data science strategy
TRANSCRIPT
Armando Vieira @lidinwise
Good Data Science Bad Data Science
Armando VieiraData Science Consultant
London, Sep 2016
Armando Vieira @lidinwise
The Scenario
• Today 1% software use AI - in 2018 it will be 50%
• AI is achieving human level accuracy in image, video, voice recognition and text
• 90% data was generated last 2 years• Smart devices are connecting everything
However only a few organizations are taking advantage of these forces. Why?
Armando Vieira @lidinwise
Data Science - the fluffy side
“We want to extract value from our 10 Tb of Data”
“We need an applied Data Scie
ntist”
“We want to become data centric organization”
“Data Science will transform our business”
“Our Hadoop cluster handles any data”
Armando Vieira @lidinwise
What’s not Data Science
• It is not Science
• It is not Data
• It is not IT
• It is not about unicorns
• It is not about money
Armando Vieira @lidinwise
How to design a Data Science strategy?
DS strategy should be designed to take advantage of the forces unleashed by AI and data available to refocus the business through careful redesign and integration on data driven processes.
As in a business strategy, it requires a deep understandingof the business and the technology.
There is no template
Armando Vieira @lidinwise
How?
• Requires a long term vision• Backed by highest level decision makers• It requires careful engineer of business
processes• Need an experienced data scientist advisor• It is normally painful• Can only be partially outsourced
Armando Vieira @lidinwise
Wrong data science
Department Diagnostic Why it failed?
Digital Marketing Not cost effective Outsourced
Operations Non integrated CLV Fragmentation
Sales Too many products “Not a DS problem”
Fraud Hard rules - easy to trick Incomplete data
Pricing Use more parameters Too complex to integrate
Short-term thinking , not prepared for secondary uses data, legacy data, no team,Underestimate effort, lack of management support and budget
Armando Vieira @lidinwise
Good Data Science
• From chats to CLTV• Automate CS• Networks effect• Explore, test and learn• Feedback loop
Armando Vieira @lidinwise
DS Check List• Cross functional teams?• Openly discuss failure? How many failures in DS?• Data is ready and consistent?• Open source friendly? Use Github?• Where do you store data: DW, Data Lake, Cloud?• Does any manager understand what are you doing?• Are they ready to learn or unlearn on wrong
Assumptions?• Do DS have a seat in the decision room?
Armando Vieira @lidinwise
How to make it happen?
• Start at the highest level• Have your long-term strategy ready• Recruit a small, but smart team• Don’t underestimate the effort. DS is painful• Start proxy deliveries and long-term goals• Communicate your vision
Armando Vieira @lidinwise
Problems
What data to consider?How to formulate the problem as a DS problem?How to sell the outcomes?How to implement it?Simple vs complex – gains in productivityMaintenance, cost, scalabilityStability – stationary
Armando Vieira @lidinwise
The AI revolution
AI is contributing to a transformation of society “happening ten times faster and at 300 times the scale, or roughly 3,000 times the impact of the Industrial Revolution”.
Armando Vieira @lidinwise
“I was a skeptic for a long time, but the progress now is real. The results are real. It works!” - Marc Andreessen
Armando Vieira @lidinwise
Conclusions
• DS is about changing the culture of your organization• Its not magic & should not be cosmetic• DS is a two side sword: it can potentiate your
business or become a money sink• Put the buzz aside and build a strategy• If you don’t have culture, start to build it• Read my book “Business Applications of Deep
Learning” – 2017.