Contenu connexe Similaire à 1000 track 1 groves_using our laptop Similaire à 1000 track 1 groves_using our laptop (20) Plus de Rising Media, Inc. Plus de Rising Media, Inc. (20) 1000 track 1 groves_using our laptop2. Discussion Topics
• Why implementation & productization is so important?
• Organizing your business to operationalize data science –
technology, process and people
• Key Takeaways
• Q&A
3. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Why is Operationalizing & Productization so Important?
Because of three dirty words……..WE DONT CARE
We have delivered 50 new predictive models.................WE DON’T CARE
We have hired 100+ new Data Scientist…………………WE DON’T CARE
We used Deep Learning to solve the problem…………..WE DON’T CARE
We increased revenue by $X or saved $Y ………………WE DO CARE!
The C-Suite and Board only care about improving profit through making or saving $ This
is only possible with “Actionable” analytics that have been implemented or
operationalized.
4. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Visual
needed
Gartner predicts that 2017 will see 60 percent
of big data projects fail. They won’t go beyond
piloting and experimentation phases, and will
eventually be abandoned.
Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
5. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Data Science & Analytics Today
Clear mismatch between
supply and demand
resulting in a 5:1 roles to
candidate ratio 83% of
organization are struggling
to meet their data and
analytic needs from a
staffing perspective
Demand is
High; Supply
is Low
Senior analytic
searches have
increased 39%
since 2013
86% of executives say their
organizations data science
and analytic efforts have
only been somewhat
effective
More then 25% say they’ve
been entirely ineffective
Expectations
are High
“Hype cycle”
is creating
hard to meet
expectations
6. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Gartner Technology Hype Cycle: Where is Data Science & Analytics?
7. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Why Data Science & Analytics Initiatives Fail?
1. Culture
2. Lack of appropriate leadership
3. Lack of appropriate organizational
structure to support data science
analytics
4. Lack of required resources
5. Inability to monetize analytics
6. Poorly set expectations
7. Lack of ability to operationalize
analytics
FAILED
8. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
What do companies need to be
successful in Data Science &
Analytics?
Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
9. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Technology: Companies Need an Integrated “Factory” Approach to Data
Science & Analytics – Increase Speed Without Sacrificing Accuracy
SENTIENCE CORE
Testing, Modeling
Development, Tuning
Enriched Data
Monetization
Data as an
Enabler
Data-as-a-Service
Monetization
Bulk Data
Productization
DATA FACTORY
ANALYTICS
FACTORY
Value Chain
Acquire, Cleanse, Ingest / ETL,
Annotate, Quality Mgmt., Vend
SENTIENCE PROCESSING STACK
Data Governance – Quality Assurance – Rights Management
Data Lake
Rules Engine,
Process
Streaming Device Data
DATA
SOURCES
IoT+ Data
10. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Technology: Leverage New Technology to Automate, Increase Speed
and Reduce Cost
• Cheap storage and compute via
proliferation of cloud technology
• Open source technology
provides free/cheap kick start
capabilities
• Integrated platforms becoming
the norm
• GPU technology provides
immense compute capabilities
opening up new opportunities in
Deep Learning
• Enhanced user interfaces and
self service capabilities
democratizes more basic
analytics – descriptive and basic
predictive
“Cheap storage and compute has changed
analytics forever by reducing cost significantly
and making the ‘impossible’ a reality.”
– CAO Fortune 100
$$
11. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Process: Establish an “End to End” Data Science & AI Workflow
Identify and validate analytics opportunities
aligned to SBG STRAP. Align on user
stories. Understand high level analytics
solutions and commercial viability.
PRE-IDEATION
Develop business case with SBG. Conduct
opportunity analysis. Identify specific data
needs.
IDEATION
Create project plan. Establish cadence of
project. Finalize requirements.
DISCOVERY
Report to SBG on initial findings and on
iterative outputs. Produce demonstrable
product for UAT.
MODELING
DATAACCESS
Output:
Data evaluation report
Data acceptance certification
Implement testable model. Conduct QA,
Validate model.
PRODUCTIZATION
Develop Messaging and Collateral. Create
Launch and marketing Plan. Train Sales and
SBG Customer Service. Determine KPIs.
GO TO MARKET
Put in place a plan for supporting and
monitoring the models and data access and
currency. Implement Customer service plan.
SUPPORT
Enhance / update the solution
- Customer Solution Adoption
- Model Accuracy Management
MONITORING
12. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
People: Evolving Data Science & Analytics Resource Mix / Continuum
EXPERT
ADDITIVE
OPERATIONS
RESEARCH
STATISTICAL
MODELS
MACHINE
LEARNING
TECHNOLOGY-
BASED
DEEP
LEARNING
13. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
People: Finding the Right Skillsets for the New “Data Scientist” is Hard
Common Data Science Deliverables
(aka Actionable Insights)
• Anomaly detection
• Pattern detection
• Segmentation
• Classification
• Prediction
• Scoring and ranking
• Optimization
• Forecasting
• Simulation
• Automated processes
• Data driven decision-making
• Sensitivity analysis
COMMUNICATION
STATISTICS PROGRAMMING
BUSINESS
Head
of IT Analyst
Salesperson
Great
Data
Scientist Number
Cruncher
AccountantHot Air
Comp
Sci
Prof
Good
Consultant
Drew
Conway’s
Data
Scientist
IT
Guy
R
Core
TeamStats
Prof
Data
Nerd
Hacker
14. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
People: Companies Need an Optimal Organizational Model
“Centralized units and
centers of excellence
outperform other
organization models on
coordinating analytics
initiatives, sharing and
developing analysts’
knowledge, and
deploying analysts
strategically.”
– Analytics Magazine
15. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
People: Honeywell's Data Science Center of Excellence
Chief Data Scientist
& Analytics Officer
Solution Architects
Data Scientists
Data Engineers
Data Architects
Software Developers
Implementation Analyst
Data Engineers
Data Science
Development
Data Engineering Productization Governance
16. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Key Takeaways
• Take action to avoid slipping into the “Trough of Disillusionment”
- Data Science & Analytics needs to deliver value quickly, the clock has already started
- Set expectations appropriately because there is no magic bullet
• Data Science & Analytic leaders need to push our organizations to change
- Embrace new technology led analytics for speed and efficiency in development and implementation
- Shift the mix of our teams to leverage the “new age” Data Scientist – software meets math
- Build new holistic processes and methodologies around the emerging technology to ensure
operationalization and maximum impact of analytics – its not just about the analytic development
• Being good at math is table stakes for Data & Analytic Scientist
- Need to refine your business skills including communication and focus on the impact of analytics to
business strategy
- Need to become quasi “technologist” to stay in front of the technology tidal wave and leverage new
innovations in analytics
1
2
3
17. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Last Thought:
The math is
only the
tip of the
iceberg……
Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
18. Honeywell Confidential - © 2017 by Honeywell International Inc. All rights reserved.
Thank You!
Questions? Comments?
Bill Groves
Chief Data Scientist & Analytic Officer
Honeywell International
william.groves@Honeywell.com
302-981-3060