Attribution Measurement, What’s Old is New Again

Attribution Measurement, What’s Old is New Again

Too often, marketers carry out activities and then wonder whether they were worth it. Channel specific metrics are great at measuring tactics but fall short in shedding light on the entire picture. Agency leaders, for years, we have been preaching the need for cross channel measurement, at an individual level, and finally the brands are on board and ready to invest, but is it too late? Is cross-channel measurement dead? Certainly Multi-Touch Attribution (MTA) as we know it today will be fragmented at best. MTA is a form of measurement that relies on user-level tracking that evaluates a user’s exposure and reaction to advertising across different channels and/or ad formats. How can MTA survive if the foundation by which the users are measured will become obsolete in the next 12 months. The cookie, device ID, App ID and most virtual consumer identifiers will not be shared across domains - this is a fact and happening. 

Numerous studies have proven that combined advertising channels and touchpoints contribute to increased brand awareness, drive user engagement and ultimately result in higher sales. However, without cross channel measurement, how can we be assured that advertising across channels is not redundant or wasteful?

Luckily one form of cross-channel measurement, Media Mix Attribution (M3) does not rely on cookies, device IDs,.., rather M3 leverages aggregate Ad delivery and media spend data over time. M3 has been around for many years. In fact Consumer Packaged Goods and Retail companies have used it, for decades now, to evaluate and attribute all forms of advertising, promotions and marketing to in-store sales. Recently M3 has evolved to include certain user-level data as inputs (meaning, beyond the next 12 months there will still be access to user-level data within some “walled gardens” it just won’t be mapped to any other platform or channel).

Maybe we can just stop spending endless resources, time and orchestrating a multitude of panel discussions to recreate a new version of the cookie and/or a device ID and practically accept the fact the while consumers want relevant content, free news and other forms of entertainment offered through an Ad-supported model - consumers do not want their every action and/or move tracked. Rather, we can work to further advance the M3 approach to make it even more robust, near real-time and more widely accepted as a privacy-safe and highly effective solution to measure and attribute value to advertising across channels. This approach, the walled-gardens will not be able to take away from the industry and this could become the new gold standard from cross-channel measurement.

So how do you get started? Given all the data that is available for measurement, it is often difficult and time consuming to compile a powerful set of metrics to track marketing performance. That is why we focus on creating the metrics that matter - metrics that align the way we measure success specific to a brand’s ultimate business objective(s). There isn’t a one-size fits all model, every brand operates differently and oftentimes has different priorities for what they seek to understand about their marketing. Luckily the M3 methodology can accommodate bespoke brand initiatives and deliver a custom solution.

Case Study

Take a Canadian national insurance brand for example. The marketing team wanted to demonstrate the importance of upper funnel branded media in driving lower-funnel performance, such as: search discovery, website activity and ultimately quotes. Digital marketing is now unavoidable, especially in the insurance industry. While it is natural to shift dollars in paid digital media to increase digital leads, it would be incorrect to treat this channel as a silo. Our hypothesis was that branded / offline media not only increase online performance but also additional online leads (see example below). A simple week-on-week analysis showed that branded TV advertising had a significant impact on online quotes.

While we saw an impact on digital media results, we knew that most quotes were generated by the website and search results. Furthermore, we were after results that considered the entire media mix over the course of several periods. There were three key objectives (Exhibit A):

  1. Identify offline/online media drivers of the consumer response behaviors, such as: website visits, calls and quotes.  What are the media channels (offline/online) which drive consumer response?
  2. Determine whether the impactful media drivers vary by region. Does the impact of the media channels (online/offline) vary by region: Alberta, Ontario and Quebec?
  3. Does the time period of the campaign have an impact on the efficacy of the media mix. Does the impact of the media channels (online/offline) vary by time period: Jan-Jun Year 1 to Jan-Jun Year 2?

Exhibit A:

We had over 12 months of marketing channel data, including call center, website, online media and paid search to determine whether there was a direct and/or indirect impact “halo” paid offline/online media. We isolated paid media impact by region to control for biases and regional regulatory variances. Reliance on user-level data to develop a multi-touch data set, using cookies, mobile IDs, etc. was just not possible. Getting access to log-level data at the cookie level was and is no longer an option, not to mention that we had many additional non-cookie tracker channels in the mix (Exhibit B):

Exhibit B

The Media Mix Attribution (M3) model was able to successfully interpret the data and deliver meaningful insights to informed future investment and campaign ideas:

  • Media co-dependence: Due to co-dependence among media and website behaviors, the models are likely to select variables which represent multiple attributes (e.g. site traffic was highly correlated with unique site visits and will represent both, etc.)
  • External/Other Factors: Media Mix Modeling looked at the direct impact on quotes by considering the effect paid media has on quotes irrespective of other driving factors, such as: website behaviors, promotions, other marketing initiatives
  • Negative Relationships: A negative relationship between media and quotes did not indicate that spending with that media causes a reduction in quotes; it did mean that relative to other working media this channel is weaker
  • Multiple Analysis to draw our Conclusions: While univariate correlation analysis can identify the individual impact of media and website behaviors to quotes, it was important to evaluate the entire mix and control for additional factors and variables. Leverage both types of analyses (univariate and multivariate) for optimization

In the end, we modeled the entire online quote funnel stimulated by upper funnel branded media that supercharged the flow of users throughout the funnel (Exhibit C):

Exhibit C

Conclusions of our analysis:

The inability to leverage user-level log data did not impact the integrity or accuracy of our findings. The model incorporated all media channels to produce a comprehensive media mix analysis.

  • Offline GRPs and OOH were strong contributors to generating Desktop Search demand. We know that search discovery leads to site activity and almost half of Mobile site and a quarter of Website visits were directly attributed in starting an online quote.
  • Branded TV and OOH not only contributed to the more immediate action of search but provided a measure halo of 6 weeks, meaning we were able to further attribute digital performance stemming from upper funnel branded advertising.
  • There was a clear link between brand advertising and consumer search behavior that ultimately generated business results (Exhibit D):

Exhibit D

MICHAEL KAUSHANSKY'S BIO

Michael Kaushansky serves as President Helia for Havas & Chief Data Officer, North America. In this role he has direct oversight and responsibility for the agency’s direct marketing, data and technology and Michael is responsible for managing and advancing Helia – Havas’s Direct Marketing / Loyalty / CRM agency focused on managing, extracting insights, and activating consumer data on behalf of our clients through direct channels. Michael sets the agency’s overall approach to connecting brands with consumers through better data and content. Michael helps define our data capabilities that are a bedrock of Havas overall delivery to clients and provides insight into the continued evaluation of our suite of marketing effectiveness products.

Michael has been involved in the field of data, analytics and direct marketing for nearly two decades with a strong focus on consumer insights, media effectiveness, marketing modeling and digital analysis. Highlights include the development of media mix models, implementation of digital attribution, deployment for full-funnel optimization processes and specialized visualization dashboard to enable immediate and meaningful sharing of results, clients include: AutoZone, Choice Hotels, Liberty Mutual, NFL, Sears/Kmart and Dish Network.

Michael has held roles of increasing responsibility at Publicis, GE, Target and GSK. Most recently, Michael led all marketing analytics at Ogilvy where his work spanned several leading global brand advertisers including UPS, IBM, Nestle, Cisco and SAP. 

Michael holds a Bachelor’s degree in Mathematics from Creighton University and a Master’s of Science degree in Applied Mathematics & Operations Research from Creighton University. Michael is an advisor to Rutgers University’s Big Data graduate program. Michael makes his home in New York City with his family and two daughters.

Related Stories

No stories found.
CDO Magazine
www.cdomagazine.tech