How to set up for success with Predictive Analytics

Keen IO Data Engineers, Peter and Maria walk us through the steps you'll need to take in order to have a successful predictive analytics project.

Transcript

1. HOW TO SET UP FOR SUCCESS WITH PREDICTIVE ANALYTICS https://keen.io @keen_io
2. WHO WE ARE Maria Dumanis - @zelashoe Data Modeling Architect, Keen IO Helps enterprise customers model data that enables them to answer specific questions Peter Nachbaur -@peternachbaur Analytics Platform Architect, Keen IO Early Vikeen, helped design and build the platform.
3. KEEN : having or showing an ability to think clearly and to understand what is not obvious or simple about something. ref - Merriam Webster
4. WHAT IS KEEN IO? SMART DEVICES MOBILE APPS WEBSITES TEAMS CUSTOMERS ANYWHERE CLOUD DATABASE + ANALYTICS APIS events insights page loaded ad viewed link clicked purchase completed article shared error returned video played count sum min max average median percentile funnel extraction streaming
5. Be proactive and not just reactive - prevent or promote outcomes
6. Understand How How will you acquire and collect the right data? How will you analyze and transform it to be used with predictive models?
7. Champion - You need to ensure that you have someone on your team that understands why you are doing what you are doing and will enable you to do what you need to do Stakeholders - You need to understand who your stakeholders are to ensure that they can voice what type of information they need to make informed decisions Diverse Team - You need to have a team with various domain expertise to ensure that the many components that allow you to perform analytics can interact with each other Creating a Team
8. Know what data can be collected Architect the model to provide you answers to your questions Optimize query performance Performance is Important
9. Data Modeling Tracking Signups • User Ids, Geolocation, User Referral, Acquisition Cost, Cohort Tracking Page Shares • Cookie IDs, Shared URL, Topic Tag Video Plays • Duration watched, Time of Day
10. Know Your Tools Keen is great at modeling event data over time Relational DBs are great for capturing the present state of entities The Hadoop ecosystem is great for joining and crunching Keen and Relational DBs together
11. Deciding which analytics tool/API to use Combining data from various data sources Maintaining data integrity and minimizing duplication Sending data as soon as it’s available Understanding data privacy and who is allowed to access it Enriching data if needed/possible Scaling to accommodate business/product growth Enabling integration with previously collected data Collection Challenges
12. Once you have all this event data, then what?
13. Figure out what the data is already telling you.
14. Agree on your business problems.
15. Identify how a specific prediction could help.
16. Be sure to define success.
17. Build a pipeline to feed your predictive algorithm.
18. Unleash the beast!
19. Iterate, iterate, iterate.
20. With great power comes great responsibility…
21. Predictive Analytics cannot exist in a vacuum.
22. Peter Nachbaur Analytics Platform Architect pwn@keen.io @PeterNachbaur QUESTIONS? Maria Dumanis Data Modeling Architect maria@keen.io @zelashoe