I’m often asked how I became a data scientist. Here’s the story. I hope you enjoy the departure from normal Tech Tuesday topics.
As many have done, I learned the ropes of data science by doing, but my story is hardly ordinary, and my path was hardly direct.
When I started my career there weren’t any data scientists. Because the term hadn’t been invented yet. But data science was exactly what many of us quietly did, without the moniker of “the sexiest job of the century”.
Sometime around 2010, I left 12 years of engineering behind to dive firmly into the world of start-ups, thinking that I could leverage my data skills in the still-nascent but growing field of data science. My destination was anything but clear; I had some friends-cum-business-partners over the course of a few years and we dove head-first into creating several start-ups, all backed by data (but not always a whole lot of funding).
Before I could work effectively using data science tools and modern methodologies, I had to transition from a long-time signal processing engineer (or research scientist, or whatever you call someone who plays with numbers all day the way I did) to a bonafide data scientist.
At that time, I admired a company called Hunch (quickly acquired by eBay). Hunch worked hard to elicit and understand peoples’ (and even companies’) preferences all over the internet. One thing they did was especially interesting to me. They asked fun questions of anyone who came to their site. Clearly, they were doing a lot behind the scenes because the insights they created from the answers to these questions were fantastic!
Between these answers and information they crawled from all over the web, they could compare, for example, a Volkswagen to a Journey song. I’m not making this up. I was floored and I wanted to learn everything there was to know about questionnaire science and preferences. I wanted to understand peoples’ reasons for liking things. If I can figure this out, I thought, I can help make the online experience better for lots of people.
I had a goal, but I didn’t have any data or knowledge on how to proceed. Somehow (it’s hazy) I was itching to make some progress on the morning of St Patrick’s day. I had about 20 people coming over for a BBQ and they would be there all day. I had captive subjects!
Writhing my hands together, I scrambled for my laptop and in a couple of hours, I had written a web app. It was a 20-question “survey” that recorded the answers to all sorts of easy-to-answer questions, such as: Are you more of a dog or cat person? Do you think it’s okay for men to wear pink? Coffee or tea? Chocolate or vanilla?
I collected data from all 20 people, and before the day was done, I presented my “findings”. It was particularly interesting to compare the answers of couples who spent all their time together — their answers were sometimes very different. It was equally as interesting to see which two people were most similar or dissimilar in their answers. As night fell, we talked a lot and laughed even more. And I was hooked.
I took my love for this type of preference gathering and similarity measurement very seriously. Shortly after this backyard data gathering, I co-organized a large study with the University of Gent in Belgium where we collected more than 100k answers from hundreds of students, comparing answers to “Pepsi or Coke” with “Would you be willing to switch phone providers for a $100 coupon?” It was fantastically fun to see my backyard analytics stunt grow into something much bigger.
Three startups later, and having worked with clients like Kim Kardashian and Snookie (I actually thought they were the same person for a while!) I learned an incredibly valuable lesson: passion matters a lot.
I’m fortunate to wake up every day and do something I’m still in love with. I did it when there was no money, only fun conversation, and laughter.
As for all those questionnaires, well, lots of companies use similar tools to understand their users. Companies like StitchFix and TrunkClub (where I was for over two years) rely on them extensively to understand user preferences. These onboarding answers pair beautifully with product recommendations, another area I’m excited to see receive more attention by the data science community lately.
So what do you do if you want revenue to go up and product returns to go down? Well, ask your customers the right onboarding questions, listen to what they say, and recommend an experience they cannot resist.
If I can be of help with this, feel free to hit reply and let me know what you’re working on!
Predict What People Will Think of Your Photos
Would you be better off wearing your hat forwards or backward if you want to: attract a significant other, get a job, look smart? Ask Photofeeler! An app using A.I. trained on 100 million opinions to predict what people will think of your photos. There’s a lot to go “wow” about lately and this app is certainly one of them. I cannot stop thinking of the applications of this type of computer vision and A.I.
Paper Masks are Fooling Facial Recognition Software
Facial recognition is widely embraced as a security tool — law enforcement and corporations alike are rolling it out to keep tabs on who’s accessing airports, stores, and smartphones. There are many reasons why people may want to fool this type of surveillance with tricks like paper masks. Expect to see more in this space as facial recognition gets better/creepier
How to Ensure Your Data Science Projects are Successful Every Time
In a 2017 survey, Gartner analysts found that more than half of data science projects never deploy. This might lead some to believe that there are flaws within the data, analytics tools or underlying ML models, but that’s not generally the case. This topic is getting its deserved attention. Expect to see a future Tech Tuesday on it!