Making Data Meaningful
in a Connected World
by Lee Gruenfeld
Longer ago than I care to remember, I designed a reservation system for a cruise line. One of the things I was proudest of was the extraordinary amount of data we were able to present in reports to management. We could tell them everything they could possibly want to know about cabin inventory, which of their marketing programs was having the most impact, how effectively they were ordering and using the food loaded in each port, even how fairly the waitlist was being managed. It was a veritable firehose of information.
And trying to use it was like drinking from a firehose. I still remember the head of operations sitting in his office surrounded by stacks and stacks of printouts, an alarmed, stunned look on his face. “Alarmed” because he had all of this terrific information that he’d never had before, yet he was paralyzed and couldn’t articulate why. He simply couldn’t figure out what he was supposed to do with all of it, but he was sure he was supposed to do something.
That was the exact moment that I finally understood the difference between information and usable information. I took another two weeks to meet with various department heads, both on-board the ships and in the head office, to find out what kinds of information they needed to make decisions. It turned out to fall into two categories: exceptions and insight.
“Exceptions” were just that: things out of the norm that needed attention. They didn’t need to scan the entire waitlist of thousands of names; they needed to know when passengers were being cleared out of order, and why. They didn’t need to know the details of every food order; they needed to know when perishables were being dumped because they’d reached their expiration dates without having been consumed. They didn’t need to hear about ordinary re-fuelings; they needed to know when abnormally large amounts of fuel were being taken on because captains weren’t operating the ships efficiently.
“Insight” was about combining disparate data into actionable information. Passenger surveys are the lifeblood of the cruise industry, but perusing thousands of detailed survey responses is of questionable utility. Drawing correlations is what matters. That’s how this cruise line discovered that the number one determinant of how passengers felt about their cruise was how much they liked the food.
My little epiphany was, of course, not a novel discovery, however striking it was to me at the time. Exception reporting is an old-school concept now, and “actionable insight” is practically a cliché
But added to the current recipe for information usability is a relatively new ingredient: Context.
In the Internet of Things, and particularly in the sub-domain called the smart home, there is so much information flowing in all directions that just dealing with exceptions or trying to derive insights are no longer sufficient for the realization of all the potential value that exists. True utility is dependent on devices and systems being acutely aware of where they are, where you are, what else is going on that might affect what should happen next, and knowing what happened before when a similar situation existed.
That’s called contextual intelligence, and the best way to describe it is with an illustration. Here’s how I want my contextually aware system to “think:”
The garage door just opened. Based on their smartphone locations, the husband and his wife are in the car and leaving. We should save electricity by turning down the air conditioning, but it’s 118 degrees outside [should have mentioned I live in a desert] so I don’t want to turn it off completely. I’ll raise it to 85, and will also make sure all the lights are off, including the outside lights because it’s the middle of the day, and I’ll set the alarms. I’ll also forward the land line to one of their cell phones. Later in the day, geofencing tells me they’re nearing home. When they’re ten minutes out I’ll crank the temperature down to 77. The garage door is opening and smartphone locations tell me it’s certainly the homeowners, so I’ll disarm the alarm and unforward the phone. Now they’re in the house and the husband goes upstairs while the wife stays downstairs. In ten minutes I’ll raise the downstairs temperature by two degrees because that’s what the wife always does. And if she walks upstairs, I’ll raise the temperature up there as well because I figured out long ago who wears the pants in this family.
As you might imagine, it takes a lot of intelligence to pull off a scenario like that, and the more you think about it, the more you realize that it’s far more complicated than it might seem at first blush. The system figured out that the homeowners like an initial blast of cold in the house but then turn up the temp when they get used to it. But it also knows that, if they’ve just been out on a bike ride – detectable by their smartphone-reported average 15-20 mph speed over a route they’ve done many times before – they want the temp left down low for a longer period of time. And just because the husband turned it up himself earlier than that once or twice doesn’t necessarily mean that anything needs to change in the default scene.
The burden increases exponentially depending on how many people live in the house and how variable and conflicting their behaviors are. A lot of very serious A.I. is needed to make this work, and it needs to utilize the feedback-loop mechanisms of machine learning, including “guessing and assessing,” incorporating exceptions, and detecting and handling ambiguity and inconsistencies. The biological entropy known as human psychology is the trickiest part of the equation, but taking it into account is what completes the fully context-aware jigsaw puzzle. The “state of the user” is every bit as critical as the state of the devices, and represents a far more daunting problem.
The same concepts that apply to using home automation also apply to the task of supporting people who use it. It’s becoming fairly well-understood that difficulties in installing and configuring smart home devices and systems is a key impediment to adoption. The approaches of the past won’t work; the new paradigm needs to rely on context.
Again, an example. (Full disclosure: This comes from work that Support.com is currently doing in this area.) A homeowner is having trouble adjusting her connected thermostat. She picks up her smartphone and goes to the app she uses to control all the devices in her home. The system has already detected that there’s something wrong with the thermostat, and has flagged it, indicating that the problem is local to the device and not an issue with the overall home ecosystem.
She presses a HELP button embedded in the app and, instead of making her answer a bunch of questions to diagnose the problem, the app immediately takes her to a series of steps telling her how to change the batteries. Unbeknownst to her, the cloud-based system behind her home automation configuration has figured out that the batteries have been losing power over the past several days, and the unit finally went offline that morning. The system was therefore able to figure out, based on those contextual clues, that the batteries had died and suggested the most likely solution.
Without ever leaving the app, she follows the recommended steps, replaces the batteries, but the thermostat still won’t power up. She then requests help from a live agent.
Now, at this point, traditionally, the agent would ask her to described the problem and then probably drag her through all the steps she’s already tried. If he’s lucky, she won’t hang up and ship the thermostat back to whomever she bought it from, but this is not going to be a happy consumer.
If the agent is instead using contextually aware tools, he won’t do either of those things. He’ll immediately see not only that the batteries have died, he’ll also know that she was presented with some suggested steps, followed them, and got stuck when she snapped the cover back on and the unit still didn’t power up. It’s as though he’d been right there with her all along. Contextual intelligence allows him to skip over the useless, frustrating steps and dig into the heart of the problem. The system might suggest to him, “Let’s use her smartphone as a remote camera so you can see how she put the batteries in.” Sure enough, one is upside down and he can tell her to turn it. Voilà: The problem is solved.
But contextual intelligence can go a step further, not just handling the issue and satisfying the customer but helping turn her into a delighted one. While she’s re-inserting the batteries, the support system dips into the device history and discovers that she’s been setting the thermostat manually. A message pops up on the agent’s desktop: “Why don’t you let her know that the thermostat is programmable, then show her how to set it up?”
In much the same way that the home ecosystem uses context to manage the user’s experience, the support system uses context to more efficiently resolve issues and further enhance the value she derives from her connected devices.
Let’s face it: In the main, there’s not much difference between the individual devices offered by various manufacturers. How much “better” can you turn the temperature up or switch a light bulb from green to purple?
The real differentiation lies in how intelligently you can orchestrate the interaction of those devices with each other and with the environment, and how seamlessly you can step in and help users who are having difficulties.
In both of those cases, context is king. And contextual intelligence is the next frontier in the Internet of Things.