Farcon 2017

I had the pleasure of attending and presenting at Farcon.

Yesterday was a series of training sessions in which I ran a session on feature engineering techniques and strategies. I think it went very well and it's something I look forward to potentially running again. I got some great questions and had the chance to talk to a lot of people with diverse problem domains and how to deploy ML in their work.

Today, I got to interview Joseph Konstan in our first ever Data Skeptic live event. I also presented on the opportunity for Chatbots in eCommerce.

In between my activities, I was able to attend a number of interesting talks and presentations. In the morning, I sat in on their startup showcase. I didn't capture the exact details, but from what I heard, there are some excellent startup accelerators, labs, and programs in the Minneapolis area. The size of this sold out conference and engagement of the community makes me interested to keep an eye on this city in the future. While I'm attached to my home in LA, Minneapolis seems like a place a startup should give serious consideration to taking as a headquarters.

Joshua Brueckner, CEO, Air Taylor

Air Tailor is trying to make clothing alteration and repairs easy. Interestingly, their service works entirely and exclusively via SMS. They send you a garment bag, safety pins, and whatever else is needed. You mail in the garment with instructions, and they manage the relationship with the tailors to get your work done in a predictable and easy way. I myself have only had suits tailored, although I must admit I've discarded garments that didn't fit because the idea of finding a tailor was too intimidating. If their price point is good, I could see myself as a future customer.

Atif Saddiqi, CEO, Branch Messenger

Branch Messenger provides the technology platform the help hourly workers and their managers coordinate schedules easier. I was not surprised to learn that most shift workers use pen and paper systems. What did surprise me was just how much time and effort managers must invest to coordinate these efforts. Their messaging app reports to allow employees to photograph their schedule. They use image processing to convert it to structured data. Employees then have the opportunity to get reminders, swap shifts, and track earnings.

They've built out a large number of enterprise integrations, most noteably to me: Peoplesoft, ADP, and SAP. If I understand correctly, their revenue opportunity is by providing large enterprises a tool for efficiently managing their shift workers and potentially provide other products related to HR services. Cleverly, their way in the door is by offering this free app to employees. WIth a critical mass of employee users, it ought to be an easy way to get one's foot in the door with the employers.

Interview with Joseph Konstan

I had the chance to interview Joseph Konstan about recommender systems. We'll be airing this live recording on Data Skeptic on 09/15/17, so stay tuned.

Swannies, co-founder, Adam Iversen

Back to the startup showcase, I caught a presentation about Swannies that is manufacturing golf related apparel targeted at mellenials - a market that apparently is not strong in golf but might become so as that industry works to re-invent itself.

Current and Future Trends in AI Content Creation

There weren't any seats available for Frank Bell's interesting talk on AI Content Generation. Full disclosure Frank's a former colleague of mine, and I hope perhaps a future colleague as well. Although I missed his talk, I'm going to see if I can get him to post some of the highlights here on the dataskeptic.com blog.

Being Strategic with Analytics in the Face of Digital Disruption

Rebeccah Stay shared interesting examples of how many large organizations are integrating analytics. One that seemed particularly novel was McCormick (yes, the spice people). They spun out an organization called Vivanda whose product is called FlavorPrint. It's a taxonomy of flavor identifiers. As I understand it, consumer packaged goods companies could use their system to help craft new products which are well targeted from to appeal to different demographic groups.

Under Armour, Nike, and Reebok also took centerstage. At Underarmour, 700 million dollars was spent ($46 per user acquisition) to introduce health tracking apps. There's a natural consumer appeal for the quantified self community. The data play is, in my opinion, still a frontier with a lot of potential.

Every so often, someone asks a question that's direct and very effective at getting things on track. My own favorite is "what decision do you need to make?" Rebeccah raised the excellent question: "What data do you have that is unique?" I can say very confidently that no one is going to gain a strategic advantage by inventing a unique proprietary algorithm or developing a proprietary technology. Not enough companies are thoughtful enough about what unique strategic advantage they have based on their data.

Customer First Data Science Through Analytics, Product Management, and Customer Experience

Data Difference podcast

Dave Mathias (@davemathias) talks about the common but not often spoken about that disappointment that some organizations have with the fact that they haven't seen the ROI they wanted from an investment in data science. There are many insights from the product management world that Dave presented which could be very useful in helping organizations get a better return on their investment.

Dave's talk was an insightful walk through his seven recommendations for customer first data science:

  1. Cool vs value - The use of a hot new technology or algorithm has no intrinsic value. It's the impact and value delivery the use case brings to an enterprise.

  2. transparency vs creepy - With enough data it's easy to take automated actions that are creepy to users. Perhaps the solution to this problem is more transparency in what data is collected and how it is used.

  3. A little empathy goes a long way - putting yourself in shoes of others goes a long way

  4. Get out with the customer - sage advice, IMO. Field trips are often fun and moreso, insightful in how end users experience your product or service

  5. Building bridges, not walls

  6. Understand yours and others' world lens - On a personal level, I can say that the biggest successes in my own career always had their roots in this insight

  7. Life is a journey, understand your customer's

Recommender Systems: Beyond Machine Learning

Joseph Konstan, whom I had the pleasure of interviewing earlier that day, gave a talk that was an excellent extension of our conversation in the morning.

A particularly interesting point he made relates to the use of dimensionality reduction. To many data scientists, this is so natural a choice we often make it without much thought. Indeed, dimensionality reduction techniques are often effective. Researchers in movie recommendations have stated that there are around 11-18 dimensions for measuring interest. Yet, they're unable to tell you anything interpretable about what exactly those dimensions are. While there is something good to be said about measurable empirical results, if the dimensions don't admit any intuitive notion and remain purely mathematical constructs, how much confidence should we place in them?

In a nice demonstration, Joseph showed the recommendations he received from a few popular sites. In each case the sites showed him the same recommendations as last time - the exact recommendation which did not cause him to buy anything.

The talk was a really great survey of the novel findings and results in the recommender system field and what researchers have been developing.

As is evident from the talk title, a key thesis was that data mining is not the same thing as recommendation. The most useless thing a recommender can do is recommend a product that you have already consumed and have no intention of consuming again. This key insight highlights how typical data mining techniques are not well suited for the unique challenges that recommender systems face. By their very nature, they seek to help with discovery. Metrics of diversity and serendipity are critical, yet not common objectives in most machine learning.

A particular anecdote that resonated with me related to Joseph's grandfather whose highest compliment was "that's not bad". Some people's 3 starts is more appreciated by them than another person's 4 starts. We all have our own sense of scaling and a world class recommendation system needs to understand and deal with this.

As you will hear in our interview, Joseph is a very funny guy in addition to being a good presenter. It's well worth catching one of his talks.

My talk on Chatbots

If you're interested in what I have to say about Chatbots, look elsewhere! My presentation was exclusive to the FARCON attendees, and tailored to that audience. Yet, I have more to say about Chatbots so stay tuned. I was surprised anyone attended my session, because I had some truly stiff competition with other great talks happening in other rooms.

The Importance of Data Strategy and Governance for Data Science Success

Winding up the day for me was a talk from Edward Chenard, Founder Cyberian Data. A quote I appreciated from early in his talk was: "We often don't quantify our technology use. We just assume new is good."