Addressing the challenges and issues with businesses struggling to deliver successful Data Science in environments - Measure Camp, Bucharest (2 Nov. 2019)
1. Data Science Strategy & Structure
within organisations
Addressing the challenges and issues with
businesses struggling to deliver successful
Data Science in environments
Sjaun Wong – 2 Nov. 2019
Measure Camp Bucharest
2. • Over 12 years experience in Data Science Engineering, Business
Intelligence, Digital Analytics and Big Data.
• Currently, leading the technical delivery of the Marketing
Analytics Big Data Platform for Samsung Electronics UK.
• Former Global Head of Business Insights for GVC,
a FTSE 100 gaming company in the UK.
• Previous analytics roles for Comic Relief, Havas Media Agency,
Betfair, CallCredit and Brave Bison.
SJAUN WONG
Analytics Director - Digital Science Consulting
@SjaunWong
SjaunWong
3. We are data consultancy specialising in:
⬛
Data Science / Machine Learning
⬛
Big Data
⬛
Business Intelligence
⬛
Data Modelling and Design
⬛
Digital Analytics
⬛
Digital Transformation
⬛
Single Customer View Technology implementation for
marketing stacks
Digital Science Consulting
4. Agenda
@SjaunWong
SjaunWong
• Trends & Challenges: Big Data & Data Science
• Data Science Team Structure Options
• People, Process and Technology Matrix
• The advantages of owning your analytics strategy
5. Trends & Challenges: Big Data & Data Science
@SjaunWong
SjaunWong
* Source: Gartner Blog, Our Top Data and Analytics Predicts for 2019: https://blogs.gartner.com/andrew_white/2019/01/03/our-top-data-and-analytics-predicts-for-2019/
* Source: Gartner Blog, Our Top Data and Analytics Predicts for 2019
80% of AI projects will be run by
wizards (i.e. Data
Scientists) whose talents will not
scale in the organization (by 2020)*
* Source: Gartner Blog, Our Top Data and Analytics Predicts for 2019
80% of analytic insights will
fail to deliver business
outcomes (by 2022)*
20% reduction in IT specialists
required for AI-enabled
automation in data management
(by 2023)*
* Source: Gartner Blog, Our Top Data and Analytics Predicts for 2019
70% of AI workloads will use
application containers or be built on
serverless programming model
necessitating a DevOps culture
within embedded teams (by 2023)*
* Source: Gartner Blog, Our Top Data and Analytics Predicts for 2019
6. @SjaunWong
SjaunWong
Trends & Challenges: Big Data & Data Science
Source: Domino - Key factors on the journey to become model-driven: A Survery Report of over 250 Data Scientists:
https://www.dominodatalab.com/wp-content/uploads/DSM-Survey-Report.pdf
7. For Commercial team options for Data Science @SjaunWong
SjaunWong
There are 3 broad groups, below is Google’s* definition (edited for brevity):
● Centralised
`In this structure, the “center” establishes and documents all guidelines and processes e.g. a multinational
organization with many offices around the world may not be able to support a robust data science group
in every market. Larger companies also tend to use the center-of-excellence model, but it can come with
a risk: You might lose your connection to local markets.`
* 3 ways to build a data-driven marketing team: https://www.thinkwithgoogle.com/marketing-resources/data-fueled-marketing-team-strategy
● Decentralised
`The organization embeds analysts within individual divisions throughout the company. That way, analysts
can gain intimate knowledge of a team’s priorities and processes.
The model allows top management to remain focused on overall goals and strategies without getting
bogged down in micromanaging data.
But one downside: people may lose touch with the big picture if there isn’t a clear, integrated data
strategy.`
● Hub & Spoke / Hybrid model
`The hub-and-spoke model is an excellent hybrid approach that blends the best parts of the previous
models. In this model, there’s a central point of contact or team, as well as embedded analysts.
Meanwhile, the analysts within each brand or division implement them and return results to the core team,
usually with a dotted-line reporting structure.This model encourages coordination between divisions and
central management, while also empowering local teams to innovate, explore, and take risks.`
8. Data Science Team Structure Options @SjaunWong
SjaunWong
For Commercial teams there are broadly three types of team structures:
Time to
delivery
(of projects
from new)
6-12 months minimum:
As many corporate
functions need
integrating with new tech
& team members
3-6 months minimum:
⚫
Requires a self-service (container)
environment culture
⚫
Focus on commercial performance and not IT
Total Cost of Ownership etc.
1-3 months / 6-12 months:
⚪
With experienced teams this can be quick but
very use case specific (not reproduceable).
⚪
Legacy tech. issues means DS teams focus on
fixes, management, upgrades, maintenance etc.
Support
⚪
All support is provided
by IT teams
⚪
More focus on Data
Science rather than
DevOps & maintenance
Support is provided in tiers:
⚫
Hardware & software (OS) support by IT
⚫
Applications / data software by DS team
⚫
Additional out of hours, 1st-line of support can
be provided by IT
DS team fully responsible for support:
⚪
Hardware, software and application
maintenance and support
⚪
Over time your DS team will build out central
IT functions.
Centralised Hub & Spoke (Hybrid) model Decentralised
Control
Work is prioritised on
global objectives.
⚫
Little reliance on IT, primarily environment
support (hardware and software installations
and data stream integrations).
⚫
Business functions have more control and
managed centrally.
⚪
Complete control over the business function
objectives.
⚪
Team completely responsible for their
environments which is a maintenance and
management overhead.
Team
Knowledge
⚪
DS niche experts.
⚪
Domain knowledge and
onboarding takes time
⚫
A good mix of niche DS knowledge and
domain knowledge.
⚫
Alot of the application and hardware
management is abstracted (with self-service)
⚪
DS Teams tend to be less niche
⚪
Strong domain knowledge.
⚪
Require generalist knowledge of multiple
languages, platforms etc.
9. @SjaunWong
SjaunWong
Team
members
⚪
More career development
⚪
More research-focussed by
investigating new DS
solutions
⚪
Drives innovation but many
research areas may lead to
dead-ends
⚫
More agility.
⚫
Less upward career development
⚫
Opportunities for horizontal growth
into other business units encourages
sharing of knowledge.
⚫
Research is more closely related to
results.
⚪
Teams are difficult to scale across the business.
⚪
Multiple DS teams across the business create
more redundancy and fewer global synergies
due to specific team needs, limited budgets,
ownership and budget control etc.
Centralised Hub & Spoke (Hybrid) model Decentralised
Management
Usually reports directly to
C-suite (CTO, COO etc.)
⚫
Lead Data Scientist reports both into
the head of the department
⚫
Has a dotted-line to tech. teams e.g.
project / product management.
⚪
Lead data scientist reports to head of the
department only.
⚪
Budget controlled by the commercial team,
they only share resources if sharing budgets.
Collaboration
⚪
High-level of collaboration
amongst DS team but less
with business units
⚪
Usually, DS team is 1 small
component of IT / Ops
divisions
⚫
Lunch-and-learn sessions are
prevalent.
⚫
These sessions lead to Town Hall
presentations to the entire organisation.
⚫
This educates and builds interest
across other commercial teams driving
growth DS across the business
⚪
Formal collaboration across teams rarely
occurs and sometimes actively discouraged.
⚪
Where teams have a commercial impact they
can be elevated to hero-status amongst senior
managment and across the business
There are broadly three types of team structures you could use:
Data Science Team Structure Options (cont..)
11. What are the advantages to owning your
analytics strategy?
@SjaunWong
SjaunWong
● Encourages a test & learn innovation culture across the team
● Better governance and compliance - Trust me, this could make or break your
project
● Earns you a seat at the Senior Management Team decision table ≈ More budget
● DS agility and better domain knowledge, hence, better business solutions.
● 80% of DS is data preparation not modelling so focussing on this area for
self-service amongst DS is critical.
● Team satisfaction is much higher - commercial teams deal with the political
agendas, challenges etc.