The dilemma of analytics for the internet of things
On the occasion of Snap Inc’s pivot from photo-sharing via Snapchat to the camera hardware of Spectacles, CEO Evan Spiegel deemed that we were living in a post-rectangle world.
At first, I thought the declaration had come too soon. After all, most of the internet still lives in rectangles. But upon further consideration, I figured my bias was too backward-looking. Platform shifts look like this. In 2009, for instance, Mixpanel reached a similar conclusion: The party line had been that user actions happened on desktop and laptop computers as page views and link clicks, when in actuality users were spending more and more time tapping on apps on their mobile devices.
Nobody had developed an analytics solution that accounted for this paradigm shift, however. As a result, product managers were using out-of-date, generic web metrics, like pageviews, instead of more dynamic actions, like “Song Played” or “Room Booked.”
Tracking these events presented greater tracking challenge for builders. It forced us to get more creative with the questions product managers posed to their data. But it ultimately was the right thing to do for products. Now we’re at the verge of another paradigm shift, and it’s time to follow Spiegel out of the rectangle.
There’s a lot of bathwater here, though, so hold on tight to the baby. Web and mobile will still play essential roles in the internet-enabled world, but there’s reason to believe the market is pushing consumers toward products that look increasingly like Snap’s Spectacles, or the Apple Watch, or the Google Home. Non-rectangles like these will ingest more data than any standard application would.
And each of these products gatecrashed tech headlines in 2016, which confirmed something: the Internet of Things is no longer a two-years-out, sci-fi fantasy. The internet-enabled world is already here, and it’s only growing. The number of connected devices, currently about 10 billion, will grow to 34 billion by 2020.
But for all things that seem like far-fetched fantasy, there are real technical questions to address. The first part of making this dream a reality? Mastering the data.
The volume problem
IoT is the biggest data challenge facing product folks today. Connected devices represent a very literal information overload, and the volume they generate could have PMs rethinking how they approach data.
The type of data IoT solutions rely on is time-series and sensor-based, as opposed to the more static forms of data that sites and apps rely on. It’s not enough to know that a FitBit or a Nest Thermostat was on and active. PMs want to know the how, when, and, if possible, why of actions.
But this creates a lot of real-time granularity in the data. Every action sent from a connected device has to be in proper sequence and bring a host of other data along with it. Because of this, most IoT devices will send far more data than the average technical product. They already are.
Gartner reports, “IoT deployments will generate large quantities of data that need to be processed and analyzed in real time.” Self-driving vehicles alone could generate 2 petabytes of data per car yearly, or 2 million gigabytes. But that’s nothing compared to the entire ecosystem. ABI Research hypothesizes that by 2020 data volumes across connected devices will hit 1.6 zettabytes, or roughly 1.6 trillion gigabytes.
That’s a lot of noise. It would be easy for a product manager on the Spectacles project to look at the sales numbers behind the release, shrug, and consider the product a job well done without ever taking a peek beneath the hood to see what’s actually happening.
But the opportunity in coupling product analytics with IoT devices is too great.
The difference with analytics
So, why should product managers bother scaling Mt. Data?
According to a 2016 study by MIT Sloan Management Review, companies with strong analytic capabilities are three times more likely to get value from the Internet of Things than those without. From that same study, IoT providers report that, more than talent or security, the number one priority for improvement in their organizations is analytics capability.
Neither people nor secrecy will be the competitive differentiator in the IoT arms race. Data-driven insights will be.
But for PMs to build out a strong analytics capability, they will need to invite all of the baggage that comes with IoT data. Time-series data will be a necessity for gathering valuable insights, which means the infrastructural problems accompany it will be too.
In building an analytics solution, buying one, or choosing some kind of hybrid, product managers will need an infrastructure that can ingest IoT data accordingly, and also an interface that allows them to segment devices for valuable insights.
In addition to the data defining actions as users interact with connected devices, there’s all the incidental data that comes along with IoT. Many such devices are always logging sensor data: Things like temperature, telemetry, and speed. The post-rectangle world entails a whole host of properties that computers and mobile devices don’t touch on.
When futurists talk about IoT, we often leap to thinking of homes and cars acting autonomously, but that kind of innovation simply isn’t possible without a strong analytics layer. Analytics isn’t just about humans deriving insights. It’s also the competency by which machines will make decisions.
A popular thought experiment of IoT is the refrigerator that is able to detect when milk is low and order a new carton. While impressive, such a use case is not exactly smart. But with machine learning, an IoT device could take this another step. Instead of simply asking “Is milk low?” the fridge could interpret its owner’s usage patterns and order cartons based on that cadence.
Our IoT dreams never get off the ground, however, without a strong analytics capability. That’s why it’s important for product leaders to get out of the visionary headspace and into the data, even if the numbers seem costly and hard to interpret at first. Scaling IoT products is going to be more expensive and noisy upfront than any product that came before it. That’s why it’s essential not to cower where the data is concerned, but to invest heavily and get analyzing.