How retailers can use small experiments to improve product choices
Small experiments with product assortments can help retailers learn customer preferences and reduce revenue loss.
Retailers can learn customer preferences by experimenting with different assortments. The cost of each experiment and the operational burden of changing product offerings can make extensive experimentation impractical.
The study uses the multinomial logit model. This model considers known product revenues and unknown customer preferences.
The researchers aim to minimize worst-case regret. This is the expected revenue loss of the final assortment relative to an optimal assortment under the true preferences.
Offering a single experimental assortment to just one customer can substantially reduce regret. This depends on how the products' revenues compare.
A single fixed experimental assortment offered to many customers can yield no improvement or nearly complete regret reduction. The study also looks at multiple experimental assortments chosen in advance.
Key facts
- A single experimental assortment offered to one customer can reduce regret.
- The study characterizes the minimax regret when no experimentation is possible.
- Multiple experimental assortments can allow worst-case regret to vanish as customer observations grow.