
Agents that buy and sell • Maes, Guttman, Moukas • MIT Media Lab
Overview • Physical vs. digital negotiations • General agents overview • Current electronic purchases • Software agents properties and types • Agents as mediators in e-commerce • Future directions
Physical vs. digital negotiations • “Real-world negotiations accrue transaction costs that may be too high for either consumers or merchants” • “At the speed of bits, agents will strategically form and reform coalitions to bid on contracts and leverage economies of scale”
General agents overview • 1st generation agents • Filter information • Match people w/similar interests • Automate repetitive behavior • 2nd generation • E-commerce ==> revolutionize • business-to-business • business-to-consumer • consumer-to-consumer
Current electronic purchases • Automate • Information • Products, vendors • Orders, payment • Merchandise buyer: manually • Seek merchants, products • Enter purchase & payment info
Software agents properties and types • Personalized • Continuously running • Semi autonomous • ==> Optimize buying experience • E.g., monitoring agents …
E.g., monitoring agents • Monitor • Quantity • Usage patterns • Invoke buying agent when low
Buying agents • Collect info: vendors, products • Evaluate offerings • Make decision to investigate: merchants, products • Negotiate terms of transaction w/merchant • Place orders • Make auto payments
Agents as mediators in e-commerce • Buying behavior models & theories • Six fundamental stages of buying process • Role of agents as mediators in e-commerce • Technologies / techniques in agents • Negotiations • Agent systems
Buying behavior models & theories • CBB: consumer buying behavior • Other theories and models • Nicosia • Howard-Sheth • Gugel-Blackwell • Bettman info-processing • Andreasen
Buying behavior models & theories (cont.) • Approximation & simplification • Define similar fundamental stages • Help identify where agent technology apply • Categorize existing systems • Buying process: Six fundamental stages
Six fundamental stages of buying process • Need identification • Product brokering • Merchant brokering • Negotiation • Pruchase and delivery • Product service and evaluation
Need identification • Buyer aware of unmet need • Can motivate through product information
Product brokering • Retrieve info: help determine what to buy • Evaluate product alternatives • ==> Buyer-provided criteria • ==> “Consideration set” of products
Merchant brokering • Combine consideration set w/merchant-specific info • Evaluate merchant alternatives subject to buyer-provided criteria • Price • Warranty • Availability • Delivery time • Reputation
Negotiation • How to settle on terms of transaction • Fixed • Common in consumer products • Negotiable • Common in business-to-business • Duration and complexity
Pruchase and delivery • End of negotiation • Other time (later?)
Product service and evaluation • Post-purchase • Product service • Customer service • Satisfaction evaluation • Overall experience with decision
Need identification • Repetitive • E.g., supplies • Predictable • E.g., habits • How … monitors
Monitors • Continuously running • Monitor sets of • Sensors • Data streams • Take actions • When prespecified conditions apply • Examples …
Examples • Stock market • E-commerce • Amazon.com • Notification agent (“eyes”) • New book by • author • category
Product evaluation • PersonaLogic ... • Tete-a-Tete (T@T) ... • Firefly ...
PersonaLogic • Define product features • Filter unwanted products • Constraints on features • Constraint-satisfaction engine • Return list of products • Satisfy shopper’s “hard constraints” • Prioritize by soft constraints
Tete-a-Tete (T@T) • Comparable techniques • Multiattribute utility theory • Also • Merchant brokering • Negotiation
Firefly • Automated “word-of-mouth” • Collaborative filtering • Compare ratings w/others’ • Identify “nearest neighbors” • Users w/similar tastes • Recommend products rated highly by neighbors • Not yet rated by shopper
Firefly (cont.) • ==> Serendipitous finds • Opinions of like-minded people • Music • Books • More difficult to characterize • Web pages • Resraturants
Technologies / techniques for product evaluation • Constrained-based ... • Collaborative filtering ... • Rule-based ... • Data mining ...
Constrained-based • PersonaLogic • Tete-a-Tete
Collaborative filtering • Firefly • Other
Rule-based • Broadvision, Inc. • Personalize product offerings for individual customers
Data mining • Patterns in customer purchasing behavior • Help customers find products • E.g., Engage
Merchant evaluation • BargainFinder (Andersen Consulting) ... • Jango …
BargainFinder (Andersen Consulting) • Online price comparison • 9 merchant Web sites (at least) • 1/3 blocked • Don’t compete on price only • Also value-added services • Others asked to be included • Want to compete on prices
Jango ... • “Advanced BargainFinder” • Solve merchant-blocking • Requests originate from requestor’s site • Not agent’s
Technologies / techniques for merchant evaluation • Current: build comparison shopping agent • Largely manual, tedious • Virtual database • E.g., Junglee, Inc. • Learning • ==> Semi auto composing of “wrappers” for Web sites • Future ...
Future ... • XML • Mobile agents • ==> Comparison-shopping agents • Flexible • Open ended • Easier to implement
Negotiations • Settle on • Price • Other terms of transaction
Negotiations: current • Business-to-business • Yes • Retail • Mostly fixed • Only last 100 years
Dynamic negotiation of product price • Benefits ... • Impediments ...
Dynamic negotiations: benefits • Don’t need to determine a-priori value of goods, services • Take to marketplace • ==> Limited resources allocated fairly • To those who value them most
Dynamic negotiations: impediments • E.g., auctions • Geographical colocation at auction place • Complicated, frustrating • Extended period • Not fit for impatient / time-constrained • Generally cost too high for both
Dynamic negotiations in digital world • Impediments gone • E.g., • OnSale • eBay’s AuctionWeb • No geographic colocation • Yes: manage own negotiation strategies • Agent tech. can help
Agent systems that negotiate • Auction Bot (U Mich) ... • Kasbah (MIT MediaLab) ... • Tete-a-Tete (MIT MediaLab) ...
Auction Bot (U Mich) • Internet auction server • New auctions • Auction type • Parameters • Clearing times • Method for resolving tie bids • Number of sellers permitted
Auction Bot (cont.) • Buyers & sellers bid • Multilateral distributive negotiation protocol • Advantage: API • ==> Users create own SW agents • Autonomously compete in AB marketplace • Users encode own bidding strategies
Kasbah (MIT MediaLab) • Online • Multiagent • Consumer-to-consumer • User (buy or sell) • Create agent • Give strategic directions • Send off to centralized marketplace
Kasbah (cont.) • Proactive • Seek out buyers / sellers • Negotiate on behalf of owners • Goal • Complete acceptable deal on behalf of user • Subject to set of user constraints
Kasbah (cont.): user constraints • Initial bidding (asking) price • Lowest (highest) acceptable price • Date to complete • Restrictions on parties to negotiate with • Price change over time
Kasbah (cont.) • After match, only valid action • Buying agents offer bid • No restriction on time, price • Selling agents • Binding “yes” • “No”