler holding a paper magazine and trying to use it like an iPad, swiping and pinching to zoom, and it didn’t work so she just looked at it thinking, ‘It must be broken,’” says Rogati. “It was obvious evidence of a new generation of digital natives. Today we’re witnessing a new revolution of data natives who expect their world to be ‘smart’ and seamlessly adapt to what they want.”
There’s incredible appetite for products that will anticipate every need and want. In this exclusive article, Rogati shares how companies can harbor their resources and take a new approach to rise to this occasion.
Understanding the ‘Data Native’
Being a data native goes beyond tech savvy or digital engagement. It’s not just that you like your information served up on screens or being comfortable with the tools. The digital revolution happened when the balance tipped in favor of people growing up surrounded and shaped by computers and the Internet. Rogati believes we’re in the middle of a similar yet separate revolution that’s entirely about data.
“A data native is someone who expects their world to not just be digital, but to be smart and to adjust immediately to their taste and habits,” she says. “For example, a magazine should not only be digital and interactive — it should be personalized. It should tell you what you need to know based on your interests, location, preferences. The expectations have shifted.”
A digital native may be comfortable programming their own thermostat, but a data native expects the thermostat to program itself. Digital natives might use the Starbucks app to order their morning coffee but data natives want the Starbucks app to order their favorite drink at the right time automatically. And it doesn’t stop there. They want that app to be context aware so it knows when to order the same-old same-old, and when to mix it up so they can try something new.
Digital natives were concerned about what they could do with technology. Now data natives want to know what technology can do for them.
This attitude has been hastened by the explosion in network devices. According to Cisco’s most recent study, the number of networked devices will be triple the global human population by 2019. McKinsey says the Internet of Things is predicted to crack open the economy with a new market worth $6.2 trillion by 2025. Just three years ago, Home Depot offered 100 different smart home devices. Now it stocks well over 600.
The wristbands Rogati worked on at Jawbone were designed to ride this wave, not only by helping people dutifully quantify themselves and motivating them, but also turning on their coffee makers when they wake up, switching on the AC and turning lights off when they fall asleep.
All that said, there’s plenty of evidence that products are not keeping pace with rising expectations. “Your GPS hasn’t learned your preferred routes. It still shows restaurants that are thousands of miles away. Ads still aren’t perfect at knowing who you are and what you want. The other day one was telling me to get my degree online in only 7 days, even though it knew my PhD took 7 years.” says Rogati. “Then there’s your phone’s autocorrect turning things like ‘LOL LOL LOL’ into ‘lollipop oligopolistic’, a phrase that I’m pretty sure nobody had uttered before. The data to make it all smarter has been collected, but we’re still not there yet.”
It’s an interesting quagmire. Traditionally, new technology has expanded people’s notion of what’s possible. Now the crowd is already dreaming of products no one knows how to build yet.
Data Products: How to Catch up with a ‘Smart’ World
Start with the right working definition: “Data products provide context and personalization using data collected from you, others and the world,” says Rogati. The key to building really good ones is shortening the feedback loop that allows the product to continually take in a lot of data really fast.
Giving people recommendations to interact with and learning from their choices is one of the clearest examples. The more recommendations you serve, the smarter you can get by recording and reacting to people’s preferences. This is why 35% of Amazon’s revenue comes from recommendations — and why 75% of Netflix content is consumed based on recommendations.
At Jawbone, Rogati and her team were tasked with using and serving data to help wristband buyers move more, sleep longer and eat healthier. They did that by automatically detecting and classifying workouts, making food suggestions, and helping the device serve as a “smart coach” that challenges you to drink 3 more glasses of water than usual, take 1,000 extra steps, or go to bed 10 minutes earlier.
“That’s on the individual level, but when we take things to scale it gets really interesting,” says Rogati. “If enough people are all doing this at once, we can sense whether people respond to a more encouraging tone in the app or more of a drill sergeant tone. We can see how people are motivated by cooperating or sharing with others — like letting people agree to run 100 miles together or competing against each other.”
If we do this right, all the data science is behind the scenes. It’s not providing more charts and graphs, it’s about delivering a deeply personal experience.
So where does this start?
Data Products Start with Data
Before you can run any analysis, build a recommender system, or start training a machine learning model, you need numbers to dissect. The goal shouldn’t just be to collect massive amounts of data, but more so a wide variety, says Rogati. This means you should instrument your app to log as much as you possibly can, because some data can be lost forever. For example, “It’s not enough to log that a user clicked on a product recommendation — you have to know what else was being recommended, the order of each item, and position on the screen, ” says Rogati. “You need to record versions of your algorithms, parameters, strings that are exposed to the users because all of that could change in a couple months.”
Reliable Data Flow
The best instrumentation and the best machine learning algorithms don’t help if you don’t have a reliable data flow. “If you drop events or if your infrastructure isn’t sufficiently fault tolerant or scalable, you’re looking at the wrong numbers,” she says. That’s true of dashboards and internal analytics — but it’s even more important for data products.
The moment you show data back to users, the consequences of breaking your data flow are severe. It’s missed sales because you didn’t make good recommendations. It’s your app crashing because a queue was backed up. It’s losing the user’s trust.
Clean Data — and Fast Iterations
Many articles have been written about the importance of data wrangling and cleanup. “Data scientists spend 80% of their time cleaning data” is heard often enough that it inspired its own parody (‘Data scientists spend 80% of their time complaining about cleaning data’). Rogati, however, wants to see data scientists embrace it:
Data prep is not beneath you. Good data prep is detective work; it takes intuition, experience, ingenuity and pragmatism.
“The effort is well worth it — it can have a bigger impact on your results than your choice of algorithms,” she says.
The real challenge is that you can’t possibly anticipate all the different ways your data is wrong — which makes fast iteration (both at the low level data wrangling and at shipping products) absolutely imperative.
The human in the loop
For both data products and traditional analytics, user experience is a crucial piece of the puzzle. Even if your data is clean and organized on the backend, you have to have a reasonable user interface. For reporting and analytics, this means a dashboard that’s not too painful or discouraging to explore. In order to really dive into the numbers you’ve collected and draw a diversity of ideas and creative insights from your team, everyone has to feel comfortable with the way they are diving into the data and getting results out. A poor interface is a strong disincentive.
For data products, the user experience is a gating factor. “The user interaction needs to be smooth, intuitive and robust because it’s being handed to people who are going to misinterpret it, click on the ‘wrong’ things, or have different expectations.”
Great user experience and great data are what make products feel smart.
Those are the two sides of the data product coin and they reinforce each other to create a virtuous feedback cycle. What does this look like in practice? You want easy to use software and hardware that can seamlessly integrate into people’s lives so that it generates higher volume and better quality data.
Rogati provides an apt example from her experience: “Say you’re logging your meals in your Jawbone mobile app — quick autocomplete becomes really important because it helps people get the job done faster. The faster people can log their info, the more they’ll do it and the more quality, consistent data you’ll end up with. You’ll avoid misspellings or using thousands of different version for the same concept if you can autocomplete people’s thoughts right away.”
Harkening back to that virtuous cycle, more and better data is what, in turn, makes autocomplete (and data products in general) work better and faster — and makes it feel smooth and smart to users. “This is why the best data products need reliable data flows, fast iteration, and tight, implicit feedback loops — all in the service of a better user experience that feels truly ‘smart’.”