Unless you’ve been living under a rock, it should be fairly apparent that there have been major factors impacting marketing measurement in recent years.
Privacy regulation, cookie consents, the deprecation of third-party cookies, and stricter privacy solutions implemented (ITP, ATT, and probably some other acronyms) have quite literally up-ended the world of (European) marketers, and especially eCommerce/D2C players that have previously relied on accurate, yet invasive measurement.
To boot, there’s a constant debate about the legality of market-dominating analytics tools and platforms – but that is a fight I do not wish to participate in. And a few weeks ago, there was an interview of Les Binet on The Drum with a title claiming that digital attribution is dead – which expectedly stirred a healthy debate.
As technical capabilities have improved in leaps and bounds, advanced solutions such as marketing mix modeling are more accessible today than as little as five years ago. This alleviates some of the pain for marketers, but is it enough?
The bottom line is that measuring marketing effectiveness is hard. There is no silver bullet, no one-stop-shop to address each need and use case. Fortunately, constructing a workable solution is possible – it just needs a bit of time, effort, and patience. I’ll introduce my way of thinking which consists of three levels; helicopter view, street view, and the microscope.
Let’s begin with the zoomed-out helicopter view. In my opinion, this is where Marketing Mix Modeling, or MMM, sits.
If you’re not familiar with the concept: Econometric modeling, better known as Marketing Mix Modeling (MMM) is a statistical analysis method (e.g. bayesian) that helps brands quantify their baseline sales and marketing activities’ incremental sales impact. It allows brands to determine the return on marketing investment and whether budgets are optimally allocated.
MMM involves analyzing large amounts of granular sales and marketing data from a wide range of sources (from ERPs, commerce platforms, analytics tools to advertising channels to name a few) to model the impact of marketing activities, but also create forecasts of future sales performance based on changes in marketing strategy and budget allocation.
TL;DR: MMM provides a holistic, top-down view into the sales contribution of marketing channels, campaigns, or tactics.
Sounds great, right? My view is that MMM is or will be a core component in any marketer’s toolkit going forward, but it does come with… challenges.
For starters, modeling requires a large amount of data, usually the more the better. To get meaningful and accurate results from MMM you’ll need clean and granular data, preferably from a longer period (+ 2 years). You can feed new data to the model continuously, but spotting larger shifts or patterns often requires time, making operative, day-to-day decision-making based solely on MMM challenging.
What this means in practice, is that MMM provides an excellent overview of how your marketing efforts are doing, and which ones are contributing to your bottom line in the big picture. It can additionally help marketers create scenarios to forecast future performance.
However, due to the sheer volume of data and time required, using MMM for making operative, real-time optimization decisions is not optimal. This brings us to step two...
Digital attribution has a somewhat controversial reputation. It’s kind of a known secret that attribution hasn’t been that accurate for a long time, maybe ever, but marketers have diligently continued to use it regardless. Possibly because there hasn’t been a reasonable alternative, or just because that’s what most marketers are used to.
For the uninitiated, attribution modeling is a way to determine which marketing touchpoints are responsible for a conversion. Multi-touch attribution considers multiple touchpoints and then assigns credit to each of those touchpoints based on the amount of influence it had on a customer’s decision to convert. First and last-touch attribution, on the other hand, assign all credit either to the first or last touchpoints.
Doing the above accurately, reliably, and scalably is incredibly difficult, or most likely impossible. Users move between devices and browsers, cookies are blocked. At the same time, approximately third of your web visitors and more than two thirds of Apple users opt-out of tracking. It is fair to say, digital attribution is not representative of the full truth across the buyer's journey from first interaction to eventual purchase.
Once you are aware of these constraints, and understand the data you’re seeing only represents a fraction of the truth, digital attribution can have a role in short-term, day-to-day optimization. Results are available either in real time or within a few hour delay, which allows making adjustments to tactical, bottom-of-funnel activities daily on campaign and ad group levels.
There have been advances in digital attribution as well. Conversion modeling is the use of machine learning to assess the impact of marketing efforts when a subset of conversions can’t be directly linked to ad interactions. This helps create a more complete and accurate picture of your campaigns’ performance. Tools such as SegmentStream evaluate each visit and predict the user’s probability to convert in the future. When this probability is sufficient enough — a “modeled conversion” is created and immediately attributed to the traffic source.
Factoring in the recent advances, digital attribution still isn’t dead but simultaneously it’s still quite fundamentally flawed as a method for crafting a full, 360° view of your marketing’s effectiveness.
I think you can see where I’m going with this, so rather than MMM and digital attribution being mutually exclusive, I see them as complementary to each other.
The final level of in-platform measurement is the most granular, and the most hands-on.
In-platform measurement refers to the performance of a specific channel or platform, and as such, doesn’t share any data between other platforms. In this instance, we’re talking about channels like Meta or Google on the paid side, your email marketing’s built-in reporting (think HubSpot or Salesforce Marketing Cloud).
On this level, we’re mostly concerned about data volume and quality. Most of these platforms employ algorithmic optimization of some sort, and the more data you’re able to supply, the better. When any of these platforms know which placement, format, asset, offer or copy results in a conversion, they’ll most likely do a better job than any human at optimizing your marketing activities.
As you might have noticed, we’ve barely discussed creatives and content. And that’s effectively what you should look at, obsessing over and continuously optimizing on the channel level. MMM most likely will not have sufficient data volumes to model the impact of individual ads, attribution wasn’t exactly designed for that purpose either so what you’re left with is the data housed within each channel.
You’ve probably pieced it together now… You’ll need all three levels to be successful.
Phew, that took a while! If you’ve made it this far I’ll leave you with a few bullet points to wrap this up:
Marketing Mix Modeling can help marketers understand the long-term impact of their marketing efforts on sales, not just the short-term impact. Don’t focus only on digital channels, or only paid ones, aim for a holistic dataset.
You don’t have to stop using digital attribution, just be aware of its constraints and capabilities.
On the platform-level focus on data quality and volume! Use server-side tracking and conversion APIs to boost signal strength, and spend more with your creatives.
Can’t trust your data and optimization is a nightmare? Drop us a line.
...or cut a few corners and book a meeting directly.