Brands reach their consumers via an ever-increasing number of channels. The ability to accurately parse out the effects on buying behavior of the various touch points, and understand each channel’s impact on an individual level, is becoming critical.
If marketers don’t attribute sales and leads to the appropriate channels, their understanding of various expenditures’ contributions will be skewed and will result in marketing decisions that don’t produce desired outcomes. Nevertheless, many marketers still rely on a "last touch" approach for assigning credit to their marketing activities.
Fractional or partial attribution allocation methods are a step in the right direction. But these tend to be judgment based and produce questionable accuracy. Furthermore, most marketers are overwhelmed by how to get started.
Fortunately, there is a new approach, or framework, that addresses multichannel attribution. This framework leverages promotional data in combination with modeling techniques and universal control groups. Using this approach allows marketers to better discern, at an individual level, each channel’s influence on demand generation.
Components of this alternative framework include:
1. A multi-channel marketing test and control design establishing baselines for contribution by channel. For example, direct mail might produce three times the demand of email; email might produce two times the demand of banner ads and so on.
Marketers determine these by holding out channels to various control groups to measure the incremental impact on sales. The metrics subsequently form the basis for channel weights in the attribution algorithm.
2. Construction of channel-based incremental response models, which are used to calibrate channel weights at an individual level. This helps better gauge channel contribution based on each customer or prospect’s specific channel propensities.
Consider Jane Doe, a sample customer who has been determined by her past behavior to be highly responsive to email – say 2.2 times as likely as an average customer to respond to this channel. If the average customer’s baseline for email response is 100, Jane Doe’s uplift model index for email would be 220 (2.2 x 100).
Now, suppose a marketer has determined email has a channel weight of 0.65 – that is, email is only about two thirds as productive as all the direct channels in the marketing program. Jane’s individualized email channel weight is 1.43 (0.65 X 2.2).
In a similar fashion, the marketer can calculate individualized direct mail channel weights. For Jane Doe, the individualized direct mail channel weight is 4.29, or three times that of email. Therefore, when Jane Doe makes a purchase, direct mail gets three times as much credit for the purchase as email. For every $100 Jane Doe spends, $75 would be attributed to direct mail, and $25 would be attributed to email, assuming Jane Doe was exposed to both direct and email efforts.
3. The development of response curves for each direct marketing channel, such as direct mail, email and banner advertisements. Response curves allow marketers to adjust the weight given to channels based on the last time a customer or prospect was touched by that channel, relative to the purchase date. The closer the marketing touch to the purchase, the greater the contribution.
Response curves can also be used to determine the appropriate attribution windows to use in the attribution algorithm. The data may show, for example, that email contributes to purchase behavior for up to two weeks, while direct mail contributes up to a month.
4. Universal control group design and implementation, which is used to account for the effects of mass media and other demand generators. This ensures that direct marketing channels are not over-credited in the attribution algorithm.
While this attribution framework requires some investment and patience, marketers adopting such a data-driven approach will make wiser decisions and uncover greater return on their marketing dollars.
John Young is senior vice president of analytics for Epsilon
Network
