AFTER 30 YEARS AND THOUSANDS OF CAMPAIGNS, WE’VE LEARNED A LOT. TO PUT OUR LEARNING TO WORK, SIGN UPFOR OUR BLOGS AND NEVER MISS A POST!
Attribution Conundrum – Solved!
July 24th, 2018
Uh, not really. If anyone tries to tell you they have an omni-channel attribution model impeccably derived, you should run, run away. There is no Holy Grail for attribution.
At the recent DMA Marketing Analytics Conference in Atlanta, we participated in many conversations about attribution. The wide-ranging consensus? There is simply no ideal solution.
And with the increasing restrictions on collecting and using personally-identifiable information (PII), it will be even harder to find one. However, you must make the effort to define your channel attribution as best as you can, making reasonable adjustments for constraints on your model.
So let’s start defining channel attribution with some basic analyses, such as:
How many of your new online donors received a direct mail piece prior to their gift?
How many of your direct mail campaign responders received, opened, and clicked-through an email in the days leading up to receiving the mailing?
“There is no Holy Grail for attribution.”
These initial analyses are good for reference, but they may not tell the full story. Online contributions may have increased with your mailings, but was your social media investment held constant. Can you even track who saw your ads? Some campaigns are oriented more to engagement or cultivation, so how far back should you consider?
Nevertheless, baselines will ultimately help you determine if your model gives you a reasonable answer or not.
With that in mind, start building datasets containing:
All promotional and transactional activity for donating households for a given period prior to the donation
Probabilistic promotional values for channels like DRTV, social and print that cannot be tracked and stored (e.g., 30% chance the TV add was viewed, 70% chance the Facebook post was viewed, etc.)
Layer in the effectiveness of each promotion, either at the campaign or channel level. Ideally, the effectiveness would be a function of the investment level to mitigate a spend more/get more result
Each of these data elements requires decisions that impact the accuracy of your attribution model. Your best bet is to establish a few reasonable options for timeframe, probability, and effectiveness based on your initial analyses, run them all, and look for convergence. The ultimate goal is to model the proportional allocation of each channel and touch point, even if some of the data values in the dataset are more subjective than objective.
Having these types of datasets provide an excellent knowledge base for tackling your attribution questions, but if you still want to push farther into the attribution world, you can explore Cross-device ID (XDID) technology and other software options. These can be pricey, but they allow for robust online tracking of individuals and the possible matching of offline activity to online connected devices. These data can “fill in the gaps” for some of the values that would normally be estimated with probabilities, but can require a daunting investment.
No matter where you are in your quest for marketing attribution, first take incremental approaches appropriate for your organization. There will not be an Indiana Jones bringing home an attribution vessel with supernatural powers!
By Jeff Huberty|Executive Vice President of Analytics and Partner