A Hybrid Approach to Data Science Project Management
-
Upload
elaine-k-lee -
Category
Data & Analytics
-
view
541 -
download
0
Transcript of A Hybrid Approach to Data Science Project Management
Building a Data-Driven WorldTM
Open Data Science Conference
A Hybrid Approach to Data Science Project ManagementElaine [email protected]@elaineklee
2Open Data Science Conference#ODSC
Organizations want to be data-driven but many obstacles stand in their way:• Communication not trickling up to executives and key decision makers• Silos between departments, making it difficult to share and
collaborate on analysis• Data ingestion (ETL or Extract-Transform-Load) is difficult and time-
consuming• Lack of meaningful, yet customizable visual reporting• Inability to flexibly scale up or down technological needs at a
reasonable cost• Inadequate or overwhelming learning resources about data science
A Common Problem With Many Faces
3Open Data Science Conference#ODSC
Where should Enroll America direct its insurance signup efforts?
Mapping the Uninsured in America
4Civis Analytics | Proprietary and Confidential
As a company, Civis traces its origins to the 2012 Obama for America analytics team.
We built a scientific understanding of each voter. Our data science influenced every strategy and tactic: voter targeting, messaging, media buys, and fundraising.
This meant the campaign could allocate resources where impact would be greatest.
We ran the first individualized presidential campaign
Civis Analytics | Proprietary and Confidential Open Data Science Conference#ODSC
5Civis Analytics | Proprietary and Confidential
Today, we leverage data science to help our clients in politics, non-profits, and the corporate world.
Civis Analytics | Proprietary and Confidential Open Data Science Conference#ODSC
Open Data Science Conference#ODSC Open Data Science Conference#ODSC
An easy-to-use,end-to-end, incredibly extendable, data science platform in the cloud for teams who want to make great data-driven decisions to drive their organizations forward.
Introducing Civis
7Open Data Science Conference#ODSC
The Civis Approach
ProductConsulting R&D
Applied Data Science• Tackles the toughest data
science problems we can find
Data Science R&D• Generalizes and
automates the solution for many scenarios
Software Engineering• Integrates solutions into
user-empowering software
• Highly collaborative departments• All departments contribute to both our services arm and product
development
8Open Data Science Conference#ODSC
The Civis Approach
Our unique team structure allows us to solve your biggest problems
with custom solutions and the technology to scale them.
9Open Data Science Conference#ODSC
Strategies and philosophies• Teams based on Civis’s product and consulting needs:
• “Built around code”• Semi-annual departmental day-long off-sites to plan upcoming R&D
initiatives• Academia-influenced: evidence-based approaches to finding and
reporting best solutions• Software development-influenced: standups, code review• Favorite tools:
Data Science R&D
R&D
Modeling Methodology
Unstructured Data Engineering
10Open Data Science Conference#ODSC
Tools
• Share and discuss data science news
• Receive feedback from colleagues using our tools
• Discuss implementation• Lower communication costs
compared to email
Data Science R&D
11Open Data Science Conference#ODSC
Tools
• Prototype new workflows• Used like a log book to record
and present results• Share preliminary results with
members of other departments
Data Science R&D
12Open Data Science Conference#ODSC
Tools
• Department heads set milestones, check progress, and make project staffing decisions
• Collaboratively plan development on new functionality or organizational processes (e.g. recruiting)
Data Science R&D
13Open Data Science Conference#ODSC
Tools
Strategies• Designate “tag team” on R&D as
default R&D resources for client engagements• This is the Modeling Methodology
team• Other R&D teams’ members may
be staffed on engagements depending on expertise required
• R&D team member always serves as the Consulted in the RACI model
• Transparency about challenges is paramount
R&D <-> ADS
14Open Data Science Conference#ODSC
1. Assemble a project team of R&D data scientists and Applied Data Scientists
2. Work with Enroll America to refine requirements and come up with a plan of analysis, ultimately resulting in the design and execution of a phone survey on a sample of individuals, followed by building a predictive model for the rest of the country.
3. The Applied Data Science Manager has weekly calls with Enroll America and status meetings with the project team.
4. The project team delivers the predictions and analysis to Enroll America.
R&D <-> ADS: A Case Study
Mapping the Uninsured in America
The project team completes a postmortem and determines these activities could be automated: model building
15Open Data Science Conference#ODSC
Tools
Strategies• Designate teams at the interface
to triage issues and plan new development:• R&D: “Engineering” team• Tech: “Modeling” team
• Use module or project-specific chatrooms to get answers to ad-hoc questions quickly
• Identify opportunities to form cross-functional teams, e.g.:
• Developing apps using the Platform’s API
• Knowledge sharing on best practices
R&D <-> Tech
16Open Data Science Conference#ODSC
1. After the postmortem for the Enroll America engagement, R&D begins prototyping automated modeling functionality and discussing its implementation with the Tech department.
2. R&D’s Engineering team finishes the prototype and works with Tech’s Modeling team to integrate it as a new feature in the Platform.
3. During integration, ad hoc discussions occur on GitHub and Hipchat to address usability questions, e.g. resource usage and input/output specifications.
R&D <-> Tech: A Case Study
Mapping the Uninsured in America
The integration team successfully builds and integrates the Build Model module in the Platform.
Open Data Science Conference#ODSC
Our approach to data science consulting and product development is enriched by valuable perspectives of our
employees, who come from a wide array of backgrounds, making our project management
strategies a hybrid of more conventional techniques.
Conclusion