Outline of the generic process for an end-to-end data science project, beginning with definition of business requirements and ending with value-add brainstorming.
2. 2
At a glance
Requirements
Research
Rationale
• Define business problem
• Determine scope
• Develop high-level strategy
• Choose platform
• Craft algorithms
• Construct solutions
• Vindicate investment
• View / present / publish
• Value-add
• pain points, opportunities
• knowledge, tools, products
• implementation, success
• in-house vs Cloud, O/S vs ™
• pre-processing, modelling
• investigate, hypothesise, test
• explain, prove
• make the data tell the story
• discover, predict, exploit
Technology
Techniques
Tactics
Validation
Visualisation
Vision
3. 3
Define the business problem
Background
Business environment
Generic challenges, best
practice, current R&D
Specific challenges
Problem Statement
What do we want to
understand / fix / build ?
Requirements, Research, Rationale
• High-level client SME’s give
overview
• Low-level SME’s provide
operational constraints
• Root cause analysis
• Decision-support tools
• New product / New industry
4. 4
Determine the scope
Investigation
Questions & answers
Decision-Support Tools
Specificity / generality
Tactical / strategic
Data Products
Library / application
Stand-alone / add-in / service
Requirements, Research, Rationale
• What do we need to reveal?
• How far shall we let the data take
us?
• Domain of application?
• Who is the user?
• Future-proofing?
• Extensibility?
• Commercial quality?
5. 5
Develop the high-level strategy
Data
in-house / open / proprietary
static / longitudinal / streaming
structured / unstructured
Methods
supervised / unsupervised ML
Deep Learning
Outcomes
what should success look like?
Requirements, Research, Rationale
• quality / completeness
• proxies, compound inference
• Big Data
• technological constraints
• theoretical constraints
• metrics / KPIs
• before / after / ongoing
9. 9
Vindicate Investment
Early wins
small, relevant use cases
proof of concept
ROI
sanity checks, KPIs
effectiveness of solution
Empowerment
operational intelligence
strategic intelligence
Validation, Visualisation, Vision
• useful insights
• immediate actions
• immediate value
• lifetime value
• how to convert insight into
maximum business benefit?
10. 10
View / Present / Publish
Proof of solution
field tests
ROI estimates
Graphic presentation
static / interactive
knowledge transfer
Story telling
essential points
implications
Validation, Visualisation, Vision
• test recommendations
• test predictions
• web-based models,
dashboards
• infographics
• explaining the past
• predicting the future
11. 11
Value-Add
Fine-tuning
data selection & pre-processing
parameter adjustment
Expanded use cases
other products, processes
Future directions
new applications
new markets
Validation, Visualisation, Vision
• actuality vs prediction
• feedback from users
• brainstorming with the
wider business
• licensing opportunities
• disruption