The Importance of Onboarding Data
Tech Tuesdays

The Importance of Onboarding Data

By July 23, 2019February 2nd, 2020No Comments
Reading Time: 4 minutes

The week, we dive into onboarding – what it is, who needs it, why it’s so important and some A.I. use cases.

What is onboarding? Onboarding is usually the first interaction an e-commerce company has with its clients. It’s the let’s get to know each other stage.
Who needs/uses it? E-commerce companies, generally any business that provides personalized services.
Why is it so important? Getting onboard right is critical, as it solves the “cold-start” problem, where a business knows little or nothing about a new user. Businesses generally have lots of questions for users, such as:
  • How much do you usually spend on jeans (fashion)?
  • Do you like horror movies (entertainment)?
  • What kinds of coverage do you need (insurance)?
It’s got to be conversational – users expect that the info they provide will result in a highly-personalized experience. If fact, I’ve seen instances where asking more onboarding questions results in greater customer lifetime value (LTV); users who spend more time in the onboarding process anticipate a better future experience. It should be a two-way conversation. Doing that at scale is where AI comes in.
How to consider it / what to do with it / what it informs / how data science deals with it: AI is used in several ways:
  1. User classification to qualify users – this is essential to businesses with a human in the loop where it’s important to know which users qualify for human assistance.
  2. Personalized recommendations – before we know anything about a user, recommending popular products is about all we can do. That quickly changes with relevant onboarding questions, facilitating more accurate recommendations, for example, using collaborative filtering.
  3. User segmentation – comprehensive onboarding combined with purchase history can provide powerful user segments. We use a technique called multiple correspondence analysis combined with hierarchical clustering to identify user segments. This informs content creation for sales and marketing.

If this is interesting to you, hit reply and let’s have a chat. Alternatively, I’ll be giving a half-day talk on this and other topics, August 21st at the TDWI conference in San Diego.

Sign up for Tech Tuesdays

Of Interest

StitchFix:
With revenue up nearly 30%, its stock up over 60% in 2019 with a $3B valuation, StitchFix is proving that their business model is working. And that business model involves a lot of A.I. to facilitate automated sales. Stitch Fix CEO Katrina Lake believes that the precision of Stitch Fix’s algorithms helps manufacturers more closely align supply with demand. 100% of what they sell is through 1:1 personalized e-commerce. None of what they send out is actually chosen by a consumer. A.I. driven personalization is key to their success. They consistently delight their customers with A.I. To do this, they employ over 100 data scientists. And interestingly, they thrive in what Lake calls “small data”. It’s actionable and precise. They’re not out there boiling the ocean. And in a time of “big data”, this can seem counter intuitive (and like a breath of fresh air). Hear Lake talk about this and more.

New Data Science Journal:
Want to get into the details of cutting edge data science? The MIT Press and the Harvard Data Science Initiative (HDSI) have announced the launch of the Harvard Data Science Review (HDSR). The open-access journal, published by MIT Press and hosted online, features leading global thinkers in the burgeoning field of data science, making research, educational resources, and commentary accessible to academics, professionals, and the interested public.

Can’t Let it Be:
Any idea who wrote your favorite Beatles song? Well, if you don’t know, don’t worry, the Beatles might not know either! But A.I. may be able to settle a few arguments. It’s commonly known that many Beatles songs were written by either Lennon or McCartney, yet attribution usually shows “Lennon-McCartney” (both of them). So, who’s responsible for writing those songs that have double attribution? And within those songs, did one write the verses and the other write the chorus? What about the bridge? Given this conundrum, the paper’s authors set out to build a classifier to answer just those questions. They include text and musical elements into a very well constructed analysis. It’s technical, and worth a read.

Extreme Speech:
There’s a lot flying around lately about extreme speech and the part that big tech companies play in its dissemination. Watch Jonathan Greenblatt talk about the responsibility of big companies to address these growing concerns. Greenblatt is the CEO of the Anti-Defamation League (ADL), one of the most respected civil rights organizations in the country. The implications to A.I. are real and tricky. Companies like Facebook say they can’t easily detect extreme speech (or pictures/videos) since there’s so little of it compared to other content types, what data scientists call a class imbalance. There is a real push to solve this class imbalance problem.

Kibbe Fashion Types:
Update on Kibbe fashion types. After quite a few positive responses from the mention of Kibbe Types and our work in classifying users, the BDS team is working on a simple interface where users can upload a photo to receive their Kibbe face type prediction. We’re working this week to include current celebrities for comparisons and to include fashion recommendations. We’re excited to show you soon!

Find the Fake:
Finally, if you liked the deep fake video from last week of Jim Carry acting in The Shining, here’s something you may also like, as University of California, Riverside has developed a deep neural network architecture that can identify manipulated images at the pixel level with high precision. The battle is on!