Atlas Insights Series: Gauging the influence of ads over time

By Rebecca Sharpe, Atlas Insights |

We’ve already demonstrated that measuring results beyond the last click gives advertisers a more complete picture of what drives sales, but a key question remains: How far back in time are ads actually influential? Intuitively we believe that an ad impression’s impact diminishes as time passes, but is that really the case? This paper helps to quantify that intuition and argues that ads viewed up to two weeks in the past are predictive of a person’s purchase behavior.

By considering the consumer conversion patterns1 for four Atlas advertisers in the travel, health and e-commerce verticals during June 2014, we found that the optimal “lookback window” should be at least 14 days. That is, when assigning conversion credit via attribution models, ads delivered up to 14 days prior to the purchase should be included for credit assignment. For each advertiser, we used statistical modeling to evaluate the relationship between the amount of time elapsed since ad exposures and the likelihood of a consumer purchase.

Specifically, we used a logistic regression model to assess whether there is a significant correlation between a consumer’s likelihood to convert on a given day and whether ad impressions were delivered to that person on the preceding days. We express the ad exposure time lag in terms of the number of days elapsed. For example, if a person buys on a Monday after seeing an ad on Sunday, the time lag equals one day. We also controlled for other elements of the online display ad delivery that we expect to be correlated with purchase, for example, the device on which the ad was delivered (mobile/desktop), the campaign, publisher, ad frequency, etc. Controlling for these other factors in a regression framework enables us to isolate the impact of time. The result is an estimate of each time period’s contribution to the likelihood a person makes a purchase.

For each of the four advertisers considered, the strongest association was found for ads delivered on the same day—ads seen with a time lag of zero have the highest association with likelihood of purchase. Although same-day impressions have the greatest impact on the probability of a purchase, for all of the advertisers in the study, we also found a significant association between purchase likelihood and ads delivered over the seven preceding days. In fact, we found a significant relationship between purchases and ad impressions delivered as far back as two weeks for some advertisers2.

Lookback windowAdvertisers seek to make the most informed decisions about which elements of their media are working and which ones aren’t. This study demystifies one of the key assumptions that goes into that evaluation: how much history should advertisers consider? The results suggest that considering about two weeks of history is reasonable. If you consider too short a history, you are truncating important information, and if you consider more than two weeks, you are probably weighing yourself down with unnecessary data.

1 We define a conversion as an instance in which a person triggers an online Atlas action tag following the recent delivery of an online ad to that same user.

2 This study considered “lookback” windows of 1–14 days.