Share
Embed
Wasting Money with Data Science
Slides of the presentation I gave at the Amsterdam Dataiku meetup on 01/05/2019
lanzani
May 01, 2019
Transcript
31 slides
-
Wasting Money with DS Giovanni Lanzani — Director of L&D
@gglanzani on Twitter -
What people got 7 • A lot of POCs •
A lot of screenshots / presentations / dashboards on a laptop • Nice stories to tell to their network, about those screenshots and especially those dashboards • Headaches with data and infra even more scattered -
But… 8 • We got a data scientist working on
trees, and forests • Neural Networks! • Deep Learning!!! -
What do companies actually need 9 • Put things of
value into production -
Oversimplifying 11 Requirements Data Sources Exploration Modeling Products Feedback Data
scientist ML engineer Data engineer Data engineer !" Customers -
Kaggle Curse 12 • gdd.li/toldYouSo • Many data scientists approach
the problem at hand with a Kaggle-like mentality: delivering the best model in absolute terms, no matter what the practical implications are. • In reality it's not the best model that we implement, but the one that combines quality and practicality: a continuous balancing act • Netflix competition -
Skills 16 • Participate in actually building production quality systems
OR being proficient enough in R or python to hack together a prototype on a very small dataset? • Supply of the second group keeps growing while demand is flat or shrinking • Especially as executives get burned by “data scientists” who don't know how to help them build things of value -
The value-chain of Data Science Action Value Prediction Data Model
Optimize Measure “Knowledge becomes power through action!” 18 -
More than just a technical challenge Action Value Prediction Data
Model Optimize Measure Business Tech Organizational challenges • Align with business processes • Accept and take action • Measure business value Technical challenges • Collect and store the data • Build a predictive model • Optimize and make actionable 19 -
Start at the value and work your way back Action
Value Prediction Data Model Optimize Measure Value Proposition End User Business objectives 20 -
Finding the fit between the end-user and the analytics value
proposition • What are the business goals you are (currently) aiming for and what are relevant KPI’s? For example, increase of net profit of traditional trade outlets. • What predictions and optimizations can help for achieving these business goals? For example, the prediction of potential outlet profitability. • Who is the end-user that takes actions based on the predictions and optimizations? For example, a sales manager can allocate coolers by using predicted outlet profitability. • What relevant data is available for building the advanced analytics powered product? Ideally very detailed data, delivered as frequently as possible, and with an available historical set of 4-5 year. Can be both internal and external data. 21 -
Example: Business objective Predicting probability of churn of high-value customers
that will optimize overall marketing spend by identifying critical customers that are likely to churn and more efficiently allocating spend 22 -
Before you start the project • Understand the business value
for the stakeholders • Know the end-user you need to empower in creating value • Develop a good value proposition for the end-user 23 Value Proposition End User Business objectives -
Aim for developing a data MVP 24 Delightful Usable Reliable
Functional Model Pipeline Data Delightful Usable Reliable Functional Model Pipeline Data Model Not this… ...but this! -
6-8 weeks to go from idea to data MVP 25
Data & Pipelines Insight Visualizations Predictive Model Data MVP Business Values Data Sources E(T)L EDA Modeling Product- ionize Evaluate Sprint 1 Sprint 2 Sprint 3 PoV -
Starting up your project 26 • Always have a kick-off
meeting with all stakeholders • Plan stand-ups, sprint-reviews and data delivery a.s.a.p. • Engage stakeholders through frequent demos • Scope you project and manage expectations: • Start small and fail fast • Aim for a Minimum Viable Product • Have a clear ‘Definition of Done’ • Plan to test your model in practice a.s.a.p. and measure the value -
About getting the data 27 • Make sure start collecting
the data a.s.a.p. • Get a S.P.O.C. for all the important data owners • Check data quality as soon as you receive data • Make sure your data is GDPR compliant (if in doubt ask) • Remember: filtering out a column is easier them requesting one! • Finally, led the business tell what you should do; the data will tell you if and how you can do! -
Workflow characteristics 28 • Get to a working (simple) model
as quickly as possible • Be agile; short cycles, incremental improvements and result driven • Iterative increments: • mining extra data sources • extension of the input datasets • continued feature and/or target engineering • optimization of the predictive model • Frequent demos of the ‘improved’ model to stakeholders -
The need for analytics translators 30 • The link between
the business case and the data teams is crucial -
The role of analytics translators 31 And how can you
help! • Ensures that business challenge is addressed • Translates between business and analytics • Helps identify use cases • Drives implementation of solutions • Advocates on data driven decision making