Industrializing data science: a view from the trenches
- 1. a view from the trenches INDUSTRIALIZING DATA SCIENCE
- 3. Data science projects without a clear industrialization path are just expensive math
- 4. The data science product life cycle EXPERIMENT PRODUCTIDEA INDUSTRIALIZE How to overcome the barrier
- 5. DATA SCIENTISTDATA ENGINEERDATA ARCHITECT BUSINESS Key roles for successful data science DATA TRANSLATOR
- 6. The anatomy of a data science product X �y DATA SCIENTIST ? DATA TRANSLATOR 𝛌𝛌 MODEL
- 7. The anatomy of a data science product 𝛌𝛌 MODEL SERVINGDATA MODEL TRAINING MONITORING DATA ENGINEER DATA ARCHITECT BUSINESS € INSIGHT ACTION VALUE
- 8. From proof-of-concept to product INDUSTRIALIZEEXPERIMENT 3challenges BUSINESS TEAMS INFRASTRUCTURE PRODUCTIDEA FLEXIBILITY RELIABILITY
- 9. There is a difference between interest and commitment When you’re interested in doing something, you do it only when it’s convenient. When you’re committed to something, you accept no excuses, only results. ~ Ken Blanchard Get business committed and not just interested
- 10. P U S H BUSINESS The challenge ? Resources DATA SCIENCE TEAM
- 11. DATA SCIENCE TEAM P U L L BUSINESS PRODUCT MANAGER / OWNER The solution BUSINESS CASE, PROJECT, TARGETS, BUDGET, EXPERTISE Create business pull
- 12. DATA SCIENCE TEAM 1. Who are our end users? 2. Are we solving their problem? 3. Do they want our solution? 1. Is the solution aligned with our strategy? 2. Is there a clear business case? 3. Do we have budget? The solution BUSINESS Asking the right questions
- 13. You build it, you run it Work in multidisciplinary product teams
- 14. EXPERIMENT TEAM BUSINESS The challenge ? Hand-overs
- 15. INDUSTRIALIZATION TEAM The challenge EXPERIMENT TEAM ?? ? Hand-overs
- 16. PRODUCT TEAM BUSINESS The solution CONTINUOUS DELIVERY CONTINUOUS FEEDBACK Product teams
- 17. The solution PRODUCTIDEA BUSINESS PRODUCT TEAM Product teams: evolving composition
- 18. The solution 𝛌𝛌 𝛌𝛌 𝛌𝛌 𝛌𝛌 Product teams: ability to operate and improve PRODUCT TEAM
- 19. And they lived happily ever after Separately Aim for a modular architecture
- 20. DATA ARCHITECT BUSINESS 𝛌𝛌 MODEL SERVINGDATA The challenge ? ? ? Embedding the product
- 21. DATA INTEGRATION LAYER 𝛌𝛌 MODEL SERVING The solution Model-as-a-Service BUSINESS SOURCES CONSUMERS
- 22. PRODUCTIDEA A data science product pipeline 1. Agree on the life-cycle stages of a data science product 2. Install stage gates with measurable criteria 3. Establish and assign responsibilities at each stage 4. Align technological roadmap 5. Execute and evaluate
- 23. Key take-aways Get business committed and not just interested Work in multidisciplinary product teams Aim for a modular architecture Instate a data science product pipeline
- 24. +31 (0) 168 479294 Coltbaan 4C, Nieuwegein @bigdatarep www.bigdatarepublic.nl /company/bigdata-republic info@bigdatarepublic.nl DATA SCIENCE | BIG DATA ANALYTICS | BIG DATA ARCHITECTURES