For marketers, one of the great promises of the Internet is that it is a highly measurable medium. They can track customer clickstreams and identify which campaigns generate clicks and conversions. This in turn can help them increase RoI from marketing campaigns. All this is true in theory. In practice, measurability of the digital medium has been a double edged sword.
Consumers are typically exposed to advertisers across a number of online channels. For example, a consumer may be exposed to an advertiser\u2019s display ads at multiple websites followed by sponsored search ads for a number of search queries. So a conversion may be the result of a series of ad exposures. This raises the key question of attribution: which ads get credit for a conversion and how much credit do each of these ads get? The answer to this question is critically important because a number of key decisions including ad budget allocation across channels as well as bids for ads depend on sales and/or brand value generated by each of these channels. 1 Although the issue is well documented, current solutions are often simplistic. For example, a common practice is to attribute the conversion to the ad exposure that generated the user\u2019s last ad click prior to conversion. Another practice is to attribute the conversion to the ad exposure that generated the first click from the user. In other instances, firms apply exponentially decreasing weights based on time of ad exposure. These methods of attribution penalize prior exposures and give greater credit to ad exposures that occur just before the sale. As all the ads collectively influence the consumer\u2019s decision to make a purchase, it is difficult to disentangle the contribution of various ads towards the eventual sale. But if it is done incorrectly, it defeats the whole notion of data-driven ad optimization. And in practice, most firms often do it incorrectly. Thus, although online advertising is measurable, the gains from its measurability are ultimately somewhat limited.\n\n
There are three main approaches I advocate when thinking about how to deal with the challenge of attribution:
- Reassess the need to do attribute conversions: Brand advertisers should care more about clicks than conversions. That is, they should ask which campaigns are generating clicks and visits to the website rather than which ones are generating conversions and purchases. Thus, for brand advertisers, it may be better not to do any attribution rather than do it wrong. Further, keep in mind that the goal of evaluating conversion rate is to evaluate the quality of clicks. One can compute this using alternative metrics that are less susceptible to the kinds of biases associated with conversions. One such metric is average time on site. Suppose the user clicked on ten different ads and eventually purchased only once, the average time on site is available for all ten ad clicks, thereby allowing us to meaningfully compare the quality of all the clicks.
- Test robustness to multiple attribution rules: When using simple attribution approaches (e.g. attribute conversion to last click), it is important to compare insights by using multiple attribution approaches. For example, do the rules attribute to first click, attribute to last click and attribute evenly to all clicks all shed similar insights about performance of various ad campaigns. If so, attribution is less of an issue.
- Model Latent Awareness: When budget allocation and bidding policies are driven by data on conversions AND different attribution rules are generating different insights, it calls for a more sophisticated approach. One such approach is to model latent awareness and evaluate how ad impressions and clicks affect consumer\u2019s brand awareness and how brand awareness in turn impacts consumer search activity as well as conversion propensity. These techniques can be complicated but may be necessary if the above two principles are not adequate. Interested readers can read more on the technique in Section 06 of this paper
In short, while the digital medium is highly measurable and therefore capable of delivering precise quantitative insights to marketers, one needs to be cautious about incorrect measurements and inferences. Errors in attribution can undo any gains obtained from the granular individual-level data in digital channels. It is important to reassess the need to attribute conversions and to test the robustness of the insights to multiple attribution techniques.
How are you using data on conversions and purchases in digital channels? Has attribution been a factor you have considered?
1. For example, see a discussion on this topic by Analytics expert Avinash Kaushik at http://www.kaushik.net/avinash/2008/03/standard-metrics-revisited-5-conversion-roi-attribution.html. \u21a9