Data Science + Magic = Profit LILIA GUTNIK @superlilia

An alarm goes off in a nuclear power plant

credit: United States Nuclear Regulatory Commission, 1979

Then another. And another.

credit: United States Nuclear Regulatory Commission, 1979

The first was a small mechanical failure

Cascading failure

Cascading failure of humans

credit: " cascading failure of humans " –Roman Mars, 99% Invisible Podcast

Everything Else vs Data Science

Part 1: Identifying Customer Pain & Designing a Potential Solution

People don’t pay for algorithms. They pay for solutions.

Interviewing Users

credit: Interviewing Users: HOW TO UNCOVER COMPELLING INSIGHTS by Steve Portigal

“Tell me about a time when…”

credit: Interviewing Users: HOW TO UNCOVER COMPELLING INSIGHTS by Steve Portigal

“Imagine the ideal situation…”

Interviewing Users HOW TO UNCOVER COMPELLING INSIGHTS by Steve Portigal

Product Vision: Imagine a good place

credit: Schur, M. (Writer, Director). (2017). Michael’s Gambit [Season 1, Episode 13]. In M. Schur (Executive producer), The Good Place. Universal City, CA: NBC.

Product Vision: Ambitious future

Schur, M. (Writer, Director). (2017). Michael’s Gambit [Season 1, Episode 13]. In M. Schur (Executive producer), The Good Place. Universal City, CA: NBC. Data Science + Magic = Profit

What is your product vision?

credit: http://futurama.wikia.com/wiki/File:Fry_Sleeping.png Schur, M. (Writer, Director). (2017). Michael’s Gambit [Season 1, Episode 13]. In M. Schur (Executive producer), The Good Place. Universal City, CA: NBC.

Predicting a value? Suggesting an action? Recommending something?

credit: http://futurama.wikia.com/wiki/File:Fry_Sleeping.png Schur, M. (Writer, Director). (2017). Michael’s Gambit [Season 1, Episode 13]. In M. Schur (Executive producer), The Good Place. Universal City, CA: NBC.

Supervised Machine Learning

credit for concept: Linda Zhou, Western Digital

Supervised Machine Learning: Classification

credit concept: Linda Zhou, Western Digital

“Using ____________ data I can predict (classify) __________”

credit: Jeremy Schiff, MLConf 2015

credit: United States Nuclear Regulatory Commission, 1979

It appears you are experiencing a nuclear meltdown

credit: United States Nuclear Regulatory Commission, 1979

It appears you are experiencing a nuclear meltdown

credit: United States Nuclear Regulatory Commission, 1979

Part 2: Finding Data, Prototyping, and Testing

Finding data

350 million incidents 589 million notifications 1.7 million people

Finding data

350 million incidents 589 million notifications 1.7 million people 28,000 manual merges 1/100 of 1% of incidents in dataset

Can we find the data elsewhere?

Collaborative Filtering

Collaborative Filtering for Implicit Feedback Datasets, Y.Hu, Y. Koren, C. Volinsky, 2008 Eighth IEEE International Conference on Data Mining

Mini version of the vision

Part 3: Measuring Success, Iterating on Design, and Shipping

Explainability v Complexity

Explainability v Complexity

30 seconds 3 hours $200.00 @superlilia Data Science + Magic = Profit

Explainability v Complexity

Juggling requirements

Juggling requirements

Complexity Accuracy Performance Precision Sellability Scale Customer Expectation Resilience Recall Explainability

Final Thoughts

Data Science + Magic = Profit

Data Science + (Well Understood Customer Pain + Thoughtful Design) = Profit

Data Science is part of the solution

Data Science is not the entire solution

Creativity Customers Design Experiment Data Science Define success Iterate Make it understandable Ship it

Data Science + Magic = Profit by Lilia Gutnik @superlilia

References

NRC Annual Report, 1979, United States Nuclear Regulatory Commission •“Lessons from Three Mile Island: the Design of Interactions in a High Stakes Environment”, A. Roesler, University of Washington, 2009 •Machine Learning for Designers, Patrick Hebron, 2016, O’Reilly Media, Inc •“Collaborative Filtering for Implicit Feedback Datasets”, Y.Hu, Y. Koren, C. Volinsky, 2008 Eighth IEEE International Conference on Data Mining •Human Interpretable Machine Learning - the Need and Importance of Model Interpretation, D. Sarkar, Intel, 2018, kdnuggets.com •Satisficing and Optimizing Metric, DeepLearning.ai, A. Ng, coursera.org, •Recommendation Architecture, J. Schiff, MLConf 2015, New York, NY •Simplify Machine Learning Pipeline Analysis with Object Storage, L. Zhou, westerndigital.com, 2018