Please enable JavaScript.
Coggle requires JavaScript to display documents.
NCR Analytics Data Science Roadmap - Coggle Diagram
NCR Analytics Data Science Roadmap
Cashier Risk Scorer
Transactions Scorer
General Features - 1 week
Tenders
Time Features
Items identity (in terms of prices, departments, identity)
Item scan time + how
Items weights + quantites
is loyalty
Cash Office Transactions - 1 week
cashier over/short
no sale
office over/short
Drawer opens with high round values
Technical
Turn all data analysis to use the new transaction provider
documentation
Add data validation before training - 1 week
returns count
Suspended count
and everything else
including infrastructure
Items analysis - 1.5 weeks
Override Reason
Consider the overridden Item
Check the actions history for items with negative discount
new price with even amount (5.00$)
Consider the items identity
combination with overrides
Item scanning vs. keyed rate (relative to itself)
Abnormal item price (relative to itself)
Item entry method dist.
cashier aggregation
Returns - 1.5 week
Items identity (in terms of prices, departments, identity)
Return by TouchPoint
Even Amounts (5.00$)
Return-refund relationship
Return-Sales relationship
Suspended - 1 week
refer to cases of different cashier/touchpoint for the recall action.
resumed vs. voided research
suspension time analysis
Coupons - 0.5 week
coupons only trans (greater than X$)
manual entered coupons (for future customers)
Coupons types
High value items combined with many coupons.
Evaluation - 1 week
Stability measurement
Models comparison
Models overlap measurement
human evaluation - X
Tailor-made test-cases
Modeling - 1 month must + 1 month optional
literature survey (what models are relevant?)
try different models
tune each submodel
bagging methods
Research Buffer
check performance for another customer (with new trained model and without)
data quantity estimation for new customer
Stoplift exploiting
Interpretation - 2 weeks
baseline approach
try different approaches and evaluate (how?)
Cashiers Scorer
Models - 4 weeks
Global vs. Per Store model
over which time range / number of transactions we should modelize cashier behaviour?
how frequent we should produce scores?
research buffer
Features - 2.5 weeks
Void Transactions
few cashiers had an unproportionally percentage of voided trans
sales per hour (working rate)
Manager Interventions
Tenders + Tenders Exchange
Loyalty
customer-cashier "matrix"
Items weights + quantities
the cashiers transactions scores
"feature" sequences
Evaluation - TBD
loss definition
simulation / data collection
Interpretation - 1.5 weeks
most risky transactions
Data Analysis - 1.5 week
Clustering (after feature construction)
Touchpoints analysis (identify SCO + analyze touchepoints groups)
Managers analysis + cashiers cerdentials
Department analysis
Feedbacks - 2 weeks (TBD)
AI Platform
Monitoring
Research
Development
Templating
add to auto-deployment
Feedbacks
Design
Infrastructure
Customers management
Features Caching
Unittests
Components
Data Validators
Interpreters
Models
Features
Data Transformers
Deployers
Evaluators
Experiement
Enhance Deployment
Extract training to ML Engine
Separate the data querying from its processing
make airflow run from the step it failed
Training - make artifacts generic
Integrate Distributed Computing
CI/CD/CT
CI/CD for training workflow (upload dag + dependencies automatically)
CI/CD for inference workflow
Supporting different types of problems
Visual Models
on-premise models
Recommendations
Auto ML
Graphs
Fine-Tuning
Learn airflow thoroughly and see how to better exploit it.
Learn Kubeflow thoroughly and see if it outperforms airflow.
Scrutinize TFX
Advanced
UI management tool
Configurable Project
labeling
Misc
job that deletes unnecessary models and folders from GCP
Create Projects Template
Sales Prediction
Revenue Prediction
Store Level
Department Level
Different time resoltuions
Interpretation
Recommendation
Item Level
Promotions Recommendations
Revenue estimation
Store level / Global level
relevant time
Items associations research
complementary items
alternative items
Ordering tool
DD Manager Interventions
Review the past work
Data collection
Business Analysis
customers conversations
apriori knowledge
variance est.
"no-touch" rules
approval sensitivity types
intervention necessaity
consider rush hours
Data Analysis
amount
cashier identity
loyalty id
items in basket
time of day/week/year
Project Design
threshold management
Measure past solution
Modeling
Evaluation
Deployment (on-premise?)
Management
Recruiting
Horizon 2