Agent-Mediated Electronic Commerce:
Issues, Challenges, and some Viewpoints

H. Nwana, J. Rosenschein, T. Sandholm, C. Sierra, P. Maes, and R. Guttman
Proceedings of the Workshop on Agent Mediated Electronic Trading (AMET'98)
Minneapolis, Minnesota, May 1998.

What are the properties of agents and agent systems that are useful in commerce applications?

Software agents are programs to which one can delegate (aspects of) a task. They differ from "traditional" software in that they are personalized, continuously running and semi-autonomous. These properties make agents useful for a wide variety of information and process management tasks. It should come as no surprise that these same qualities are particularly useful for the information-rich and process-rich environment of electronic commerce.

What particular challenges does electronic commerce pose to agent technology?

Electronic commerce poses a diverse array of challenges and opportunities for agent systems. From a Consumer Buying Behavior perspective, we see agents playing roles in three primary CBB stages: Product Brokering, Merchant Brokering, and Negotiation corresponding to what to buy, who to buy it from, and how to determine the terms of the transaction respectively. It is in these roles that agent system designers have the challenge to accurately identify and the opportunity to prescribe innovative ways of doing business. Of course, there are other opportunities for agents in ecommerce such as helping automate supply chain management and numerous other back-office tasks.

What requirements must be met before we can expect to see fielded systems?

The answer depends on the problem and market, of course. We already see agent systems playing visible roles in retail markets (e.g., Firefly, Jango, Dell's computer configurators, etc.) as well as stock markets (e.g., OptiMark Technologies, E-Trade, etc.) However, for agent systems really to become ubiquitous, some pieces need to be in place first.

The two general goals of ecommerce is (1) interoperation and (2) automation. In many cases, there is a dependency of automation upon interoperation. For example, in order to help automate the management of supply chains, there needs to be a semantically interoperable language and protocol for coordinating the parties involved. Unfortunately, there is currently a lack of common languages and ontologies for ecommerce interoperation.

Although HTML web-scraping may get us by for certain problems (e.g., product information retrieval in retail markets), it is not sufficiently robust to base important business processes upon. As a community with a long history of language and protocol design, we have the opportunity to help define such languages and protocols for ecommerce systems. OPS and ICE are two such examples of this.

Related to this is the need for an open (distributed) registry mechanism which helps tie businesses and ecommerce components together. If not EDI Value-Added Networks (VANs), then we need to find another avenue for managing the evolution of these registries -- e.g., CommerceNet.

In what specific contexts might agents be trusted and empowered to conduct negotiations and make agreements? What are the implications of this? Indeed, how would we build trust and reputation into agent-mediated commerce systems?

We already see agent systems being used to negotiate and make agreements in consumer-to-consumer (e.g., eBay's AuctionWeb) and business-to-business (e.g., FairMarket) and stock markets (e.g., E-Trade and OptiMark Technologies). In order to trust such systems, the agent owners must be assured that the agent will not compromise private information and deviate beyond its constraints. This can occur in the form of insured guarantees (e.g., E-Trade), audits (e.g., TRUSTe), clear controls of strategic and other private information (e.g., OPS/P3P), as well as reputation services (e.g., BizRate and Kasbah's Better Business Bureau).

The implications of agents helping automate negotiations among numerous parties are profound. The first wave will substantially help streamline business-to-business transactions, reducing transaction costs at every stage of the supply chain. At some critical threshold, new types of transactions will emerge in the form of dynamic relationships among previously unknown parties. In this next wave, agents will strategically form and reform coalitions to bid on contracts and leverage economies of scale -- in essence, creating dynamic business partnerships that exist only as long as necessary. The potential is an environment where companies will be at their most agile and marketplaces will approach perfect efficiency.

What kind of warranty has to be provided to customers in order to engage them to delegate their agents to do everyday commerce tasks? How do we develop software that provides these warranties?

For certain, consumers must be assured that their agents will not compromise private information and deviate beyond their constraints. This includes account numbers for payment mechanisms. Essentially, consumers must trust their agents and the environment in which agents carry out their tasks.

As agent designers, we can help instill trust in our agents by having them exhibit (mostly) predictable behaviors and having them be able to clearly explain their actions and decisions. In addition, we must be careful not to inflate the users' expectations beyond our agents' capabilities as this can lead to a lack of trust in our systems.

What will be the marketing or economic consequences of the automation of commerce services and transactions using agent technology?

Thanks to agents such as Andersen Consulting's BargainFinder and NetBots' (now Excite's) Jango, we see the threat of agent technologies to traditional business models. An important question that these systems raise is how do we balance the needs of both consumers and merchants so that we don't turn markets into ineffective arms races of blocking and anti-blocking devices? Is it enough to say "deal with it" or is there also an opportunity to provide smarter tools to balance both sides of the equation? Our Tete-a-Tete system, for example, is helping to differentiate merchant offerings while still providing powerful cross-merchant product comparison tools to consumers.

From a marketing perspective, agent technologies can help realize the elusive one-to-one marketing. Personalization technologies such as automated collaborative filtering and other learning methods can (passively) help consumers identify their needs (stage 1 of the CBB). This can take the form of well-selected banner ads, or better yet, via more innovative means -- e.g., personalized shopping catalogs, effective cross-sale and up-sale offers, etc.

Agents will also have a large impact economically. Agents will likely help shape the new economy by taking over the roles of some traditional intermediaries (e.g., discount stock brokers) as well as reduce the need of other positions (e.g., customer service and sales). However, agents will also be instrumental in creating new markets (more viable consumer-to-consumer and B-goods). Most importantly, agents will help increase sales through the better matching of consumer needs with merchant offerings as well as reduce transaction costs through the semi-automation of business (and inter-business) processes.

What standards need to be in place for agent-mediated electronic commerce to be a reality?

Security, privacy, communication, and ontology standards need to exist for agents to be most effective. What is also important are processes for interoperating in the face of competing standards (which is the more likely scenario) and distributively managing the evolution of these standards. Standards-by-committee will surely be too slow for the potential dizzying pace of ecommerce.

There is also a need of clear legal jurisdiction on agent-mediated business deals and contracts. For example, what audit trails must agents leave in their digital wakes?

What roles will certification authorities, trusted third parties and security/authentification functions play in agent-mediated electronic commerce?

Cryptography and related certification and authentication services will play vital roles in agent-mediated ecommerce. Trusted third-parties will be used as market-makers (e.g., auction "houses") as well as provide related services such as reputation brokering, financial services, escrow services, etc.

How do we ensure our electronic commerce systems are stable, scalable and have adequate performance times?

This is where software engineering skills come into play. However, ensuring all of these is obviously impossible in an environment as diverse, unpredictable, and unstable as the Internet (as we know it today). (This is one argument in favor of secure EDI VANs.)

What state of the art technologies are available to allow agents to 'understand' multimedia information like sounds, images, video and text in order to facilitate commerce within a complex multimedia information system?

Image processing, speech recognition, information filtering and retrieval, and other content-based technologies are slowly making progress to help extract semantics from multimedia. These may one-day help agents find items of interest to their owners or to help accomplish a task.

An alternative approach is to build statistical models of the relationships among multimedia objects (as opposed to their contents). Automated collaborative filtering, for example, enables the recommendation of previously unknown products -- e.g., songs, movies, etc. -- to a consumer through the aggregated ratings of numerous respondents.