By Andrew Shaughnessy
Artificial Intelligence is transforming the way we do business, and Chattanooga’s tech startups are in the thick of it.
How Chattanooga’s Tech Startups Are Transforming Business Using AI and ML
As Amazon’s Alexa and Apple’s Siri find their way into homes across the country, Artificial Intelligence (AI) is fast becoming an integral part of nearly every business sector, from data mining, entertainment, and finance to the automotive and food industries. Machine Learning (ML), a subset of AI able to be “trained” on huge amounts of data and make predictions based on patterns found in that data, is turning out to be a particularly productive offshoot. In Chattanooga, companies are taking advantage of the trend and breaking ground with new technologies that utilize AI or ML.
A powerful offshoot of AI, and the driving force behind the future of Artificial Intelligence is Machine Learning (ML). With ML, algorithms are created so that machines are trained to process huge amounts of data and then learn from feedback to think like human beings – essentially creating human-like Artificial Intelligence. In Chattanooga, companies are taking advantage of the growing demand for AI and breaking ground with new technologies.
“At this moment in technology, if what you’re doing isn’t using machine learning techniques, then you’re not on the cutting edge. You’re behind,” says Noel Weichbrodt, who serves as principle software engineer for Chattanooga startup Pylon. “This is an industry-wide trend, in the same way that having an app is an industry-wide trend – or having a website was.”
In fact, according to researchers with Accenture, a global management company, AI could double annual economic growth rates and dramatically boost labor productivity by 2035 in a number of developed countries, including the United States. Major players like Google, Amazon, Apple, and Microsoft have devoted significant resources to AI research and development. Much of this technology is already all around us – think about when Amazon suggests books based on your browsing and purchase history, or when Spotify curates playlists based on what you have listened to. Every action you take on Amazon or Spotify creates a data point that enables their program to make inferences and give suggestions. Newer applications include Amazon and Google’s voice recognition software – ask Alexa a question, and it will do it’s best to recognize your speech patterns, translate it into text, and deliver an answer. The value here is that Alexa “learns.” The more you talk, the more data points the device receives, and thus the better it will be able to recognize and decipher your speech patterns and do its job.
The heart of AI and ML’s potential impact is not that they can do things that humans cannot, it is that they can do them much faster. As AI applications absorb huge amounts of data, crunch numbers, search for patterns, and provide insight much faster than people, they free human beings from simple, generic tasks so that they can better use their time and creativity. The possible applications and implications of this are vast, and the potential impact over decades to come, for good or ill, is difficult to overstate.
“If the technology’s development is a wave, we’re surfing through the barrel,” says Weichbrodt. “We’re not sure where all the advantages and insights are going to end, but it’s already making a huge impact.”
Here in Chattanooga, our entrepreneurial city’s tech startups are not letting this moment pass them by.
“At this moment in technology, if what you’re doing isn’t using machine learning techniques, then you’re not on the cutting edge. You’re behind.”
A busy, working mom walks into her kitchen. Children burst through the door and run around the house. Her arms are full of groceries and school backpacks, and it’s almost dinnertime.
“Alexa, open Tasted,” she says.
Amazon’s voice-activated AI device turns on its tiny brain and listens intently.
“Hello! And welcome to Tasted,” says Alexa’s electronic voice.
“What’s for dinner tonight?” the mom asks.
Alexa suggests several meal options – each one is nut-free (one of her children has a peanut allergy), calculated to serve the correct number of people, and tailored to the family’s specific food preferences. It’s all information that Tasted has picked up over time through conversations. Once mom has picked out a recipe, the device proceeds to text her a grocery list of necessary items, and then talk her through the recipe, with visual aids simultaneously appearing on her iPad, allowing her to make the recipe hands-free.
That’s the dream of Chattanooga-based AI technology company Pylon.
“The vision for Tasted was always to make it easy and pleasant to cook for yourself or your family and friends,” says Weichbrodt. “The reason it’s not as easy now is because you have to Google everything, and Google doesn’t know your context and preferences. Google doesn’t know your allergies. Google doesn’t know that you prefer Italian food over Korean. Yes, they can try and infer all of this, but really what they have is a giant database of recipes and it’s on you to browse through and pick the things you think are best.”
With Tasted, the idea is that the “assistant,” whether Amazon’s Alexa, Google Home, or something else, will learn and remember your preferences, allergies, and context, eventually reaching the point where you can just ask it: “What should I cook for dinner tonight?” From there, its suggestions will fit your specific needs and tastes nearly perfectly.
Though Tasted is still in development, the possibilities for this sort of AI-based, conversation-driven technology are just around the corner. Already, Pylon has developed a prototype for Bartender, another conversational app for Alexa or Google Home, which talks the user through how to make various mixed drinks.
“It’s really “Search-Find-Acquire-Do,” explains Pylon co-founder Mike Tatum. “We’ll probably build other variations going forward … health care, security … We’d like to be seen as the best platform for any companies that help consumers find content that makes their lives easier, faster, more convenient, and cheaper.”
Bellhops started small, but the Chattanooga-based moving company now operates in more than 15 cities nationwide. That sort of large-scale operation makes monitoring the quality of moves and workers a potentially daunting task. That’s where machine learning comes in.
In order to provide the best customer experience possible and improve the quality of moves, Bellhops wants to better understand the performance of their workers to best match them to jobs. After each job, Bellhops asks customers to provide feedback and rankings on each mover’s performance – things like punctuality, attitude, and whether anything was broken during the move. That data is aggregated by the company’s ML tool to build a profile for each bellhop. It gives them ranks and grades, allowing the company to observe their behavior and skillsets as compared to their peers and how they change over time. The more data that is added over time, the more accurate the analysis and predictions.
“We get smarter while we sleep,” says Bellhops CTO Scott Downes. “Machine learning is allowing us to get more efficient with matching the right people to the right jobs.”
Machine learning also plays a key role in the complicated task of efficiently matching workers to jobs. “There’s a lot of seasonality and periodicity with moves,” says Downes. “Everybody wants to move on Saturday morning at 8 a.m., and we don’t have an infinite supply of workers.”
Rather than employing full-time movers, Bellhops employs college students part-time. They have the ability to set their own schedules, inputting their specific availability for any given week into the Bellhops app. It’s a constantly changing story of supply and demand. Data is continually added by customers (demand) and movers (supply), and the machine learning algorithm takes that data and churns out optimal scenarios – matching movers to jobs nearly instantaneously.
“This is the point where a lot of moving companies break down,” says Downes. “They’re not able to operate at scale. It’s easier to imagine providing quality when you know the five guys who work for you and you keep track of them on your clipboard. With machine learning we’re able to scale up to thousands of moves without having to worry about a number of individual managers understanding the details of the performance of each individual.”
Now, Bellhops is researching ways to use ML image and video analysis to accurately estimate move times. In theory, the customer will be able to take their smartphone, walk around the house with the camera, and send the video to Bellhops to plug into the program. The ML application would be able to identify and catalog objects, counting how many sofas, beds, chairs, tables, and so forth need to be moved, then automatically file that inventory into the system to determine how long the move should take. As with the employee profiles, the more these data inputs are compared with actual move times, the more accurate the program will become at predicting how many and what people are needed along with move times.
“This is transformative technology,” says Downes. “Every company should be looking at these developments in AI and ML as an opportunity to free up human beings to do what human beings are good at. We shouldn’t be limited by technology, we should be enabled by it. For us, it’s not just about saving money, creating efficiencies, and making more dollars, it’s about changing people’s lives at a really vulnerable, sensitive time – moving day.”
“We’re big on culture, on establishing sustainable companies that take care of people, because people are the most important asset
a company has.”
For Gabe Weaver, partner at Chattanooga-based technology company Very, it all started several years ago when a talented software engineer quit out of the blue.
“There were no warning indicators,” says Weaver. “We have a pretty open culture and provide ample space for feedback, so we were like, ‘That’s interesting.’ And then I thought, ‘I wonder if we can tell whether people are happy or unhappy based on emoji usage in Slack?’”
Rather than just sighing with resignation, hiring another engineer, and moving on, Weaver began to explore ways to apply ML to the problem – gathering data from Slack channels, searching for clues, and looking for better insight into what makes people burn out. They spent about six months prototyping, and eventually came up with a system that analyzes the messages from Very’s public Slack channels in nearly real time.
“It takes a couple of seconds to process each message because it has to go through a couple calculations, and then it gets pushed into a big data set that the ML algorithms can learn from,” says Weaver. “The more data we add from every team that uses it, the smarter it becomes and the more accurate analysis it can produce.”
The resulting product, called Foresight, has now processed around six million messages from a dozen different teams within Very, and boasts an 85% accuracy rate for predicting people who are going to quit in the next two to three months.
The basic idea is that everybody has a unique fingerprint reflecting the particular way they communicate. By gathering data points like grammar, length and frequency of messages, and emoji usage on Slack, Foresight is able to map a fingerprint for each individual and watch how it changes over time. Next, Foresight’s ML algorithm looks for correlations, clustering data points together based on similar characteristics and producing sentiment analysis, i.e. determining the writer’s attitude in relation to the message.
“The machine kind of decided on its own to divide the data into seven different groups,” says Weaver. “Taking those different data points, we came up with what we call a stress metric: grammar errors, sentiment, how many hours they’re working in a week, and how long people take to respond to messages. You can essentially put all these factors together and build a model of what leads to burnout.”
Weaver hopes Foresight can be used to provide a feedback mechanism for employees to see patterns that indicate stress and better take care of themselves. Of course, in the wrong hands, sentiment analysis begins to sound a bit overbearing.
“It’s kind of a mixed bag,” Weaver admits. “People have offered to pay us a ton of money to do things with this that we don’t want to do because we don’t believe organizations should be run that way. We’re big on culture, on establishing sustainable companies that take care of people, because people are the most important asset a company has. If a company is going to use our tools to get ahead a little, that’s fine, but not if it’s at the expense of everyone who works there.”
One large organization that is seeing a lot of turnover with their socialwork employees wants to use Foresight to head off burn out. A candy manufacturer is interested in using it to keep tabs on floor supervisors’ health in order to be more predictive in mitigating risks. Because sentiment analysis can be applied to any type of written communication (not just Slack), Very has also explored applying Foresight to video conferences, transcoding emotion from voice signatures.
“It can cost two to three times a person’s salary to replace them,” says Weaver. “If a company can use this to build the right culture and reduce turnover, it will generate more revenue.”
While Chattanooga-based tech company Skuid is not developing its own ML technology from scratch, they aim to be a bridge, connecting customers to existing ML technologies and services.
“Right now, the industry has some pretty powerful tech that basically empowers smaller companies to use large scale AI and ML without having to build their own infrastructure,” says Skuid Principle Software Engineer Zach McElrath. “Skuid is really playing the role of matchmaker.”
For manufacturing industries, customers could post photos of a broken part to a company’s Skuid app and kick off an Amazon or Salesforce-based ML image recognition process to identify the part, assess how it is broken, and recommend next steps.
For prioritizing floods of customer service calls, tweets, and emails, sentiment analysis of speech or text could prioritize responses based on a customer’s urgency or anger “score.”
Another common example is lead scoring in sales, where the key metric is converting leads. You can’t follow every lead, so deciding which leads to prioritize and which leads to ignore is huge. In this case, the under-the-hood ML tool would take data points from previous interactions with potential leads and turn them into a score – who is most likely to be converted – allowing salespeople to prioritize their response to leads based on their score.
“Companies have so much data, but they don’t know what’s valuable,” McElrath adds. “This is basically about looking at all that and saying, ‘What are we really trying to achieve as a business and what variables in all of our data are going to achieve that for us? Skuid is enabling you to build those relevant ML insights into your application without writing code.”
The upfront cost of building the infrastructure for machine learning is very high and requires a great deal of expertise. By serving as a bridge to ML, Skuid is lowering the barrier of entry.
“In terms of all the things you can do with this, the sky’s the limit.”
It’s clear that AI is already making a massive impact by allowing companies to harness the potential power of big data in thousands of different ways. The technology is amazing, it’s cutting edge, and we are only just beginning to see all the ways it can be applied.
Yet, as with so many significant technological advances throughout history, some worry that the automation and increased efficiency that AI and ML bring with them mean that jobs will be lost.
But that, at least according to Chattanooga’s machine learning entrepreneurs, is missing the point.
“One of our partners took an eight-hour process and condensed it to 15 minutes with Skuid,” says McElrath. “People look at that and say, ‘Isn’t that killing jobs?’ And I say, ‘No, it’s enabling creativity. It’s allowing people to focus on much more important things.’”
“People are really good at finding new ways to deliver value to a company,” he adds. “If you free them from doing busywork for seven hours and 45 minutes of the day, what else can they do for you? People are creative, that’s who we are. So, if we can use machine learning to help us focus on the right things, that’s really exciting.”
As they evolve, AI will reshape industries by changing millions of jobs, reducing operational costs for businesses, and improving productivity. Whatever direction AI and ML take us, these Chattanooga companies will be right there in the thick of it.