Media mix modeling for mobile apps: a privacy-safe answer to marketing measurement?
Is media mix modeling the answer for mobile apps user acquisitions leaders who want to measure and optimize marketing and advertising?
Here’s the UA leader’s dream:
- Immediate, accurate marketing intelligence
- Actionable insights on channel, creatives, bids, and budgets
- Privacy-safe, Apple compliant, GDPR-happy, CCPA-OK methodologies and mechanics
- … all of which result in smart insights that generate massive growth
It’s a tall order for any methodology of marketing measurement. Can media mix modeling deliver that for mobile marketers?
Let’s dive in.
Media mix modeling is a senior citizen in marketing (not that that’s a bad thing)
There’s no really kind way to put it: media mix modeling is old.
Often referred to as marketing mix modeling, it’s been around since literally the 1950s and 1960s. Neil Borden, a professor at Harvard Business School, first coined the term marketing mix in 1949, and big brands used the concept in the emerging age of mass media to build statistical models of marketing effectiveness.
In that form, media mix modeling was effective at answering questions like these:
“If I spend $10 million on a TV campaign on the three big networks, what impact will it have on my sales of Cheerios?”
To build those answers, media mix modeling needed data: prior sales, ideally for years. Impacts on those sales. Seasonal trends. Competitive actions. Economic indicators like the consumer price index. Pricing comparisons. Availability data. Wider cultural movements, and so on. Marketing mix modeling typically used (and uses: it’s still in operation) intensive analysis of lots of data from lots of different sources over a long period of time. When it works, marketing executives get answers to the big “what if” questions like what will tossing $10 million at ABC, NBC, and CBS do to our bottom line?
This reliance on scale and history and long periods of time would seem to disqualify MMM as a methodology for the much faster-paced and lower-scale mobile user acquisition campaigns. And, indeed, when I recently talked to Adobe about their brand-new still-in-beta marketing AI-driven media mix modeling product, that suspicion seems to be accurate.
It works well: the system allowed one customer to cut ad spend by 50% while still growing 10%.
But there’s a big caveat: you need to be spending $50 million a year.
Well, some of the biggest mobile publishers are at that level, and beyond it. But certainly not the average mobile publisher. Most mobile marketers have much smaller budgets and need to know what’s going on with their $100,000/month spend or their $50,000/month budget.
Mobile marketing has been drunk on data
Let’s just be honest. Mobile advertising has been drunk on data: immediate data, accurate data, actionable data. Mobile growth professionals haven’t had to bother with pre-digital modes of marketing measurement … partly because they seemed ancient and inexact and cumbersome and expensive and slow.
But mostly because digital offered a dream: a pure and clean and bold dream of exact knowledge and perfect awareness and scientific marketing resulting in no wasted dollars.
(Fraudsters, of course, loved that dream.)
But despite challenges — and there are some beyond fraud — that dream required a level of tracking that 1960s marketers would be amazed at, if not shocked. It started with cookies on the web in 1994, continued with hard-coded device IDs on smartphones in the early 2010s, shifted to more user-controllable advertising IDs shortly thereafter, and is currently trending to largely disappear in the early to mid-2020s.
No data to gobs of data to much less data, all in about a quarter of a century.
But for the complete to-date lifespan of the mobile-first computing era until now, we’ve been in an era of data glut.
“We’ve had a data addiction in mobile,” says Brian Krebs, who runs MetricWorks and builds media mix models. “Because it has always been available.”
Hello again, media mix modeling (the times they are a-changin’)
Because those times are changing, driven by Apple, iOS 14.5, government regulation, and changing social attitudes, media mix modeling is coming back into the conversation. If you can’t get as much granular data via IDFA on iOS, and you’re wondering what might happen with AAID/GAID on Android, maybe there’s another way.
Maybe it doesn’t take a $50 million budget, and maybe it also doesn’t take as much crazy math and exogenous data about the surrounding world and competitor’s actions and economic shifts as we once thought.
If you pour in first-party data like day-of-week trends, vertical-specific trends like time of year seasonality, broad industry trends for app install frequency, plus specific-to-you events like getting featured by Google or Apple or having your game reviewed by GamesBeat or PocketGamer, that’s actually enough, says Krebs. Add to it the marketing activity for your app — spend and impressions — and you’ve got a good basis for MMM for mobile user acquisition.
You do need to vary spend from time to time — a steady-state drone makes efficacy hard to pinpoint — but that happens naturally with changing competition for impressions and availability of supply.
The reality might be that media mix modeling is better for mobile user acquisition than it is for its originally intended purpose: big, slow, spendy campaigns by massive national and global brands. And it may show incrementality better than traditional methods as well.
“It was surprising to us to be perfectly honest: MMM just works better in mobile,” Krebs says. “It’s simpler … you need fewer data points.”
But is it good enough?
What media mix modeling won’t give you
Look. A technique doesn’t have to be perfect to be useful. Anyone who thinks last-click attribution is a perfect measure of marketing performance is, frankly, delusional. And yet we’ve been largely using it for the last decade with, arguably, good effect.
Because it’s been good enough.
So the question is: is MMM good enough to drive decision-making?
Before we answer that, here’s what media mix modeling won’t provide.
Perhaps the three words that capture it best are immediacy, granularity, and clarity. Immediacy was of course the incomparable glory of the IDFA (which still exists, clearly, for 10-20% of iOS traffic) and still is the shining north star of GAID-based attribution. (It’s lacking — to an extent driven somewhat by your own decisions — in Apple’s SKAdNetwork framework, of course. You could reset the SKAN postback to as long as 7 days.)
Knowing what works as soon as possible is critical to quick optimization. Knowing what doesn’t work as soon as possible is critical to not wasting budget.
Media mix modeling for mobile doesn’t take the weeks and months it requires in the big brand consumer space. But it’s not quite as fast as a postback, either.
Granularity is similar: IDFA/GAID and even SKAdNetwork give you pinpoint precision data on a number of factors: source and campaign, a few configurable factors for SKAdNetwork, plus of course much more granular user-level data and cohorting for IDFA/GAID. There’s less data from SKAdNetwork, but smart customization and hybrid postbacks with encoded values for multiple events and/or timers can help.
MMM will provide less in each of these areas, including a little less clarity into helper networks, assists, and last touches, especially for mobile growth marketers who are accustomed to running with many different ad partners simultaneously. (Yes, you can argue that all those networks are casting hooks into the same river, and that’s true. But it’s hard to argue that all the hooks — and the bait on them — are of identical value, and are all cast into the very best parts of the river where fish congregate.)
Incrementality will also struggle with creative optimization, which of course you can address by manually limiting creative variations by campaign to see impact over time. And bids and budgets on individual platforms will be a little more opaque to MMM.
So … a hybrid model: next-gen attribution
Media mix modeling isn’t a silver bullet, someone who sells media mix modeling for mobile user acquisition teams told me. But it is useful. And, alongside an MMP, it adds context and insight.
The reality is that the age of tracking is ending.
IDFA is largely gone, fingerprinting is against Apple’s guidelines, and Google will be making some privacy changes over time as well.
So you do need next-gen attribution.
That means impressions, clicks, and costs, sure. Installs, when you can get them. Creative insights, as much as possible. Channel and partner-specific results, as available. Upper funnel data and lower funnel data. First-party data from within your own app: new users, engagement patterns, sign-ups, purchases. Bids and budgets, and normalization and standardization across all your data sources.
The reality is that it’s getting tougher out there. Last click on an advertising identifier was simple. Now, mobile growth professionals need a complete marketing data infrastructure, not just a mobile tracker. And you need that, by the way, across more than just mobile ads. There’s out of doors, TV, web, and other channels that are starting to matter.
Combining all that signal while simultaneously silencing the noise to generate insights for growth: that’s the next challenge. And MMM has a seat at that table as part of next-gen attribution.
Talk to us
Interested in learning more about how you can futureproof your marketing growth with next-gen analytics and attribution?
Let us know. We’ll be in touch.