Designing a Better Shopbot

This paper describes optimizing the design of a shopbot (=shopping robot) which considers,

• the intrinsic value of the product,
• the disutility from waiting, and
• the cognitive cost associated with evaluating the offers retrieved

Note: The cost of evaluation refers to the cost of going through the list of offers (not of evaluating the offers themselves).

Considering these criteria the authors determine, whether the consumer maximizes his utility by,

• buying at his favourite store,
• the agent, or
• an optimized version of the agent.

Agent design Agent variables:

• stores to query, (waiting time)
• response timeouts, and (waiting time)
• which items to report (cognitive costs)

Reducing waiting and time and cognitive costs reduce the total utility of the returned items (offer set).

Query times: fat tails (10% of the queries time out after 3 minutes) Radom Utility Model The utility is composed of three parameters:

• Utility of the product
• Costs of the waiting time (query times) and processing time (time to spawn a thread)
• Cognitive effort of evaluating the answer set

Utility is the sum of the utility provided by the attributes of the product and a stochastic component epsilon covering unobservable factors and random revaluation errors. This compensatory utility model allows considering the effect of multiple utility-factors on the total utility (and therefore for example considering trade-offs). The authors use an extreme value distribution with a zero location parameter and a scale parameter theta.

Cognitive Costs are estimated using a heuristic introduced by Shugan (1980):

C=(A-1)(P-1)

Waiting times are modelled as exponentially distributed (page 196). A gamma distribution estimates the response time of requests (provided that the store responses(!)).

The paper also considers consumer heterogeneity and the resulting different assessment of the utility of a product. Afterwards the authors split the decision process into two stages, describe the number of different combinations per stage and provide a solution for choosing an optimal strategy.

Literature

• Choice with uncertain sets (the decision maker decides which alternatives to consider (for the decision) with imperfect information
• Information Overload and Cognitive Costs. Previous research shows, that consumers are willing to trade off cognitive effort for accuracy.
• Usability studies. Delay of more than 10 seconds yields a loss of user attention (Nielson 2000, p44); Users place high emphasis on download times (Udo and Marquis 2001).
• Cost of waiting time. Konana et al. (2000) show that there is a direct trade-off between waiting time and cost. Experiments by Dellaert and Kahn (1999) show that waiting negatively affects the evaluations of Web sites.
The authors assume a value of time of 0.01 $/seconds (based on a yearly wage of $70,000).