Falsification is a significant risk within any supply chain. The longer the process or the chain, the higher the risk that the customer gets a falsified product or service. Including data science as a tool to extract knowledge may provide insights about the crucial points where the falsification process of products and services takes place within a supply chain. In several papers, this problem is discussed and the main methods used are RFID (Radio Frequency Identification) and CSFs (Critical Success Factors) as well as the emerging block chain technology. The data science, machine learning, and artificial intelligence aspects remain less discovered compared to others and provide a high potential to solve the traceability issue. By incorporating data science and robots in crucial falsification points within a supply chain, traceability can be achieved. Robot will either achieve the tasks with high falsification risk or inspect the output of these tasks once achieved by humans. If the products and services contain unique features that robots can learn through supervised learning and which humans cannot easily falsify, falsification risk can be significantly reduced. By achieving a robot level performance that is higher than the human level performance on this aspect, it will become difficult to unethical people to falsify products and services and their originality will be consequently maintained and sustained.
3. 3
Introduction
Star Conferences 2019
Global Supply Chains
Traceability
Counterfeiting
Data Science
ML & AI
Products / Services
Financial
Information
Did I get a Fake Product
4. 4
Case 1 – Sanlu Milk – Domestic
Star Conferences 2019
Melamine
Fool Protein Tests
300,000 Sick Babies
6 Babies Died
Dairy Industry Devastated
5. 5
Case 2 – Australian Beef – Global
Star Conferences 2019
Cryovac Bag
Different Beef
1Kg Counterfeit vs 1Kg Real
Worth Billions of $
Customers Complaints
6. 6
Case 3 – Amazon – Online
Star Conferences 2019
No Regulation
Third Party
90% to 100% Gross Margin
Little Differentiation
40% of Paid Items
Disappointed Customers Little Incentive for Regulation
7. Literature - RFID
7Star Conferences 2019
Radio-Frequency IDentification
Product Side -> Counterfeiting
Unique Identification
Increased Visibility
Track & Trace
Maintain Data
Counterfeit Tags exist across billions of RFID Tags
8. Literature - Blockchain
8Star Conferences 2019
Linking Blocks Through Cryptography
Information Side -> Traceability
Reliability
Transparency
Disintermediation
Durability
Redundancy – Signature – Consensus
10. Insight – Information Side
10Star Conferences 2019
Information Side
Learn about SC
Data Mining
Set of Points
Identify Critical Points
Available Data
Data Preparation Data Mining
Knowledge ExtractionData Science
11. Product Side
Insight – Product Side
11Star Conferences 2019
Learn about the Product
Machine Learning: Classification, Clustering, Regression
Apply within Critical Points
Intelligent Agent Checks or Executes Tasks in Critical Points
Humans Learn
Teach Agents Better Level Performance
Risk Control & ReductionML & AI
12. Application – Case 1 – Sanlu Milk
12Star Conferences 2019
Issue Between Farm & Factory
Samples from Different Points
Data Science for Critical Points
(Un)Supervised Learning
Apply through Domestic SC
Traceability
13. Application – Case 2 – Australian Beef
13Star Conferences 2019
Teach Agents
Data Science for Critical Points
Classification or Clustering
Apply through Global SC
Issue within the Chinese Context
Learn Australian Beef Features
Regulation
14. Application – Case 3 – Amazon
14Star Conferences 2019
Reviews Data Selection
NLP Model
Classification or Regression
Risk Value on each Product
85%
Identify Critical 3Ps
Peer Review
15. Benefits & Limitations
15Star Conferences 2019
Benefits Limitations
Both Sides Advantage
Humans vs Machines
Data Selection
Data Availability
Current AI & ML State
Technological Trend
Stakeholders Involvement
Investment in Agents
Editor's Notes
Why this topic is important?
First Case – Sanlu: Domestic Case, China, 2008.
Nearly 40% of paid items from Amazon are from 3P merchants.
Talk briefly about other methods (ERP, Contracts, Inspection, etc), how to track information or products in a supply chain?
Slower than Centralized Databases because it carries three additional burdens: Redundancy, Signature Verification, Consensus Mechanisms.
-Redundancy: Whereas centralized databases process transations once (or twice), in a blockchain they must be processed independently by every node in the network.
-Signature Verification: Every blockchain transaction must be digitally signed using a public-private cryptography scheme. This is necessary because transactions propagate between nodes in a peer-to-peer fashion.
-Consensus Mechanisms: In a distributed database such as a blockchain, efforts must be expended in ensuring that nodes in the networ reach consensus.
Movies reviews using NLP and Tfidf, then I got the idea to Scale it to supply chain.
Capstone Project in Bombardier Aerospace: Departments Analysis of tasks execution -> Lazy/Problematic.
Data Science Part:
- Learn about the Supply Chain.
- Utilize Data Mining.
- Define Supply Chain Set of Points.
- Identification of Critical Points with High Counterfeiting Probability or Lack of Traceability such as Packaging, Distribution, etc.
More into the HOW? I should find more insights.
AI Part:
- Learn about the Product.
- Utilize Supervised/Unsupervised Learning.
- Classify Real (1) or Counterfeit (0) Product.
- Apply within Supply Chain Critical Points.
- Machine check the Human output or execute Human tasks in critical points.
Can be applied to any online product or service. Peer reviews verification and selection (real ones) and then utilization.