Commentary

Improving Marketing ROI: Towards a More Equitable Conversion Attribution Model

While last-click attribution is an easy default mode for digital marketers, this practice can lead to serious marketing missteps, including inflated digital marketing estimates and misallocation of marketing spend.

The underlying issue is that a frequent assumption of online marketing -- that the last click is the prime contributor to the sale -- is flawed. In fact, there are a multiplicity of factors (touchpoints and exposures both on and offline) that have helped pave the way for that click. Unfortunately, last-click attribution models create an illusion of "marketing science," when in fact the results are often grossly overstated, resulting in erroneous findings that can dramatically affect marketing ROI.

Marketers instead need to consider a more sophisticated analytical approach to tackle the issue. The best approach utilizes a staged system of multivariate equations to determine the relative contributions of different elements across the marketing mix. Through this advanced analytic approach, one can quantify the true effect of investments in upstream media vehicles such as TV, print, display and e-mail in driving consumers to search or to marketers' websites and subsequent conversion. Ultimately, the insights derived from this technique lead to smarter, more effective spend allocation decisions for both online and offline sales success.

The staged approach: a more equitable attribution model
Most digital attribution models, regardless of whether predicated on last-click assumptions, perform analyses that assume the effect of digital media (and search in particular) in complete isolation from other factors. Generally speaking, they don't consider the impact of macroeconomic and seasonal factors or interactions with offline media like TV, radio and print ads. Unless marketers create a more holistic model that weighs all the other factors that can also impact digital results, they're not getting an accurate assessment of their online performance.

A better way is to take an approach that analyzes online effects and digital elements side by side with all the other factors that drive brand sales. Additionally, you need to introduce a staged approach to quantify the interactions between online and offline marketing vehicles. This piece is critical in order to get to a true assessment, and to understand the extent to which offline vehicles are driving visitors online.

Here's a scenario to illustrate how this works. Say someone sees a Hotels.com special offer in a TV ad, then three weeks later wants to book a hotel room for a trip to Santa Fe. They go to Google and type in "hotels in Santa Fe" and up pops a Hotels.com paid search ad. It offers $50 off a room at a Holiday Inn in their destination city if they book now, so they click and purchase. Most digital attribution models would give full credit to that last click, when in fact the TV ad helped create awareness and contributed to the sale.

A staged approach would instead look at how TV, search and all other marketing drivers correlate to sales, with an assignment of how much sales were driven by each. This involves using a system of sequential equations. The first set of equations isolate and quantify the number of transactions driven by each media vehicle. This is then overlaid with a second stage of equations that quantify how offline marketing and display ads contributed to online website and social media visits/interactions as well as search and display clicks. This gives you a more complete picture of how online engagement was driven through other channels, and allows for proper reallocation of attribution. To get a true depiction of ROI, you need to understand not only who or how many converted through online media but also what drove them to the online channel.

Traditional attribution models skew conversion results
In some instances, we've found that 30 percent to 40 percent of search clicks were driven by upstream media vehicles. Using the standard attribution model, a marketer would have significantly overstated the ROI of search and understated the impact of display ads and offline media. With a staged approach, you reattribute a portion of sales from search back to factors that drove the searches. Typically, even after performing this reallocation, search still proves to be a very efficient tactic. The true benefit, however, is that you get a more equitable representation of the ROI of offline media and online display ads.

This staged analysis model gives both researchers and agency planners alike ammunition to go back to management and say, "While we need to continue to maximize our sales delivered through search, it can't be done at the expense of investment in upstream media vehicles. If we take our foot off the gas pedal for those tactics, we'll also see deterioration in search performance." Using a less sophisticated approach, faulty conclusions might have resulted in poor investment decisions, undermining both search and offline media performance.

Recommend (6) Print RSS
6 comments about "Improving Marketing ROI: Towards a More Equitable Conversion Attribution Model".
  1. John Ardis , March 24, 2010 at 10:44 a.m.

    Dan, how do you go about gauging the offline influence. For instance, the example you give on the Hotels.com TV ad - is it through pre/post surveys that you determine the offline influence, or some other method?

  2. Paula Lynn from Who Else Unlimited , March 24, 2010 at 11:37 a.m.

    Unless I am tracked for a very long time on everything I see and do, you will never know, nor are you entitled to know anything about me.

  3. Anto Chittilappilly , March 25, 2010 at 4:35 p.m.

    Great article, Dan. There are few points you make in your article that we strongly agree with and try to drive home in the conversations we have with people like you in the industry, as well as our clients: Last-click attribution leads to serious marketing missteps and misallocation of marketing spend, and you can’t find the attribution effects of certain media in isolation from other factors.

    While the multivariate equations you discuss can help you to create a model, you may need a more comprehensive approach to ensure accuracy of your attribution. The model must use the most granular data available. Online channels are very rich with granular data up to the impression log level for each individual user. There is a great value in getting the touchpoint stack of individual users. It is also important to use the data of converters and non converters in the model to understand the difference between them. Finally, understanding the importance of each attribute is very critical. Some attributes (includes macro economical attribute) are important to some marketers and but not for others.

    Besides the most optimal media mix, a good attribution system should be providing several tactical deliverables like the most optimal offer sequence or creative rotation, optimal frequency cap, and insight into how to synchronize the campaigns across different channels to leverage the affinities between them. For example a particular display creative may have higher affinity to provide a greater than average lift to particular search keyword, or trafficking the creative and keyword with the correct time-lag could produce greater ROI to both campaigns.

    Wish you all the best

  4. Daniel Eggleston from Synovate MMA , March 26, 2010 at 9:05 a.m.

    Hi John,

    The primary method we employ at Synovate MMA is econometric modeling of aggregate sales data. There's no need to link together respondent or cookie-level data from online and offline sources when using aggregate sales data for analysis. The econometric models relate variation in both traditional and online media execution across time to variation in sales trends and establishes the unique impact of each marketing vehicle. Hope this clarifies.

    Dan

  5. Christopher Brinkworth from Ensighten inc (acquired TagMan) , March 30, 2010 at 4:26 p.m.

    An excellent whitepaper from Havas was just released on this. Link is below.

    My view on all this, is still that this is all 'retrospective'; as in - "This is what we 'should have' done". TagMan works with these results from a digital perspective to 'action' that data in real-time through 'dynamic awarding' -or, (sorry, another new term) "Applied Attribution"

    TagMan video: www.tagman.com
    Havas White Paper: http://bit.ly/90ES7P

    http://bit.ly/90ES7P

  6. John Grono from GAP Research , April 15, 2010 at 11:27 p.m.

    Excellent article Dan. My only question is, does this mean that some marketers and online sales people are still using last-click? How antiquated.

    Also, Anto, you do NOT need to know how each individual - be they an existing customer or a prospect, a purchaser or a non-purchaser behaves. Econometric models at the aggregate levels tell you that (for example), when you did a burst of TV over 3 weeks, with search along with POS billboards that sales went up x%. For every marketer I have ever met that is more than sufficient. They are not judged on how many individuals they individually impact and the size and scope of that impact. They are judged whether their marketing activities sell product (or generate call centre traffic, foot-fall etc).