One Fish, Two Fish: Attribution Modeling

Dr. Seuss once said that sometimes the questions are complicated and the answers are simple. Of course Dr. Seuss never managed an attribution modeling effort, where both the questions and the answers can weave a complex tapestry of marketing investment performance.  But in many ways the answer IS simple: Do you want to develop the most accurate model of media influence on business profits? Of course you do.

I've got retail and direct response on my mind this week, so join me as we delve into a topic that is exponentially more important than the amount of attention it receives within most organizations. Somehow attribution reporting is still the exception and not the rule, even among fairly large retailers and direct marketers who pride themselves on the mastery of measurement and the understanding of customer behavior. Yet the path of least resistance prevails for many marketers. After all, attribution modeling requires a significant investment of time, energy and budget, in addition to approvals from multiple stake holders within the company, sometimes including C-level executives. Changing a fundamental marketing view of the business takes a serious organizational commitment. Attribution modeling is not an endeavor to test and figure out later, but rather an ongoing challenge to improve the analytical prowess necessary to truly maximize cross channel digital media investments.



One would imagine marketers flocking to the opportunity to improve the accuracy and breadth of reporting to account for the collaborative influence and attribution of multiple marketing touch-points, not to mention the elimination of duplicative tracking across channels.

"Duplicate tracking, you say?" A significant amount of revenue is being double-counted between your affiliate marketing channel and all other core programs: search, comparison shopping, display, and even to an extent your CRM programs. Since affiliate tracking platforms are almost always separate from the suite of products used to track other DR channels, this different cookie poses a risk of double counting revenue that is also being reported by your bid management system and/or ad server.

I'd be remiss to discuss the topic of attribution without putting it in the context of scale. Attribution reporting comes in a variety of scale-flavors, from small and crude MacGuyver-esque attribution reporting, to straightforward attribution reporting from your ad server, to full-blown, comprehensive, black-box modeling of attribution based on specific campaign attributes such as recency, context, and ad formats, all pulled in through APIs from your disparate marketing tracking systems. While leading providers ClearSaleing and Visual IQ are dedicated to the science of attribution, the web analytics companies, notably Coremetrics and Omniture, are eager to become players in the space and have been improving their offerings as well.

With scale comes great responsibility, as in the responsibility to manage your company's media investments wisely. By employing the most sophisticated approach to closing the loop across digital channels, attribution modeling becomes a dynamic predictive mechanism that drives fluid budget allocation based on the combination of digital media that works over specific periods of time. While scale is all relative, marketers spending a minimum of seven figures owe it to themselves to put an attribution reporting tool in place.  That's not to excuse smaller marketers from disregarding the practice, but with with greater scale comes greater value and insights.

For larger marketers there is one more tool that can help make your attribution models most actionable -- tag management.  By implementing a tag container with management tools, you are able to establish rules for when and how your various tracking pixels fire and track activity, thus adding an additional dimension to the actionable results of your models. For example, most retailers pay coupon sites a significant amount of commission as part of their affiliate programs. It's a necessary evil, but one that rewards these affiliates for perpetuating a consumer behavior that often occurs after a purchase decision has been made. If you want to preserve margins and eliminate the double tracking of consumers who are driven to purchase from SEM or other channels, and then go back to search for "your brand +coupons" right before checking out, some tag management systems can help you implement such rules. The affiliate pixel would not fire in this instance and the affiliate would not get the credit, or commission, for the transaction.

Dr. Seuss would be taken aback by the complexities of digital attribution questions, not to mention the evolving black box algorithms in place to try and model the answers. What campaign attributes drive, contribute to, or detract from performance? How does the model account for non-paid components that are inconsistent and not within your control? Where does lifetime value fit into the equation? How do attribution modeling and traditional mix modeling work together to inform overarching media investment planning? What assumptions are we making that can lead to inaccurate modeling?

Indeed, the questions can be complicated. However, a significant investment of time in planning and asking the difficult questions is the first step to developing a successful attribution model. The practice is still nascent and nobody has all the answers yet.

Are you successfully modeling attribution? Considering it? Have you had difficulties selling it in or implementing a system? I'd love to hear about the good, the bad and the ugly of your experiences.

5 comments about "One Fish, Two Fish: Attribution Modeling".
Check to receive email when comments are posted.
  1. Christopher Brinkworth from Ensighten inc (acquired TagMan), March 22, 2011 at 2:51 p.m.

    Jason, great stuff. Im hoping I get to touch on this on the Panel at OMMA Metrics tomorrow.

    A (decent) Tag Management System and Attribution go hand in hand for several reasons.

    This video explains it easily:

    1) Tag Management Systems can connect all disparate data points/tags and unify with one global ID. Thus, showing touch points easily in one report.
    2)They can help you harvest referrer data at a deeper level without extra complex tagging (SEO, Stem level, BookMarked Direct to Site etc) and unify against other vendor touch points (per point 1) above.
    3)Yes - they can decide which tag to 'fire/not fire' - but as in TagMan's case, decide on a complex model 'how much to award' directly as a percentage into a tag.

    EG: Flat attribution model = 4 channels, 25% of credit goes directly reported into affiliate tag on flat model basis.

    One note however, is to consider other points when looking at this activity before setting these types of rules. Are you limited in what you can see currently? EG - do you have Live Person/Bold Chat as a live chat tag option on your site. Is that assisting conversions and therefore as there is a cost to it - should be rewarded as part of the model etc. What about environmental factors such as competitive activity? Is your natural search racing ahead at the moment because of some competitive TV activity you are not aware of raising awareness and forcing you onto the radar as competitive research etc.

    It's an exciting year for measurement tools when it comes to attribution/tag management and I suggest that readers read up on 'what' people are offering and what is actually live.

    Being the leading (first) Tag Management System and (only one) with real-time attribution since 2007 - there are plenty of bias but also industry focused WhitePapers/Case Studies and videos on our Site ( However, also look into what the NRF's are doing to research this growth area of measurement:

    Also - get a copy of Joe Stanhope from Forrester's (agnostic) paper on Tag Management

    Any questions - happy to answer them also.

    Chris Brinkworth
    TagMan Inc

  2. George Michie from Rimm-Kaufman Group, March 22, 2011 at 3:44 p.m.

    Jason, this is a fine piece.

    At RKG ( we've been studying this problem for years, and over the last two years have built what we believe to be the most sophisticated and flexible system on the market for smartly doing fractional credit attribution.

    We totally agree that this is one piece of the even more complex media mix discussion, but it is an important piece. A key to doing this well is to have the flexibility to create different classifications within a marketing channel. Most pressing: separating brand traffic from competitive non-brand traffic in both paid and natural search; but there may be different classes of affiliates that should be modeled differently as well.

    Chris Brinkworth, above, and I have been on the study group since its inception. Happy to talk further if of interest.

  3. Mark Hughes from C3 Metrics, March 22, 2011 at 4:36 p.m.

    Said another way, "How would you like to run your business using 6% of your data?" If the avg number of touchpoints = 15, that's exactly what's happening. 1/15 = 6%. In an era where CMO's are under fire for marketing accountability...this is an easy place to start.

  4. Gabe Elliott from Vision Media Interactive, March 23, 2011 at 2:27 p.m.

    I understand that there are many companies still trying to figure this issue of attribution and valuation out and to many, its even a new concept. Although not as well known yet, Channel Intelligence has been providing an industry leading solution for this for some time.

  5. Allyson Taylor, March 23, 2011 at 11:33 p.m.

    Making the decision to use multi-attribution is a huge step in the right direction. But the very nature of multi-attribution makes it more vulnerable to an already major problem in online conversion attribution - accurately tracking visitors. Cookie blocking/deletion is rampant - 33% of users clear them at least every 30 days. So any attribution models that rely on crediting marketing touchpoints that happened long before the last click (in many cases longer before conversion than the lifespan of a typical cookie) are going to be at the mercy of the robustness of their tracking systems. Multi-attribution simply requires visitor tracking mechanisms to be highly effective, lest important top-funnel sources lose credit they deserve. At Convertro, we're pioneering new methods of visitor tracking that go beyond standard cookies so that we can be sure we're building attribution models that have reliable data at their core.

Next story loading loading..