Published: February 4, 2014
The NodeSpan team works a late-night session preparing for the MIT $100K Accelerate Contest
Floor to ceiling white boards are a maze of flow charts, draft algorithms, and words and phrases representing issues that once spent a moment in the sun as a problem requiring a solution. The screen shows a PowerPoint in perpetual draft mode—today’s complex visual may well be scrapped by “showtime” as messages and descriptions are modified through a process of discovery.
The co-founders of NodeSpan, an MIT big data and machine learning startup, hunker around laptops in a conference room at the Martin Trust Center for MIT Entrepreneurship. Seemingly simple questions morph into complex discussions amid the murky intersection of logic-based algorithms and the quirky obscurities of language and the need to communicate all of it to a room full of people and a panel of judges.
The MIT $100K Accelerate contest is just weeks away.
NodeSpan is one team of just 30, culled from 270 applications, preparing to pitch their newly founded business at Accelerate, one of three parts of the MIT $100K Entrepreneurship Competition.
Accelerate teams will make their presentations over three nights beginning Feb. 5, with a grand finale show and announcement of the winner on Feb. 20. In addition to the $10,000 Daniel M. Lewin prize for the winning team, other prizes include a $2,000 AARP Foundation prize, and a $2,000 audience choice award. Each team presents for five minutes, followed by a Q-and-A with the panel of judges.
But unlike last fall’s Pitch Contest—the MIT $100K competition where speed rules in a judged one-minute pitch of a business concept—teams competing in Accelerate spend the weeks leading up to “showtime” going beyond just an idea. There is customer analysis to do, prototypes to take off the drawing board and build and test, and a case to build that the team’s ideas are not just worth exploring, but explored and proven viable.
To that end, organizers of the $100K provide each team with the resources to make their business concept more concrete. Each team competing in Accelerate is given a $1,000 expense account, though some in the emerging market sector may receive additional funds for travel. Mentors work with teams and the connections are made to potential customers who can provide feedback.
As the team from NodeSpan described it, those resources may be the real reward, whether the team pockets the prize money or not. Particularly valuable to them, they said, are the mentors who can connect them with potential customers, people holding the very jobs in major companies NodeSpan intends to target for sales. Mentors also provide feedback that helps to shape the team’s concept into a product customers actually want.
So what is NodeSpan building? The team is developing a machine learning and analysis tool that will help companies improve customer experience by analyzing a wealth of untapped text-based information: customer service interactions (which are increasingly online), comments on sites like Amazon and Yelp, and social media channels. According to CEO Zack Hendlin, MBA ’ 14, NodeSpan seeks to analyze it all, determine trends, and identify issues that, when properly addressed, can lead to more efficient and higher quality resource allocation benefitting the customer experience.
The goal is to utilize data to route customer contacts to the most relevant representative, to better determine training, and to develop more expedient response techniques.
“You’re able to resolve questions more quickly, get the right person to the right representative, which we learn from all the previous history,” Hendlin said. “We allow the customer to be able to figure out if there’s a related answer to their question—even if they’re using words that won’t come up in a keyword search—through the machine learning and statistics that we’re applying.”
Until NodeSpan, Hendlin said, about 80 percent of the data generated from all these sources went unused, principally because, frankly, “it’s hard.” Short of assigning employees to review it all, or developing surveys that can be expensive and unreliable, there are challenges of applying logic-based algorithms and hard data principles to something as nuanced as language. In fact, a recent NodeSpan team meeting included over an hour discussing the sentence “my heels are itchy and dry,” and the problem of ensuring the algorithms the team is developing would properly interpret the sentence as related to a skin-care product, where using key words would most often tag the word “heels” as pertaining to shoes.
“There’s an ethos at MIT that’s drilled into us that says ‘go do something big,’” Hendlin said. “But one of the challenges in being a startup is demonstrating that you’re valuable before you have customers or have data.”
Hendlin said their idea took root when he and co-founder Tejas Kulkarni met at last August’s T=0 Entrepreneurship Festival. Early conversations with a Fortune 500 company confirmed they were on the right track, as companies are unsure how to mine text in a useful way, and surveys and focus groups have pitfalls, including expense.
According to Kulkarni, a second year PhD student in the MIT Department of Brain and Cognitive Sciences and the MIT Computer Science and Artificial Intelligence Laboratory, the NodeSpan concept has “drastically changed” since the team was selected to compete in Accelerate. Access to people in large and relevant companies—their potential customers—gave them insight that both confirmed their path and took them in unexpected, and more focused, directions.
“On the technical lead side, when we began we were too broad. We were trying to be everything to everyone,” Kulkarni said. “We had an idea that seemed applicable, a lot of what we were doing as we were working on our PhDs. But a lot of what we were doing was not product-focused. It was technology-focused, it was algorithm-focused, as opposed to problem-focused. It was important for us to focus on a very narrow problem.”
The feedback served to make them more confident about their upcoming Accelerate presentation. The more they learn about their intended market, the more they tailor their product for customer needs.
“The focus changes, but that gives you a certain level of confidence,” said Kulkarni. “Especially for technical people, we can chase cool technology as opposed to building something people really want. Getting in touch with those people in a meaningful way, especially in our case as we’re an enterprise product, has reduced the nervousness.”
Gino Korolev, MBA ’15, contest director at the MIT $100K, said NodeSpan’s experience is not unique among Accelerate teams. That their business has evolved while preparing for the contest is a strength, not a weakness, Korolev said. And the program exists to allow students the opportunity to go through those key steps in startup development.
“It’s a good thing, because I think there’s a general notion that startups need to continually validate [that] whatever they’re trying to innovate makes sense for the customers,” Korolev said. “If they don’t, if you don’t talk to potential customers, it’s like trying to play darts blindfolded and trying to hit a bull’s-eye.”
“The basic idea doesn’t really change, but the interaction with customers can have a large impact on approach and specifics,” Korolev said. “We want the teams to get to something tangible, and get an understanding of how the customers would react.”
Teams meet with assigned mentors as often as weekly. In addition, according to Hendlin, there’s a sense of a community at work, especially at the Martin Trust Center for MIT Entrepreneurship and in the private sector. “It’s amazing to me, when talking to senior people at major companies, that they seem excited to help, and very curious about ideas coming out of MIT,” Hendlin said.
In addition to the ongoing process of building the product, Hendlin said his team will hone their pitch and practice the nuanced art of telling a compelling story about NodeSpan. But, he said, while winning would always be welcome, they’re already looking beyond the finale and the announcement of the winner towards NodeSpan serving customers.
“There is a need in the market,” Hendlin said. “And we’re beyond this as a mock-up.”