Intelligent Personalized Trading Agents that Facilitate Real-time Decision- making for Auctioneers and Buyers in the Dutch Flower Auctions Case by Wolfgang Ketter, Eric van Heck, Rob Zuidwijk Rotterdam School of Management, Erasmus University
Mar 31, 2015
Intelligent Personalized Trading Agents that Facilitate
Real-time Decision-makingfor Auctioneers and Buyers in
the Dutch Flower Auctions
Case by Wolfgang Ketter, Eric van Heck, Rob Zuidwijk
Rotterdam School of Management, Erasmus University
Advocate Agents - Slide 3
Dutch Flower Auction Worlds’ largest Association set up by growers
Flower industry is a giant network
– Including: breeders, growers, auctions, wholesalers, retailers, and transportation firms for im- and export.
Import from warmer countries: Israel, Kenya, Zimbabwe etc.
Export to: Germany, UK, France
Advocate Agents - Slide 5
Dutch Flower Auction cont.Retailers
Clock auction
Mediation
Growers WholesalersDutch Flower Auction
Two scenarios:– Push strategy (supply driven) clock auction (70%)
– Pull strategy (demand driven) mediation (30%)
Advocate Agents - Slide 6
Auction Locations DFA has 6 individual auction locations spread
throughout the NL, with a total of 39 auction clocks
Conducts buying transaction
Shipment of goods
Buyer is a retailerwho buys for himself
Client is a retailer
Shipmentof goods
Wholesaler who buys for client
Conducts buying transaction
1st scenario
2nd scenario
Advocate Agents - Slide 7
Clock Auction Flowers transported from cold-storage
warehouse to auction hall on carts.
2-3 clocks per hall.
Sample shown to bidders by ‘raiser’.
Buyers bid using Dutch auction: price starts high and drops fast. First person to stop the clock wins and pays that price. Invented in 1887.
Extremely fast! On average an auction clears every 3-5 seconds.
Advocate Agents - Slide 8
Clock Auction cont.
Advocate Agents - Slide 9
Clock Auction cont. One lot is divided among many commercial transactions
– Lot: is the supply of flowers from one grower
– Commercial transactions (sub lots): are different portions of one lot sold to different buyers
Units for sale Price Quantity Buyer
25 15 ct 5 H. De Jager
20 14 ct 6 T.H. Pietersen
14 18 ct 4 P.J. De Vries
10 13 ct 5 A. Jansen
5 12 ct 5 A. B. De Groot
1 lo
t
5 c
om
me
rcia
l tran
sa
ctio
ns
or s
ub
lots
E.g. 1 lot below is divided into 5 commercial transactions.
Advocate Agents - Slide 10
1
2
5
4
6
3
X
7
8
0
10
20
30
40
50
60
70
80
90
10
9
Name Example values
1. Producer Mr. Z. Boon
2. Product Tulips, roses etc.
3. Unit of currency Cents, 5 cents, 10 cents, Euro’s etc.
4. Buyer Mr. S. Klaasen
5. Units
6. Number per unit Amount of stems per unit
7. Minimum purchase quantity
8. Negative comment on quality
“Nice leaves”
9. Positive comment on quality
“Small water stains”
10. Quality indication A1 (highest quality), A2, B1, B2 (lowest quality)
10
2
Advocate Agents - Slide 11ProductsUnits
(= wagons x units per wagon)
Wagons Units per trolley
Amount of stems per unit
S1, S2, S3, and S4 are sorting methods
S1 = lengthS2 = weight, or diameter
S3 & S4 are dependent on the traded flower
Negative comments
Positive comments
Grower(seller)
Clock
Lot nr. Unit of currency
Amount of stems per
unit
Units
Minimum purchase quantity
Identification number of
the auctioned goods
InformationSub lot nr.
Advocate Agents - Slide 12
Information
Flower type and quality (including pictures and videoclip) Seller (name, background, reputation) Auction clock (price, units) Buyer (identification) Previous transactions Services (logistics, payment, settlement)
– Different buyers have different information needs…
Advocate Agents - Slide 13
Buyer Profiles 1. General wholesaler (cost driven exporter)
2. Specialist wholesaler (differentiator on quality)
3. Large retailer (medium quality retailer)
4. Small retailer (flower shop with specific client wishes)
Competing in different markets => different information needs
Information needed for decision making
– Price information
– Product information
– Transportation information
Advocate Agents - Slide 14
Decision Parameters (1) price (2) product, including quality (3) transportation costs (4) transportation time (5) upcoming auctions (6) market conditions
– More?
Based on this information, must choose price and quantity in about 4 seconds
Advocate Agents - Slide 15
Auctioneer Profile
Auctioneer governs clock trading process, by controlling five parameters
– Speed of clock
– Initial price
– Swingback
– Reserve price
– Minimum lot size
Speed vs. price tradeoff
Advocate Agents - Slide 16
Decision Parameters (1) historical prices (2) quality measures (3) upcoming auctions (4) market conditions
– More?
Based on this information, must make 5 decisions in about 4 seconds
Advocate Agents - Slide 17
Intelligent Software Agents
An agent is anything that:– perceives its environment through sensors – acts autonomously upon that environment through
effectors. A human agent
– Sensors: eyes, ears etc. – Effectors: body parts as hands and legs.
Advocate Agents - Slide 18
Intelligent Software Agents
Intelligent software agents are used for complex decision-making
Intelligent agents monitor the environment, make internal calculations, and act (or recommend actions) autonomously.
Useful for supporting trade at auctions.
E.g. for eBay auctions there are “auction bots”.
Advocate Agents - Slide 19
Example of Ebay auction agent You define your preferences (personalize) and
the agent monitors auctions and buys (acting autonomously) for you.
Advocate Agents - Slide 20
Case Conclusion Case Question:
How can personalized intelligent agents be used at the DFA to enable better decisions for the bidders and auctioneers?
Guiding question: Do agents bring a saving in costs, or a true competitive advantage?
Advocate Agents - Slide 21
Dutch Flower Auction CaseYou might want to explore…
www.floraholland.com (DFA website) Auction bidding software
– www.jbidwatcher.com– www.freedownloadscenter.com/Best/auction-
bidding.html Chapter 2 of Artificial Intelligence: A Modern Approach
by Russell and Norvig, 1995. Maes et al., 1999. Agents that Buy and Sell