How Motorola Uses Big Data Analytics to Improve Its Smartphones
When Motorola launched its RAZR M two years ago, a rumor started spreading that you couldn’t use Wi-Fi and the 4G cellular networks at the same time. It wasn’t true, but thanks to a customer sentiment analysis project Motorola was prototyping at the time, the company detected the rumor and worked to correct the misinformation before it damaged the brand.
Such is the razor-thin margin for error for consumer device manufacturers these days. Whenever a new smartphone launches, consumers pore over every square millimeter and kilobyte, looking for any sign of a hint of a flaw. Got a little too much flex in the aluminum case? Voila–it’s “Bendgate” for Apple and its new iPhone 6.
In the olden days, manufacturers would conduct consumer surveys or amass focus groups to find out what customers really thought about their products. It was time-consuming and expensive, but the results were generally accurate. Besides, there was really no other way to get the data.
But that approach doesn’t fly in today’s consumer electronics market, where entire markets, like wearables, are born and advance several generations in a matter of months, and where news of a product snafus spreads like a virus. Luckily, the communications breakthroughs that gave us the Web and smartphones allow manufacturers to take a much more direct approach to detecting consumer sentiment.
“People nowadays take everything online,” says Mike Stringer, co-founder and data scientist at Datascope Analytics, a Chicago-based provider of big data analytic professional services. “A lot of organizations still use focus groups or surveys and there’s nothing wrong with those approaches. But they miss out on what people are saying in the wild, which is ultimately what they they’re trying to estimate by doing focus groups and surveys.”
Stringer helped build Motorola’s Mobility Service and Repair’s sentiment analysis program, which seeks to extract and distill general thoughts and feelings people express toward Motorola’s wireless devices, based on what people write in chat rooms, forums, news sites, social media, comment sections of online retail websites, and other publicly accessible places on the Internet. The project began two years ago, when Motorola was owned by Google; it has since been acquired by Lenovo.
“We wanted to minimize service and repair failures and believed we could become more proactive in addressing our customers’ needs following a smartphone launch,” said Ahmad Shabazz, Sr. Manager of Business Operations & Strategy at Motorola Mobility. “We turned to Datascope to find out if it was possible to leverage real-time conversations among smartphone users across the Web to know immediately when they encountered problems.”
One of Stringers challenges was weeding out the stuff on the Internet that’s irrelevant to Motorola products, which of course is the vast majority of content on the Net. The raw data that’s input into the solution is very big, measuring in the hundreds of millions of phrases per day. Motorola isn’t measuring the entire Internet–there’s no Twitter or Facebook analyses, for example–but the data is big nonetheless.
Most of the information is of questionable value, or “crap,” as more than one data scientists has put it to Datanami over the last few months. But that high”noise” level shouldn’t prevent would-be big data researchers from looking to the Internet to find signals–provided those signals have to do with public opinion.
“The scientific approach is how you cut through the noise and figure out what people are saying and in what proportion and in what certainty,” says Stringer, who has a Ph.D in physics from Northwestern. “It’s just like with noisy telephone calls. A lot of what science and engineering has contributed to developing clear communications has been figuring out what the underlying signal is with all that noise.”
When Datascope’s text parsing engine detects a sentence relevant to Motorola on the Internet, the software’s classification algorithms then must ascertain exactly what feature the person is writing about. It could be the network connectivity, the PenTile display, the camera, or whatever else tickles a consumer’s fancy or inspires their wrath. Finally, the software (which is built on a foundation of MongoDB, MySQL, and ElasticSearch) decides whether the comments are positive or negative.
The data science aspect of the project was somewhat challenging, Stringer says. But the hardest part for Datascope was presenting the results of the analysis in a thoughtful and useful way. “One of the biggest challenges was, how do you [present information about] 10,000 phrases without dumbing it down too much?” Stringer says. “That was the really challenging part: Dealing with managers and people who are actually acting on this information.”
The word cloud turned out to be an effective tool for summarizing the words or phrases that pop up most often in public comments about Motorola’s products. “That was something we tried early on in our iterations,” Stringer says. “The users said ‘This is clearly people talking about the Razr M.'”
However, the word cloud wasn’t capturing enough about what’s useful, including the opinions and emotions and sentiment of users. So Datascope made further tweaks to ensure that the main thread of what people are saying are brought to people’s attention.
When the big data sensors detect a signal, they’re great at bubbling it up for humans to look at. What machines can’t do is determine whether the bad reviews of Motorola products are people with bad experiences, or merely Apple lovers looking for a fight in an Android forum. That’s where people must step into the equation, and what Stringer strived to deliver in the end product.
“Good data-driven tools don’t try to replace human intelligence. They just try to augment it and let humans do what they’re good at doing,” he says. “Machines are really good at trolling around the entire Internet and finding the content that people are talking about. That’s something that computers can do much more efficiently than humans.”
Despite the early rumors, the Wi-Fi and 4G cellular radios work fine on the Razr M. The high-resolution PinTile display is generally well-liked, along with its fast 1.5Ghz dual-core processor and overall thin build. The battery is held in high regard, despite not being removable, while the high-def camera has gotten mixed reviews.
Motorola continues to use the consumer sentiment tools developed by Datascope to give its product development and issue response teams information about how to proceed. At 26 months of age, the Razr M is an old man in the smartphone world, but the sentiments and feelings about that product have manifested themselves in the new Razr Maxx, which debuted four months ago to generally good reviews.
The customer feedback (or “customer outreach,” if you will) is also evident in the new Moto 360 smartwatch that Motorola launched a month ago. “They are using it to assess what people think about the smartwatch, so they’ll learn how they can continue to improve design,” Stringer says.
“We see no limit to this tool,” Motorola’s Shabazz says. “It lets us keep a pulse on our customers in a way we actually understand, and with accuracy we never thought possible before we worked with Datascope. “
Competition is fierce in the consumer electronics market, and eventually all manufacturers of a certain size will use this type of big data-powered sentiment analysis solution to accelerate product development and problem resolution. The advantages are too great and the savings compared to small data sampling are simply too good to pass up.