Organizational models for data science teams include dedicated teams, embedded scientists, and hybrid models. Key skills for data science teams include both technical abilities and soft skills like communication and problem solving. Challenges to success include executive sponsorship, training, knowledge sharing, understanding business context, and data access. A case study at Comcast developed an automated media planning tool called Pronto by translating a business need into a data science project, testing prototypes with real data, and gaining executive support through proof of concept. Keys to successful deployment included executive buy-in, collaborating across teams, measuring adoption, and focusing initially on critical use cases.
4. Organizational Models
Dedicated Data
Science Team
Serve as a
knowledge center
for Data Science
Consult with
Business Units as
Needed
May not be fully
knowledgeable on
business problem
contexts
Embedded
Data Scientists
Working directly within
the business units they
support
Highly knowledgeable
around business context
May not be able to take
advantage of knowledge
and development
efforts of a broader
technical team
Hybrid Model
A technical team of
data scientists
engaging with
embedded data
science staff
A highly technical
team seeking to
benefit from staff
with in-depth
business context
7. • Translating business
problems into Data
Science Opportunities
• Translating Data Science
findings into actionable
business results
8. Organizational Keys
• Structure Purposefully
• Select an organizational model that makes the most sense
given the context of the business goals and the resources
available
• Diversify
• Data Science teams that include a diversity of technical and
non-technical skills will be best positioned for innovation and
growth
• Don’t ignore ‘softer’ skills
• A Data Science team will only be as effective as their ability to
translate to the business context
• Establish Deep Cross-functional Relationships
• Opportunities to innovate can only be identified with a deep
understanding of core business problems
10. • Executive Sponsorship
• Ongoing Training
• Internal Knowledge-Sharing
• Focused Efforts to understand
business context
• Automation of Common Tasks
• Generalization of Applications
for re-use
• Access to Data
11. Making the Case for Data Science Development
• Case Studies are critical to gaining buy-in from the organization to pursue
innovative new projects
• Potential projects must have alignment to core business goals
• Impactful projects not related to business priorities will be DOA
• Case Studies are more about the business impact than the algorithm of choice
• Key is to prove there is a method for solving the problem at hand
• Implementation and execution details are critical
• Innovation that requires major process or staffing changes must be well
thought out before an investment is made to proceed
• A proof-of-concept test with real data can go a long way to pushing new
projects forward
12. • Bringing Data Science Applications
forward to solve business problems can
bring:
• Automation
• New Decision Processes
• New business roles (reporting,
monitoring, etc.)
• Need for training
• Resistance to Change
13. Challenges to Data Science Integration
• Fear of losing jobs
• Inability to incorporate all use cases
• Process for dealing with exceptions
• Corporate buy-in
15. Media Planning and the Need for Automation
• For many years (and even today), advertising campaigns on
television have been compiled through spreadsheets that employ
manual trial and error methods to meet advertiser goals
• A proposed television campaign must be within the advertiser’s
budget while meeting impression delivery requirements against a
target audience
• A planner with 100 units of inventory needing to place 10 units of
advertising has over 17 trillion combinations to consider in how to
best place those advertisements
20. Keys to Pronto Deployment
• Executives “bought in” after a proof-of-concept made the
business case
• Ongoing testing and development involved data scientists
and engineers sitting side by side with the account
executives that would use the functionality
• Functionality tracking identified when automation was used
vs old processes – showing product adoption success
• Initial focus aimed at handling the majority of critical use
cases
21. Deployment Results
• Algorithmically optimized Media Plans are now generated
within seconds compared to the hours it took someone to
generate a “good enough” plan through spreadsheets
• All plans are using the most updated data available – no
concerns around syncing spreadsheets
• All plan development is tracked against key metrics
automatically, allowing ongoing refinement of the system
• Over 99% of all media plans are now generated using Pronto
23. 5 Keys to Data Science Innovation
• A Data Science Organization with a Purpose
• Access to ‘cleanable’ data
• A deep focus on understanding business context
• Attention to change management
• Executive buy-in for exploration and testing