Based on the MarketShare Brief When is Advertising Unnecessary? - Attribution and Customer Intent


If a customer clicked on an ad and converted, was the ad a success? Not necessarily.

Often, customers will click on an ad as a convenient route to a conversion they would have made regardless—whether that route is a display ad that brought them to a bookmarked landing page, or a coupon link for a product they did not need to be convinced to buy.

For marketers looking to evaluate marketing effectiveness, these “convenient routes” pose serious attribution challenges. This is because it can be difficult to identify which customers followed a path to conversion because of an ad was effective, and which used an ad as a quick path toward a purchase they had planned to make all along.

As a result, marketers can easily give credit to undeserving ads, while ignoring the ads and other factors that genuinely move the needle. According to a MarketShare analysis, that attribution challenge can mean over-crediting marketing channels by as much as 20%, and under-attributing sales influences by over 10%.

If our goal is to identify individuals’ paths as a precursor to multi-stage modelling, our next question must be: are there analytics approaches that allow us to reach this goal?

First- and last-click- attribution both fail the test. Last click attribution gives complete credit to the immediate preceding touch before conversion. First Click attribution, a “close cousin” of last click, gives all credit to the first touch along the conversion chain—partially under the premise that the first measured interaction between the consumer and the brand “kicked off” the path to conversion. Because first and last click alike measure a single point in time—the moment of the click or other first interaction—they cannot be used to be build a “historical profile” of the customer. As such, they are not good candidates for predicting or controlling for customer intent.

Antoher option is matched pairs. Essentially a large test and control study, matched pairs finds sets that are identical except for one attribute, and compares outcomes both with and without that attribute. For instance, to understand the impact of targeted display, matched pairs attribution might look at conversion rates for consumers who have been exposed to all media including targeted display, as compared to the conversion rate of those who have not been exposed to targeted display but who have been exposed to all other media. According to this approach, the difference in conversion rates between the groups should reveal the effectiveness of targeted display.

What complicates this approach is that, in the wild, neatly comparing test and control groups can be prohibitively complicated. In the example above, for instance, we can’t fairly compare customers who are exposed to targeted display with customers who are not, because retargeted customers already come from a group of individuals who were likely to convert. For a true apples-to-apples comparison, we need to find not just individuals who were not exposed to targeted display, but individuals who would be just as likely to express interest as the retargeted customers had, under similar circumstances. That means, in practice, being able to accurately predict which types of consumers are likely to take very specific actions—such as searching and clicking on a sponsored link—and only using those customers in the “control” group. Accurately predicting with this level of granularity is extremely difficult, to say the least.

Further complicating matters, test and control approaches depend vitally on large sample sizes in both test and control groups. Limiting observations to large groups also limits the granularity of insights marketers can obtain. This poses a serious challenge when we’re attempting to predict individual-level behavior.

A final option is discrete choice models. Discrete choice models ask which attributes predict a given action, as well as how changes in those attributes might bring an action closer. This information might include path data, such as prior interactions with the brand’s advertising and website. It also could include segmentation data like demographics and psychographics. In other words, discrete choice models ask “who” the customer is, what actions they’ve taken in the past, and how these may work as predictors of an imminent conversion.

That level of understanding of the individual makes discrete choice modelling particularly useful for evaluating a consumer’s pre-existing intent to buy. Combining this with a multi-stage modelling approach allows for a comprehensive treatment of customers’ predispositions to act and revealed intentions based on past behavior.

For the full story, download When is Advertising Unnecessary? - Attribution and Customer Intent