Everyday, marketers search for the right combination of search advertising, social marketing and display advertising that will help them achieve a desired result – whether it be visitors to a web site or actual sales. This has made “attribution” a hot topic among interactive marketers.
Simply put, attribution analysis is a method advertisers use to understand the influence and impact individual advertisements and marketing activities are having on campaign results.
The importance of getting attribution right cannot be overestimated. It helps determine how to adjust campaigns going forward. It also provides tremendous insight into how similar campaigns can be optimized to produce better results.
The biggest problem with attribution analysis today is not all approaches are created equal. Lots of folks are using the term “attribution analysis,” but not everyone is talking about the same thing. It’s important to understand the different flavors of attribution currently available and determine which ones provide the greatest value for advertisers.
1. Last-Click Attribution: The most common attribution model is called “last-click attribution,” and its name pretty much describes the approach. Last-click attribution assigns tremendous value – 100% of the credit for success – to the click (or event) that brought the visitor to the marketer’s web site.
The problem with this approach is it ignores all other interactions and campaigns, or at least assigns them zero value. This clearly incorrect and misleading. It would be like giving credit for the Super Bowl win only to the receiver who made the last touchdown of the game, ignoring all other members of the team and all the plays that occurred before and after.
2. Even Attribution: As an acknowledgement that crediting the last click with 100% of the conversion credit is inherently flawed, early attribution practitioners devised a way to share the credit with other parts of their campaigns. “Even attribution” would apply equal credit to every event that occurred in a visitor’s conversion path. So, if the new customer were exposed to three display ads and then searched and clicked (for a total of four different events), each event would get 25% of the credit for the conversion.
In our previous Super Bowl example, it would be like dividing credit equally amongst all players who were on the field on the final play. This is simple but only a slight improvement over last click. And entirely inaccurate.
3. Predetermined Attribution Models: If the only thing wrong with “even attribution” were that credit was divvied up unfairly, then simply devising a fairer model should solve the problem. Predetermined attribution models attempt to do so by assigning a fixed amount of credit to each event in the conversion path, but those percentages are not equal.
For example, a predetermined attribution model might assign 50% credit to the last click/event, 10% to every e-mail interaction, and divide the remainder amongst any display ads the visitor was exposed to. This is like saying that on every play the quarterback gets 50% of the credit, the receiver 10%, and everyone else shares the remainder.
This would work great if every play was a pass, and every pass was completed. Unfortunately, that’s not realistic. Different plays would likely see the shares for credit divvied up in different ways each play. The same would be true for advertising campaigns.
4. Full-Funnel Attribution: An attempt to be more scientific than the previously discussed models, full-funnel attribution purports to measure all activities that contributed to the purchase, or conversion event, throughout the purchase funnel. By taking into account more marketing interactions beside the last click, such as display, search and social campaigns, full-funnel attribution provides the appearance of a more comprehensive attribution model.
However, one of the problems is full-funnel attribution only looks at a sample of the total interactions. This results in the credit being incorrectly weighted, leading to poor decisions and suboptimal campaign adjustments. For example, most full funnel attribution models look only at the paths of converted visitors, ignoring the other 99.9% of marketing interactions that didn’t result in a visit or conversion.
To see how this would impact measurement attempts, take a look at this example of full-funnel attribution in action. If almost every visitor who converted was exposed to Campaign A at one point, a full funnel attribution model might reasonably conclude that Campaign A was a huge success. The model would assign Campaign A large amounts of credit.
But this entirely misses the most important factor that attribution is trying to address: causality. If most of the unconverted visitors were also exposed to Campaign A, but failed to convert, then it’s impossible to assign any credit at all to Campaign A. A marketer using the full-funnel model would incorrectly allocate more budget to Campaign A, when in fact it had no bearing whatsoever on conversions.
5. Fractional Attribution: By far the most accurate attribution model is called “fractional attribution” or “dynamic fractional attribution,” because the model is calculated dynamically for every customer or campaign. As its name implies, fractional attribution measures the influence each part of a campaign had in contributing to the action (e.g. web site visit, sale, etc.) much like full-funnel attribution does.
But a true fractional attribution model doesn’t ignore any of the available data. Each and every display impression, click, social interaction, and e-mail campaign is incorporated into the fractional attribution model. By considering 100% of all campaign data, true patterns and trends can be identified.
For example, a wireless phone provider might run a mix of display ads, search marketing campaigns, and social media strategies as part of a push to promote a new mobile phone. Fractional attribution analysis can tell the wireless provider how much influence each and every one of those marketing events had on the outcome of the campaign by measuring 100% of the campaign events, not just the converted sample.
Some of the buyers may have been exposed to the display campaign, but by analyzing the complete set of data it may be determined that the display campaign had very little impact on purchase behavior. Wouldn’t a marketer want to know this in order to make campaign adjustments?
Attribution is clearly emerging as one of the most powerful and effective measurement tools for interactive marketers. In an advertising world that’s growing in reach and complexity, using the right kind of attribution measurement techniques will ensure that campaigns succeed no matter what new kind of advertising venue or marketing offering becomes available.