2. Sustainability
Issues
Social
Capital
Human
Capital
Business
Model and
Innovation
Leadership
and
Governance
Environment
• Human Rights and Community relations
• Customer Privacy
• Data Security (GDPR)
• Accessibility and Affordability
• Product Quality and Safety
• Customer Welfare
• Sellking practices and Product Labelling
• Labor practices
• Employee Health and Safety
• Employee Engagement
• Diversity and Inclusion
• Product design and Lifecycle Management
• Business Model resilience
• Supply chain management
• Material Sourcing and Efficiency
• Physical impacts of climate change
• Business Ethics
• Competetive Behavior
• Management of the legal and regulatory
environment
• Critical Incident risk management
• Systemic risk management
• GHG emissions / CO2 , water footprint
• Energy management
• Waste management
• Ecological impact
Value drivers for today‘s customer
3. Features and dimensions
2. Context based learning
(User metadata - age, gender, item metadata - cusine, ingredients seasonality)
1. Collaborative learning
(User / Order similarity, higher data density, rating normalization to de-bias)
Order
History
Restaurant
History
User
Feed
Our product
evolution
User
persona
User
ratings
Custom word
embeddings
Berlin based APIs
Use-specific
Features
X Search
frequency
X
Item
Features
X Weights X Updates to product
strategy
Historical Order
Frequency
X Itemized Sale
history
X
Sorted by time
X Itemized
Order history
X
Current
Search
Pre
processing
Hyper
-parameter
tuning
Model
training
Post
Processing
Evaluation
Prenzlauerberg
Model development pipeline
Recomendations
Objective
Measure conversion of user search
driven recommendations through to
check out and order fulfillment
How do we evaluate conversion? - Use hybrid models
Sample model 1
Measure Weighted Precision and Weighted Recall
Use Weighted F1 as a decision metric, where
F1 = 2 X Precision X Recall / (Precision + Recall)
Wedding
Neukölln
Sample model for creating recommendations
Hybrid models
4. User Instance
Infrastructure requirements on AWS to run reccomendation models
Redis
Lambda
Click-Stream
EC2 SNS
Current Search
Recomendations
(Recos)
Customer
Data
Order and
Inventory
Redshift Amazon DMS
Schema 1 – product viewed, clicked, added and fulfilled
Schema 2 – product detail, displayed price, search parameters
Schema 3 – availability, delivery time, customization and preferences
Generic
Recos
Add to carts Purchases
Similar
Products
Cache
Sniff out
recommendations and
possible ones with high
probability of influencing
customers
Amazon S3
Fallback
Recos
Lambda
consumer
Redshift plays key role in
Collating Schema and
ETL workbook loads
What is a
reco?
One set of probable
recommendation that will
support customer need,
prioritization is still a
business decision.