Confused about how to attribute credit for sales and brand-building success? You’re not alone. It’s a hot topic given the level of media spending in digital formats and continued media fragmentation.
But there’s a way to do it called media mix modeling. And it should be at the head of every marketer’s priority list.
Let’s start with a pair of baseline examples, advertisers with retail, call center, and online commerce touch points with their customers.
Vonage and XM Radio fit the bill. Their activities sound complex, but there is something to be learned from the way they sell their products and interact with their customers over time.
Both firms sell through retail at establishments like Target, Staples, Best Buy, Comp USA and Circuit City. You can buy at these stores and activate your account either by phone or online.
In addition, both can take your order right online, and will ship your device. You can activate your account either by phone or online in this scenario as well.
Both firms pursue a mix of online and offline marketing initiatives.
Offline Media Mix
For offline, they will have mixtures of national and local media initiatives with a combination of unique URLs (www.vonage.com/promohere), literal URLs (www.xmradio.com) or 800 numbers on their TV, print, radio, out-of-home and direct mail.
There will media planning data for the national buys that will have the number of 15 and 30 second spots by station, by day, and the GRPs attached to it for broadcast. For print, out of home, and direct mail they will have the ad weeks, days that drops happened, and the period of time ads are in play in different markets.
Online Media Mix
Online is where most people struggle in the process. There will be a mixture of performance buys, branding ads, affiliate buys, PPC search, PI Search, and Organic Search.
Marketing departments may employ different tactics to try and capture as much spend to ROI data as possible through unique URLs and literal URLs depending on channel. Most will have some sort of ad server HREF or unique URL in play. If they are trying to track as much as possible the only channel with a literal URL will be organic search.
Every agency and marketer has their point of view on how to attribute “credit” for driving a customer or building a brand. That is fine. Those rules can and should be incorporated into your models, and fine tuned over time.
Another small hurdle at this point for most people in this process is the cooperation of IT departments. As with these two examples, most companies have one or more ad agencies in play.
Most clients do not “partner” with their agency to the point of providing them enough detailed data, and periodic access to it refine strategies quickly. IT inevitably gets involved, and that is when compromises usually begin to happen.
We will use the assumption that the agency is perceived as a partner, and IT gives up the goods to get the job done.
The “Choke Point
The choke is encountered when the agency has to get the online and offline “house” in order for data integration. Let’s say the advertiser is using a few what refer to as delivery technologies like Atlas as an ad server, DART Mail, a little Point Roll, and Commission Junction for affiliate. Web site analytics companies are not critical to the solution, but can provide some customer behavior data on the site to help with clustering and segmentation.
I count four cookie spaces outside of Web site analytics. In order to get down to business and make sense of it, you must consolidate these at the impression/view level as well as click. Click is easy. Impression is where it all comes together.
I foresee millions of rows of log files that contain rich data for each channel. Stop, I know what you are thinking. You can’t do this in Excel.
Okay. To get past the first hurdle in consolidation you must synchronize/ match code all of these cookie spaces. This is a combination of passing cookie IDs/User IDs/Account IDs/Global User IDs between the delivery technologies and the clients CRM/eCommerce/Transaction Engine.
Once you do that, do not summarize the data. You will need to maintain an automated process to constantly update and append this match code across all existing and new IDs. This helps overcome cookie deletion, and multiple browsers or computers.
Your table in the database will probably be 50 to 100 columns across if you use one table, or you may have a series of smaller relational or star schemas.
Clustering and Modeling
Once you have established your match code process and it is humming, you now have the ability to look at a customer acquired or a cluster of customers acquired—say cluster A, B, C, and D. “A” being the best and “D” the worst.
Through regression and modeling exercises you will be able to start your media mix modeling.
I started in newsprint 15 years ago. Last time I checked, success did not depend on having the ¼ page ad in one issue in the top left or right corner of the page, but on having that ¼ page in the same spot for 18 or 52 weeks. That was said to create the “branding experience.” We all agree that frequency is important.
Here is where online advertisers and their agencies have been let down by the delivery technologies. These technologies and click tracking look at the “last click or view” that drove an activity on the client’s Web site. This is the exact opposite train of thought leveraged in the rest of marketing.
Some delivery technologies have R&F reporting, but they only have it for what they serve and track so it is a “stove pipe” approach, and ultimately is not ideal.
Some Fruits of the Effort
If you have match coded correctly, have not summarized your data, and have maintained those match codes you have the holy grail (or wholly grail, depending on how you look at it).
With this you can look at that cluster of customers and examine what were all the impression/view and click activity that drove them to the Web site through branding and direct efforts, and which creative, placement, key word, list, site, etc… they had in common. So for customer segment “A” you optimize your creative and media mix to drive more of segment “A”, and you leverage it to optimize out commonalities in segment “D” to not drive more of them.
Once the online house is in order, you can then fold in the offline media spend and planning data. One of the first opportunities when you integrate these data sources is identifying patterns and correlations in offline marketing activity on the media plan in combination with the impression/view through activity, click driven activity, and activity on a your Web site whereby the customer or prospect does not appear to have been exposed to online marketing efforts yet landed on your site.
You can then overlay the above offline and online data points with the retail sales, and call center data that is being captured.
With this combination of media and sale the user has purchased the XM Radio or Vonage device at Best Buy and has activated it over the phone or online. That can be brought in as well to the modeling effort.
These two companies have an added bonus of an administrative panel or online account. The moment a user logins in to use the Vonage admin panel to set their preferences or to listen to XM Radio on their laptop(s), you will be able to link that used ID with the match code.
You can then identify a cluster of customers or just one that was exposed to a series of ads over a time period online, by offline activities in their DMA or geography—i.e., that they bought at retail, and used online. Bingo, you have a direct tie purely from online branding efforts and conversion.
Now you can’t just look at one customer and make these sweeping statements. You will need to consult with your agency and establish what is the number of customers, sales, etc… that all are comfortable with saying are statistically valid.
Both the online marketing and offline marketing team will need to come to the table on this. That is what will ultimately differentiate one advertiser from another, and one ad agency from another. It is how you look at the data and make those calls.
But there are a few caveats. You can’t do this with delivery technologies that don’t make 100% of the raw log data available.
Another point to remember is that some delivery technologies provide richer data sets than others. This can happen with geographic data.
Having access to all the raw data every day from delivery technologies has been priced everywhere from free to ridiculous amounts. I tend to go with the free as more data for less is always better.
Martin Wesley is CEO of BlackFoot Inc.