Limits of Click-Level Data for a Performance Marketer

by Indranil Guha, Michael Pao, & Sunil Raman

Advertisers have always had a difficult time measuring and improving the performance of their campaigns. In the TV era, ad campaigns were believed effective, but minimal methods existed for advertisers to understand whether a specific campaign should be credited for a growth in sales. The evolution and advancement of search has provided an incredible opportunity for businesses to identify and target high value customers who have communicated a very specific intent. Further, they can track these customers much deeper through the purchase funnel to understand whether a lead that was generated by a specific campaign was converted into a paying customer. As businesses new and old try to take advantage of this new customer acquisition channel, they have encountered a new set of difficulties, with a very similar refrain.

Attribution is Tough, Even with Click-Level Data 

Even with search marketing, attribution of marketing efficacy is extremely tough. Although it is very easy to tie a customer to a specific clickstream, the breadth of the online marketing mix and constant changes to the online storefront can create significant noise that can muddy attribution. For example, Ampush Media relies heavily on maintaining a consistent conversion rate across the site. On any given day, they’re managing 500+ campaigns, A/B testing on the site, and managing inventory (matches, i.e., schools seeking students) to make sure that they can continue to pass on high quality leads to their customers. Any one of these factors can throw off the day’s conversion, however all of these pieces move fluidly, and often times are managed independently. It is fairly easy for a company to measure the impact of each of these factors individually, however it becomes difficult when trying to capture the correlations between each element.

In theory, we could build a statistical model / regression analysis to capture the correlations across each element. However, we have found that this is often not feasible or useful for 2 reasons

  1. False precision - factors such as exogenous shocks (e.g., cloud services outages) are inherently unpredictable / non-recurring, so measuring correlation is not productive. Similar issues exist with other elements 
  2. Cost - at web scale (460K+ visitors per month), cost of capturing, managing and processing a dataset for a statistical model can be very expensive. 

Last Click Visibility

Another important factor in accurate marketing attribution is the need to have last click visibility. Put another way, without control of the lead at the point of purchase, it’s impossible to know whether a lead was truly effective. In the TV era, advertisers had 100% visibility at the point of sale, but struggled to attribute that customer to the appropriate campaign. In the online world, the affiliate model has created a delineation between lead generation and lead conversion. The affiliate model serves to qualify and validate leads, however companies like Ampush struggle to understand the full impact of their qualification process and overall conversion effectiveness because they lack last click visibility. Once they throw a lead “over the fence” to an educational institution, they lose the ability to track the effective conversion of that lead. Without that insight, they are unable to understand which acquisition channels over/under-perform.

Truly driving / optimizing marketing efficiency requires 360 degree visibility on how leads perform. However, Ampush hands off leads to customers, resulting in no visibility into ultimate lead conversion except through proxy (e.g., do customers buy more leads on next cycle). In short, Ampush is optimizing for lead cost and implied lead quality. Truly driving lead performance will require optimizing both cost and direct quality measures


Popular posts from this blog

Quiz Time 129

TCS IT Wiz 2013 Bhubaneswar Prelims

The 5 hour start-up: BrownBagBrain