The age-old goal of marketing attribution sounds simple: to find out which touchpoints, channels, and/or campaigns are most effective at driving customers to conversion.
In the digital world, our ability to attribute impact is based on deterministic identifiers that let us stitch marketing interactions together. Then a model, either rules-based or data-driven, is applied to these interactions to adjust the weight of credit assigned to each interaction across the customer journey.
So attribution theoretically gives marketers the power to understand and evaluate the value of different kinds of brand interactions on a consumer’s decision to convert. The platonic ideal of an attribution model would produce a holistic view of every touchpoint in the user journey and accurately assess its importance in driving people to the end goal.
Some models have gotten closer, like multi-touch attribution, while others have well-known blind spots, like last touch.
But it is simply not possible to accurately track every touchpoint that may have influenced a conversion event. It probably never will be. And even if it was, attribution might not actually give us the full picture we need.
Attribution alone is an inherently flawed goal
Perfect attribution has been a marketing pipe dream for a long time; since the days of John Wanamaker, marketers have been obsessed with the idea of a universal measurement framework that will prove they are driving value.
But even if you could see all touchpoints and achieve perfect holistic attribution, would that give you all the information you need to build the perfect marketing strategy? After all, attribution by its very nature is always looking in the rearview mirror. It doesn’t look to the future or provide a path forward. It also doesn’t account for critical media investment signals like diminishing returns.
Measuring performance should focus on using data to understand where your next best dollar should go, not just how far the last dollar went. You need to be able to look at your data and answer forward-looking questions like:
- Where can we increase budgets to scale our conversions while maintaining our current ROAS?
- What levers can I pull to optimize campaign performance?
You can’t answer those kinds of questions if you’re only looking at a model that assigns retroactive credit on top of an incomplete data set. The cold, hard truth is that deterministic multi-touch attribution isn’t a cure-all; if that’s the only model you’re depending on to make decisions, it can’t deliver those answers.
Data deprecation is making deterministic multi-touch attribution harder
Whether or not you agree that the very idea of attribution only gets us partway to our ultimate measurement destination, we can all agree that many marketers are still beholden to attribution models. As with all models, there is no such thing as perfect. While there have been advances over the years, with Google’s value-derived data-driven attribution of particular note, there are still plenty of unknowns at play when it comes to the future of attribution.
That’s because the current state of marketing data is only making things harder. As platforms like Meta, Google, and Snap struggle to cope with Apple’s App Tracking Transparency (ATT), the CFOs of those companies would be the first to admit that data deprecation is their greatest challenge.
At least some of the practical problems with attribution are human ones: people are obsessed with connecting the dots and finding patterns, whether or not they actually exist. We often get questions from marketers about how attribution works within Google Analytics because of inconsistencies with other data sets.
You’re probably familiar with this challenge: the Facebook Business Manager UI claims the platform drove 10x more conversions than what you’re seeing reported in Google Analytics. So which one is right?
The answer is they are both wrong, just in different ways. No amount of fun math (i.e. proxy calculations that look at the delta between the two data points over time) will help you solve the equation and perfectly calculate exactly much credit Facebook Ads should get.
At the end of the day, this is a data observability problem; the data is incomplete, but we look for an answer anyway.
If that’s confusing, think about it this way: let’s say you’re on a phone call with patchy reception. For every 10 words, you miss one word. Chances are you can still understand the gist of the conversation because you have so much other context.
But when you start to lose entire sentences or every other word, you’re going to find yourself in trouble. That’s because the inputs are too limited and fragmented to draw accurate conclusions. That is exactly what is currently happening with deterministic attribution across all advertising platforms, and it’s something that no amount of modeling can totally solve.
Remember that Super Bowl T-Mobile ad where Rob Gronkowski invites Tom Brady to retire in Florida, but Brady can only hear every other word and thinks Gronk is telling him to go play in Tampa Bay? That’s attribution today. That’s the reality that brands face with data loss.
This is comparable to how ChatGPT predicts the most likely next word as it compiles responses. Often it makes sense, but sometimes it hallucinates and tells you that Elon Musk is going to be the next president of the United States.
Of course, there’s always a range of error in modeling, but if the data loss is bad enough, you can’t build an accurate model. Then the real question starts to take shape: how much can you trust advertising platforms to get right?
That’s why you need to expand your measurement toolkit.
Media mix modeling and incrementality testing can get you closer to understanding impact
Attribution is still a powerful construct, but the underlying methodology needs to change so it can evolve into the modern era. You need to ask some hard questions to figure out what kind of measurement toolkit and framework will work for your organization, like:
- What measurement do you need in place to make smart decisions about investment planning across channels and platforms?
- Where are the best opportunities to scale your existing media mix as efficiently as possible?
As an industry, we’ve gotten used to thinking of an imaginary version of perfect attribution as the end-all, be-all, but it was only ever meant to be used as a guidepost.
That doesn’t mean it’s not useful. But you need to shift your focus to the future and let attribution be a component of your decision-making, not the only arbiter.
And while there is no perfect solution, there is an imperfect one that gets us a lot closer to the goal: unified attribution combined with media mix modeling (MMM), where you use some deterministic data and model for the rest. The goal is to leverage past data to predict future investments. It’s rooted in growth, not held hostage to past performance.
To get it right, you need to invest in robust incrementality testing, which will help you both validate modeled performance data and get a clearer picture of how your campaigns are affecting the full customer journey.
Geo-based incrementality testing is vital to media mix modeling calibration. It is also the single most powerful measurement solution to determine where you’re over- or underinvested at a given moment in time.
Most brands are not very comfortable with incrementality testing. Some have done it before, but historically the majority aren’t great at it. If that’s where your brand is, you need reliable partners with a predictable methodology that is customized to the needs and quirks of your unique business challenges.
It’s time for a future-facing solution that integrates multiple tools: the performance measurement framework
One of the big challenges with traditional media mix models is speed to action. At Wpromote, we built a high-velocity media mix model and investment scenario planning tool called Growth Planner as part of our Polaris marketing platform to address both data deprecation challenges and actionability.
Growth Planner forms the core of our performance measurement framework. Essentially, it forecasts across a client’s entire year to find the optimal investment of available dollars to hit revenue targets. It also can be used for optimizations on a weekly basis so your brand can stay agile and adapt to new developments.
Growth Planner looks across all of your marketing channels and the entire funnel to maximize margins because profitability is the endgame. It tells you how to invest down to the specific tactic, down to the channel, down to the month, the week, the day.
We make sure the model stays honest and keeps getting better through continuous incrementality testing, and we can feed additional advanced data inputs like predictive lifetime value into the model to further inform investment decisions. Then we feed data from Growth Planner into data clean room analysis in key areas of investment.
Measurement is going to continue to be a challenge across platforms like Google and Facebook and media channels like CTV. If you really want to know how your marketing is actually performing, you need to start exploring privacy-compliant measurement solutions measurement.