Commentary
Does The Demise of The Cookie Resurrect Media-Mix Modeling?
- by Cory Treffiletti , Featured Contributor, November 10, 2021
The last 20 years or so have been spent making the case that digital media is great at building a brand and driving demand because of the inherent accountability it presents. Digital media has hung its hat on understanding audiences, tracking engagement -- and attributing value to both. Measurement, attribution, optimization have been the core tenets of the web. The future of these is somewhat in doubt for the open web, even if not for walled gardens.
I have always stated that attribution is like wizardry, but in fact you had to have an almost religious view of it. You believe in it, you have faith in it, and you choose one that guides the way you make decisions.
The challenge is that attribution was always driving a path toward cross-media attribution. With the demise of the cookie, I don’t think that will be possible. Even more important, digital media has become the cornerstone of all media, representing the highest spend and the highest volume of both impressions and engagement, and in the future it will be only slightly more reliable in terms of measurement than other forms of media like TV and outdoor. You will know how many impressions you generated and you will have a proxy for engagement, but across channels, impact will be very difficult to accurately and effectively measure.
advertisement
advertisement
The question I’m forced to ask: Is that actually bad?
Maybe we simply need to take a step backward for a moment when we look at digital media. Is it possible we can shift some of our attention to creating better use of the space itself rather than measuring the latent effects of the exposure? Maybe we should focus on how the message is delivered across channels and the story they tell, rather than trying to attribute value to one touch or the next. Maybe we should be looking at a higher-level of analysis that is based on a return-on-ad-spend and media mix rather than the channel-by-channel attribution models that we have been trying so desperately to shoehorn into the mix the last 10 years.
A great brand campaign drives more search. A great TV spot will drive website traffic. A great outdoor campaign can challenge perceptions and support other channels. The tactic-by-tactic measurement model tries to say that one touch is more valuable than the next, but maybe it’s the mix we should be focused on.
Media mix modeling feels like a lost art and one that should be resurrected in a cookie-deprived world. Examining the full spend, across all channels, and the resultant impact -- on top of funnel metrics such as overall engagement, traffic as well as downstream metrics like pipeline and leads -- feels better in a world where we no longer rely on a cookie. MAIDs and other “interstitial” metrics may still apply to some extent, but even those are fluid and suspect because of the moves of Apple and Google.
You can identify the trends over time, recognizing the latency effect, and then hypothesize the leading indicators where you can optimize a campaign. That optimization can happen whether you are focused on brand or demand metrics.
In my early career, we relied heavily on media mix modeling because digital metrics were not yet mature enough to trust. Most of the traditional marketers did not believe in them yet, so we always brought in media mix to the conversation to legitimize spend.
These days the tables have turned and the majority of spend is digital in nature, but we can still use MM to justify and rationalize the spend. In no situation does MM lean towards a silo’d effect, meaning it never says “go 100% digital.” But it does help a brand understand why the spend should be what it is, and it creates measurement that supports how dollars are allocated to meet business needs.
Are you invested in media mix modeling to help you understand how to best reach your target?
Good points, Cory. The problem is that many advertisers don't approach media mixing scientifically---nor do they see the need to. Instead, they make such decisions arbitrarily---without anything approaching a serious evaluation of alternatives. As an example, take advertisers who buy time in TV news,sports and/or specials---all at premium CPMs ---and, often, not necessarily reaching high concentrations of their prime marketing prospects.These account for about 45% of linear TV ad spend and their "sponsors" are wedded to those program genres---no matter what the stats say. In every case an objective media mix investigation---based on audience targeting relative to cost and even including factors like ad attentiveness will show that the advertiser is crazy to be spending so much money in news, sports and/or specials. I've seen this many times not only in the past but recently. Yet they will keep buying what they like---for "other" reasons.
To Mr Papazian- wouldnt it be logical that the CPM for TV news, sports and/or specials be higher simpy because the # of eyeballs are significantly higher, and the demographics (age in particular) are lower? When we know that Video is such a dominantly effective method of promoting a brand message, it just makes sense to reinforce that VISUAL in other channels such as TV and Billboard. Its like wearing "matching accessories"- together they all enhance an overall (positive) impression.
Dana, rating size ---or the number of eyeballs---has nothing to do with it. The networks and stations have always charged huge CPM premiums for news, sports, and specials because they know that a certain kind of advertiser wants to be associated with such vehicles and these advertisers are willing to pay far more per set of eyeballs as these are seen as image enhancers, plus there are promotional tie -ins with the sports, teams and athletes, etc. as well as general merchandising benefits that transcend CPMs or refined targeting considerations.
Bravo Cory.
In the realm of media measurement, the 'easy part' (not that it is easy) is to quantify a medium in isolation. Measuring the media owners (i.e. share) is harder but can be done. Measuring the entities that each media owner has requires more investment (i.e. programmes, titles, dayparts, sites etc.). Measuring it daily or in real-time adds lots of cost.
Measurement for all major media - using common metrics - is VERY expensive.
Then the hard part starts. How you measure all of them simultaneously with the same granularity and with the same metric definitions is such a gargantuan challenge that I doubt that funds (let alone universal agreement amongst the media owners) would be available.
Then comes the REALLY hard part!
How do you drill down to the level of detail that the advertiser/marketer is looking for. How would such a system accurately reflect and differntiate Coke drinkers vs Tide users vs Budweiser drinkers? It can't.
So I agree with Cory that MMM is the path to follow for brand advertisers. Yes, it is/will be a lot of work and require a lot of data, but the marketer will get their own bespoke model (generally explaining 70%-80% of sales movements) that they can experiment with to gain the uopoer hand over their competitors.
And John, it's not even that expensive anymore... Cloud-based vendors like Big Chalk can run MMMs for roughly half the cost of the big houses like Nielsen, Ipsos, etc. And with faster turnaround times.
Rick, I was not referring to the cost of the actual MMM, but was referring to the cost of the increased volume of wider and more granular data in order to provide more robust MMM results.
We have the opportunity to evolve what MMM is.
Digital media infrastructure changes what MMM could be, where and how it is applied. New MMM as the *beginning* of the media planning process and when combined with advertiser 1st party uploads against a Synthetic Society uncovers the audiences where the product is underindexed in sales. Only MMM can work in tandem to allocate whether overindexing of sales is from better distribution, weather, temporary price reductions, end aisle displays.
Cookies, were never that good. No conversion pixels are MRC audited and accredited for Invalid Traffic. Does it make sense to accept we have Bots on impressions, Bots on websites but there are no Bots driving conversion pixels events? C'mon, the digital players who really know the game - know that there are huge problems relying on conversion pixels which means that Multi-Touch Attribution is delusional.
So - sharpen up your direct marketing playbooks, get back into geo matched market testing and make sure your MMMs are driven by real-time APIs - digital has changed the game for MMM.
Perhaps most imporantly is something that is only walked by that graveyard, maybe accompanied by whisteling: that more granular data on audiences and their interactions with media aren't really necessary. Knowing that my customer for toilet paper is a club-footed home office radiologist who likes cheese is unnecessary and not terribly important data. While some level of salience is important to make and keep connections with customers or potential customers, knowing more about them frequently doesn't change the calculus of engagement and putting them on a path to purchase. It can have slightly more impact on the content the creative assets contain, but when it comes to the media itself: if I'm McDonalds, I mostly only care if you have a mouth; knowing more about the audience, having more data on the audience, doesn't change the basics. Digital, while it has become the viscera of the body media, enabled unuseful practices by focusing too hard on trees thereby missing forests.