Why Your Startup Needs Data Science
It’s official: Data science isn’t just for tech companies anymore. From the ethical treatment of farm animals to sleep optimization and fashion, Zank Bennett, CEO of Bennett Data Science, helps entrepreneurs utilize artificial intelligence in a wide array of industries. Working with large and small companies alike, Bennett makes complicated technology easy-to-use so even entrepreneurs with little tech experience can harness the power of AI. I recently spoke with Bennett for more insight on how business can capitalize on data science and reap its rewards.
Why should entrepreneurs utilize data science, even if their startups are not tech-focused?
For companies to be successful nowadays, they really have to nail the personalization piece, and entrepreneurs get this more than most. When entrepreneurs start companies, one of the first things they think about is how they can serve their customers, but then when it comes time to scaling what the customer wants, it gets very difficult to conceptualize how a human would do that. That’s where machines come in. Personalization at scale is what we see most from entrepreneurs. With retail, for example, we’re working to create models that predict what people want based on how they’re browsing products. So we take a product feed and look at the attributes of what people are browsing and put that together with what other people have browsed with similar preferences. With that, we can make really informed recommendations. How many different attributes can be told from a shirt or from an accessory? How do we put things together and [feature] auto-tagging and all of these different things that companies need and want?
How would you describe personalization from a data-science point of view?
Personalization is about taking a product that we think someone wants and putting it in front of that person and having a win. And so how do we do that? How do we go grab the product and show it to the person? Recently, I’ve seen companies who say they provide personalization, and what they’re really doing is they’re segmenting the audience into a small group. The first group to play with is gender, then maybe you segment by age, and so now you’ve got four different groups. That’s not personalization, it’s segmentation. Sure, it starts to get towards personalization, but it’s not very informed. It’s not what we might call intelligent use of data.
When we start getting more predictive instead of descriptive, we start to look at past behaviors and how they predict future behaviors. That’s where the really interesting work happens in the recommender space, or even in the classification space, where we might have a user onboarding with us and the user fills out a bunch of information, and then immediately we can treat the customer differently based on how we predict the customer’s going to act in the future.
There really is a big difference between just chopping up users and saying, “Oh, we’re going to treat these four segments differently and sort of guess what they might want and hone in on that” and doing some intelligent segmentation based on actual actions that customers have performed in the past and saying, “I see we’ve got some segments, we’ve got some attributes. We can plug all those into a model that will predict what someone’s going to want in this situation.”
How can an entrepreneur successfully implement data science into their company?
The number-one thing is to get data science integrated within teams. I don’t think data science should be this autonomous thing. I think it should be very well integrated in marketing, sales, product, etc. The second thing is we have to give data scientists the data that they need in a format they need it in so they can be efficient workers. There’s this idea now in some companies that we give data sciencists access to this big clump of data and just let them go at it, and that’s very ineffective. In fact, data scientists for a given application don’t need data in lots of different formats. Instead, we can provide a data lake that a data scientist can use day in and day out, 80 percent of the time. It adds massive efficiency to a team.
The next thing is being sure that data scientists can deploy their models and have a lot of support to do so. Those infrastructure pieces are part of what we call a pipeline. Data comes in and goes to data science to do something magical, and they go out and get deployed. That magical part in the middle is often what takes the least amount of time.
Do you think that data science can actually create more jobs within a company instead of replacing human labor?
With data science, we can automate tasks that can be done so much faster and so much more efficiently. When we reduce costs for a company, it seems to me that they can scale in other ways. I think there’s going to be more jobs as we make companies more efficient, not fewer. Because as we increase profitability, companies always spend to grow. They don’t just put the money in their pockets. I think that’s a misconception, especially with startups. The whole reason startups raise money is to grow, not to just save the money. If they become more profitable, they’re able to spend that money on more resources, and I think eventually that does lead to more jobs.
What’s next with data science?
I think data science will be understood a lot better, and we will take away this title of data scientist and replace it with much more descriptive titles like machine-learning engineer or statistician or data engineer. I think this general blanket term of data science needs to go away so we can be more descriptive. I also think it needs to be better integrated with companies. Data science will lose this idea that it’s this autonomous group that could come in and help anyone, and I think it will be coveted as something that can really help product or sales or marketing — but as part of those groups, not on its own.
I think we’re going to see massive changes in natural-language processing and the way we can summarize text and the way we can utilize language to communicate. I really hope self-driving cars are something that we have sooner than later, and I think that will help us so much in terms of efficiency. Some of the applications with computer vision are just amazing these days, from how we’re using it with fashion to how we’re helping cars to drive themselves. And as that gets better and progresses, I think our world will really change.