Norman M. Sadeh e-Supply Chain Management Laboratory ISR- School of Computer Science Carnegie Mellon University Supply Chain Trading Supply Chain Trading Agents: Competition-Based Agents: Competition-Based Research Research
Norman M. Sadeh
e-Supply Chain Management Laboratory
ISR- School of Computer Science
Carnegie Mellon University
Supply Chain Trading Supply Chain Trading Agents: Competition-Based Agents: Competition-Based
ResearchResearch
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 2
Outline
Morning Session (10-11:30am) What is the Supply Chain Trading Agent
Competition (TAC-SCM)? N. Sadeh Evolution of the Competition, J. Collins The CS-50 Mini-Tournament, M. Benisch
Afternoon Session (1:30-3:00pm) CS-50 Mini-Tournament Results at 2pm Supply Chain Trading Panel
T. Sandholm, J. Collins, M. Fox, M. Benisch and N. Sadeh (moderator)
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 3
Trading is Not Just for Biz Schools
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 4
Automated Trading
On the NYSE, automated trading accounts for over 50% of portfolio trades most weeks As much as 70% in some weeks
eBay proxy bidding
Ariba spend management solutions
CombineNet’s combinatorial optimization sourcing engine
Speed is the key advantage
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 5
So Isn’t This a Solved Problem?
If it’s already taking place, why is it still challenging? Market mechanisms
Multi-attribute negotiation
Competing in concurrent markets
Adapting to opponent strategies Markets with limited number of players
Delegating decision to software programs is not free of risk
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 6
Supply Chain Management
Planning and coordinating procurement, production and distribution activities From raw material suppliers to manufacturers …
to distribution centers …to retailers and consumers
Includes in-bound, out-bound and reverse logistics
Trillions of dollars annually
Can also be viewed as extending to the service sector and to digital products With different characteristics
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 7
The Need for Agile Supply Chain Solutions
Shorter product life cycles
Increased demand for customization
Global competition
Outsourcing & focus on core competencies
Companies need to rapidly adapt their supply chains as business conditions change
Companies that can quickly evaluate a larger number of options and can effectively interoperate with others will win
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 8
How Do You Evaluate Supply Chain Trading Solutions?
Evaluating techniques against a pre-specified set of solutions does not cut it
Need to capture the competitive/strategic nature of supply chain trading
In 2003, our group launched the annual Supply Chain Trading Agent Competition (“TAC-SCM”)
20 teams from around the world in 2003
Over 30 teams in 2004 and 2005– around 150-200 researchers worldwide
Game has been refined in collaboration with SICS and U. of Minnesota.
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 9
Underlying Research Objectives
Explore future possible supply chain trading scenarios Understand tradeoffs associated with
different market mechanisms and contractual relationships
Develop competitive and robust trading strategies
Study emerging behavior resulting from multiple organizations competing in overlapping customer and supplier markets
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 10
The Supply Chain Trading Agent Competition
Six PC assembly agents compete in each game Each has access to the same suppliers and customers Each agent has the same finite capacity
A game lasts 220 days, each day is simulated in 15 seconds agents bid on customer orders and negotiate supplies assemble and ship PCs Manage finances
Customers
RFQs & orders
Delivery schedule
Productionschedule
RFQs &orders
Supply ChainManagement
Manufacturers
Suppliers
BANK
RFQ RFQRFQ RFQRFQBid BidBid
[RFQ no, price]RFQ
[Desc, Qty, Due Dt, Penalty]
Suppliers
Customers
Manufacturers
Component Market Place
PC Market Place
… …
BANK
Bid BidWinning Bid
[Bid no, price]RFQ
[Desc, Qty, Due Dt, Penalty]
… …Suppliers
Customers
Manufacturers
Component Market Place
PC Market Place
Suppliers
Customers
Manufacturers
Component Market Place
PC Market Place
BANK
Bid BidBid
[RFQ no, price]RFQ
[Desc, Qty, Due Dt, Penalty]
RFQ RFQRFQ RFQRFQ RFQ [Desc, Qty, Due Dt]
Offer OfferOffer
[RFQ no, Price,Qty,Due dt]
… …
Suppliers
Customers
Manufacturers
Component Market Place
PC Market Place
BANK
Bid BidBid
[RFQ no, price]RFQ
[Desc, Qty, Due Dt, Penalty]
RFQ RFQRFQ RFQRFQ RFQ [Desc, Qty, Due Dt]Offer Offer
Accepted Offer[RFQ no, Price,
Qty,Due dt]
… …
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 15
TAC-SCM Challenges
Concurrently compete in multiple markets with interdependencies and uncertainties
Managing capacity in an environment of constantly changing component supplies and customer demand
Adapt to adversary strategies
Suppliers
Customers
PC assembly
What supplies do I buy? When?
Who should I sell to? At what price?
How should I best use my capacity?
How should I manage my inventory? Too much is costly, too little and I lose customers
I can’t afford to be late, or I pay
customer penalties.BANK
How’s my bank balance?
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 16
TAC-SCM Tournament
Qualifying rounds (8-10 days)
Seeding rounds (8-10 days)
Final rounds over 3 days
Hundreds of games pitting many different combinations of agents under many different market conditions
Copyright ©2006 N. Sadeh CS-50 Supply Chain Trading Agents Slide 17
CS-50 Mini-Tournament
A selection of 6 agents that reached the final rounds in 2005 GoBlueOval – Ford/U. of Michigan (USA) MinneTAC – U. of Minnesota (USA) PhantAgent – U. of Bucharest (Roumania) GeminiJK – U. of Gelsenkirchen (Germany) Mertacor – U. of Thessaloniki (Greece) CMieux – Carnegie Mellon U. (USA)
Total of 15 games