Morphing Banner Advertising

Urban, Glen L.; Liberali, Guilherme; MacDonald, Erin F.; Bordley, Robert F.; Hauser, John R.

doi:10.1287/mksc.2013.0803

Cited by 98 publications

(62 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Our findings have immediate marketing implications, as they emphasize the importance of the interaction between context and ad creative in optimizing online advertising campaigns. This problem relates to other work at the intersection of online advertising, online content optimization, and MAB problems (Agarwal et al 2008;Hauser et al 2014;Scott 2010;Urban et al 2014). Like those studies, we downplay what the firm learns about specific ad characteristics (e.g., which ad message or which format works best?)…”

Section: Introductionmentioning

confidence: 88%

“…The Gittins index has been applied recently but sparingly in marketing and management science (Bertsimas and Mersereau 2007;Keller and Oldale 2003;Hauser et al 2009Hauser et al , 2014Meyer and Shi 1995;Urban et al 2014). We recognize that Gittins (1979) optimally solved a classic sequential decision making problem that had attracted a great deal of attention (Berry 1972;Bradt et al 1956;Robbins 1952;Thompson 1933;Wahrenberger et al 1977) and had been previously thought to be intractable (Berry and Fristedt 1985;Gittins et al 2011;Tsitsiklis 1986;Whittle 1980).…”

Section: Gittins Indexmentioning

confidence: 99%

“…While we know our problem setup violates the formal conditions under which a Gittins index is guaranteed to be optimal (e.g., one-at-a-time updates without batching), we include it to see how much those violations affect the performance of this well-known policy, especially since it has been used recently in marketing (Hauser et al 2009Urban et al 2014). The basic UCB policy also does not optimally account for the correlations among actions, batches, and unobserved heterogeneity.…”

Section: Performance Of Index Policies and Heuristicsmentioning

confidence: 99%

“…As digital environments facilitate randomization, controlled experiments, known as A/B tests, have become an increasingly popular part of a firm's analytics capabilities (Anderson and Simester 2011;Davenport 2009;Donahoe 2011;Hauser et al 2014;Urban et al 2014). As a result, many interactive marketing firms are continuously testing and learning in their market environments; however they are bypassing a more profitable option: firms could be earning while learning.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Customer Acquisition via Display Advertising Using Multi-Armed Bandit Experiments

Schwartz

Bradlow²,

Fader³

2017

Marketing Science

198

View full text Add to dashboard Cite

Firms using online advertising regularly run experiments with multiple versions of their ads since they are uncertain about which ones are most effective. During a campaign, firms try to adapt to intermediate results of their tests, optimizing what they earn while learning about their ads. Yet how should they decide what percentage of impressions to allocate to each ad? This paper answers that question, resolving the well-known "learn-and-earn" trade-off using multi-armed bandit (MAB) methods. The online advertiser's MAB problem, however, contains particular challenges, such as a hierarchical structure (ads within a website), attributes of actions (creative elements of an ad), and batched decisions (millions of impressions at a time), that are not fully accommodated by existing MAB methods. Our approach captures how the impact of observable ad attributes on ad effectiveness differs by website in unobserved ways, and our policy generates allocations of impressions that can be used in practice. We implemented this policy in a live field experiment delivering over 750 million ad impressions in an online display campaign with a large retail bank. Over the course of two months, our policy achieved an 8% improvement in the customer acquisition rate, relative to a control policy, without any additional costs to the bank. Beyond the actual experiment, we performed counterfactual simulations to evaluate a range of alternative model specifications and allocation rules in MAB policies. Finally, we show that customer acquisition would decrease by about 10% if the firm were to optimize clickthrough rates instead of conversion directly, a finding that has implications for understanding the marketing funnel. This paper answers that question, resolving the well-known "learn-and-earn" trade-off using multi-armed bandit (MAB) methods. The online advertiser's MAB problem, however, contains particular challenges, such as a hierarchical structure (ads within a website), attributes of actions (creative elements of an ad), and batched decisions (millions of impressions at a time), that are not fully accommodated by existing MAB methods. Our approach captures how the impact of observable ad attributes on ad effectiveness differs by website in unobserved ways, and our policy generates allocations of impressions that can be used in practice.We implemented this policy in a live field experiment delivering over 700 million ad impressions in an online display campaign with a large retail bank. Over the course of two months, our policy achieved an 8% improvement in the customer acquisition rate, relative to a control policy, without any additional costs to the bank. Beyond the actual experiment, we performed counterfactual simulations to evaluate a range of alternative model specifications and allocation rules in MAB policies. Finally, we show that customer acquisition would decrease about 10% if the firm were to optimize click through rates instead of conversion directly, a finding that has implications for understanding the marketi...

show abstract

Section: Introductionmentioning

confidence: 88%

Section: Gittins Indexmentioning

confidence: 99%

Section: Performance Of Index Policies and Heuristicsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Customer Acquisition via Display Advertising Using Multi-Armed Bandit Experiments

Schwartz

Bradlow²,

Fader³

2017

Marketing Science

198

View full text Add to dashboard Cite

show abstract

“…HULB use data from a "calibration study" to simulate what would had happened had the BT Group implemented morphing on its broadband-sales website. HULB estimate that morphing would have increased revenue by $80M.Subsequently, Urban, Liberali, MacDonald, Bordley, and Hauser (2014) [hereafter Urban, et al] adapted morphing to banner advertising. The only modification in HULB's algorithm was to account for multiple customer visits to the same website.…”

mentioning

confidence: 99%

Website Morphing 2.0: Switching Costs, Partial Exposure, Random Exit, and When to Morph

2014

Self Cite

View full text Add to dashboard Cite

January 2014Website morphing infers latent customer segments from clickstreams then changes websites' look and feel to maximize revenue. The established algorithm infers latent segments from a pre-set number of clicks and then selects the best "morph" using Expected Gittins' Indices. Switching costs, potential website exit, and all clicks prior to morphing are ignored.We model switching costs, potential website exit, and the (potentially differential) impact of all clicks to determine when to morph for each customer. Morphing earlier means more customer clicks are based on the optimal morph; morphing later reveals more about the customer's latent segment. We couple this within-customer optimization to between-customer Expected Gittins' Index optimization to determine which website "look and feel" to give to each customer at each click. We evaluate the improved algorithm with synthetic data and with a proof-of-feasibility application to Japanese bank card-loans. The proposed algorithm generalizes the established algorithm, is feasible in real time, performs substantially better when tuning parameters are identified from calibration data, and is reasonably robust to misspecification. Keywords:Automated Bayesian inference of latent cognitive-style segments with dynamic programming optimization to match website designs to customers. Bayesian inference on a customer's clickstream infers probabilities that the customer belongs to the latent segments. Using these probabilities and data from past purchases, the dynamic program automatically selects the best look and feel for the website for each customer. The "morph" assignment is (near) optimal in the sense that it balances learning about the best assignment for a segment with the profit that can be obtained by exploiting current knowledge of morph-to-segment purchase probabilities. HULB use data from a "calibration study" to simulate what would had happened had the BT Group implemented morphing on its broadband-sales website. HULB estimate that morphing would have increased revenue by $80M.Subsequently, Urban, Liberali, MacDonald, Bordley, and Hauser (2014) [hereafter Urban, et al.] adapted morphing to banner advertising. The only modification in HULB's algorithm was to account for multiple customer visits to the same website. In a field test with over 100,000 customers viewing over 450,000 banners on a CNET website, they report that banner morphing almost doubled click-through rates relative to a random assignment of banners. They also conducted a laboratory test on an automotive information-and-recommendation website to test the basic concept of morph-to-segment matching. The experiment replaced the automated algorithm with direct measurement in a 4-5 week longitudinal study.Click-through rates on banners, as well as consideration and preference for Chevrolet-branded vehicles, increased significantly when morphs were matched to customer segments. The Chevrolet application expanded the definitions of customer segments to include the stage of the automoti...

show abstract