In today’s Tech Tuesday, I talk about ways to save money on data science and suggest three things you could do right now.
I’m assuming the U.S. presidential election is likely garnering all of your attention. In light of that, I’ll keep it brief so you can get back to endless results prognostications :).
Saving Money on Data Science
Want to get a lot more value out of analytics initiatives without spending a lot more time and/or money? Most companies do, and it’s entirely possible by reusing one model to service many business verticals.
An example: Let’s say the next big movie streaming service spends six months building a product recommender to show you the next great show to watch. They roll it into a “row” called “Recommended for you”. It works well, and engagement goes up as users efficiently discover new content they like.
That’s all rather straightforward. Hulu, Netflix, and Amazon Prime all have this.
But what happens when the marketing team wants to send out emails, personalized to each user, showing them what other similar users like to watch? This is called collaborative filtering recommendation.
The marketing team has no way of ripping through 3MM users and determining what they should each watch next. But the analytics team does, because they just spent the past six months developing and deploying that exact model!
In this case, it’s only a small modification to the original recommender model to get it to batch process all the shows each person might want to see. Those can be bundled up and sent over to the email service provider for tomorrow’s marketing email blast. It’s then trivial to set this up to run weekly.
And voilà, a predictive model was reused – increasing user engagement and saving money as well as time.
A new study from BCG says, “businesses that saw the biggest gains from A.I. knew when to alter [predictive] modes to suit different kinds of situations.”
Reusing models sounds obvious, but is non-trivial and often overlooked.
The key to increased value from data science models is organizational structure and alignment.
I’ve talked about this a lot at conferences and during 1:1’s with company leaders. Wider exposure of data science leads to greater adoption of predictive analytics, in turn leading to greater automation of human-in-the-loop services such as recommendations, churn prediction, and propensity scoring.
Three Things you can Do
Want to get more out of your data science spend right now? Here are three things you can do:
- Maximize exposure high up in your company by having analytics team report to the C-suite, preferably the CEO, and not the CTO
- Distribute knowledge in two directions by including analytics staff in major weekly stakeholder meetings
- Have wide and early conversations to establish analytics contracts with stakeholders and look for opportunities to reuse models from the start
These are all things I’ve done for clients in the past and they have reaped huge benefits.
I hope this helps! Hit reply if you’d like to talk about any of this.
Smart Data Visualizations: Quality Assessment Algorithm
The gap between a bad and good data visualization is small. The gap between a good and great data visualization is a vast chasm! What we need instead is a rock solid understanding of the updraft faced in our quest for greatness, and a standard framework that can help us dispassionately assess quality. This article by Avinash Kaushik will show you how to separate bad from good and good from great, using examples that we can all relate to instantly.
A new Technique Lets A.I. Learn With Practically no Data
Machine learning typically requires tons of examples. To get an A.I. model to recognize a horse, you need to show it thousands of images of horses. This is what makes the technology computationally expensive. Now, a new paper from the University of Waterloo suggests that A.I. models should be able to accurately recognize more objects than the number of examples it was trained on using a process the researchers call “less than one”-shot learning. That could be a big deal for a field that has grown increasingly expensive and inaccessible as used data sets become ever larger.
The Polls are Wrong – The U.S. presidential Race is a Near Dead Heat, this A.I. ‘Sentiment Analysis’ Tool Says
Expert.ai is a company that uses an A.I. technique called “sentiment analysis” to understand the emotions being expressed in social media posts. Their system looks at Twitter posts and other social media comments using natural language processing – a form of A.I. that can understand aspects of language and categorize the sentiment being expressed according to 84 different emotional labels. Their analysis of posts on the U.S. presidential election indicates that the upcoming election may be a much closer contest than many commentators and pollsters believe.