One of the least pleasant tasks in marketing is determining which campaigns (and parts of each campaign) are bringing home the bacon (and which ones leave you empty-handed). This is usually due to the task at hand being quite laborious, the tools developed for specific channels and not for a holistic approach, and the outcome ending up too cryptic for actionable next steps.
And still, data-driven marketing is becoming more and more dominant competitive advantage. Consumers face daunting amount of advertising wherever they go nowadays, so to distinguish themselves from the sea of messages companies need make sure the channels and campaigns are moving the needle in terms of revenue and profit, and not just impressions and clicks.
Regardless of the approach, the objective for the marketing attribution should be the same– to define whether or not a specific campaign and/or investment wasprofitable, and what were the main drivers behind this outcome.
What is attribution modeling / marketing attribution in practice? Simply put an attribution model is the ruleset that defines how to allocate the sales uplifts to different marketing activities and touchpoints through the buyer’s journey.
There are multiple use cases for leveraging the results, with the most common one being budget allocation optimization between different campaigns and media channels.
As the outcome should tell the modeler a story about profitability (or the absence of it), it helps marketers to make more informed decisions where to invest their future budgets to make the most out of them.
Whether you’re trying to understand past results or building your case for a bigger budget for 2022, there’s three marketing attribution alternatives that have different pros and cons, when it comes to easiness, data requirements and outcomes:
Multi-Touch Attribution, Marketing Mix Modeling and Multi-channel Attribution.
Multi-Touch Attribution (MTA) is one of the most common attribution models in use nowadays, largely due to its relatively easy approach to the actual attribution. Based on the MTA model, the sale is allocated to the first or last touchpoint, or something in between the first and the last touchpoints.
MTA includes various models, enabling the marketer to select the best fit for the business model. For example:
- First-Touch Models – which allocates full credit to the first recorded touchpoint – can be a great tool to identify traffic-drivers for highly transactional products that don’t require a lot of persuasion
- Whereas Last-Touch Models – which does the opposite – could work for nicely for companies interested in discovering which activities and channels convert prospects into customers during final steps of the buyer journey.
Other MTA models include for example:
- Linear MTA Model, which gives each touchpoint equal amount of credit for driving the sale. Great model for understanding that each touchpoint has its own impact towards the sale, but in reality consumers aren’t equally impacted by every ad & activity.
- U-shaped MTA Model, which gives 40% to the first and last touchpoints, and divides the remaining 20% between touchpoint between the first and the last. Preferred by certain marketers, as the first and last touchpoint tend to stick out in the consumer’s mind.
- Time Decay MTA Model, which gives more credit to touchpoints closest to the actual sales, and vice versa. Easy to explain to your CEO and CFO (as they probably like the approach of highlighting channels that convert sales), but these models neglect the fact the interest towards the product has been sparked way earlier, and the latter touchpoint could be ineffective without this initial interest.
- Custom MTA Model, which allows the marketer to weigh each touchpoint based on their own ruleset. Extremely useful for marketers that have a good understanding of the buyer journey.
As can be seen from each approach, MTA models can miss crucial information by emphasizing only certain part of the buyer journey. MTA models also tend to work on online-only marketing mix, as this method requires tagging the consumer in each touchpoint – a feature that will be extremely difficult to implement when GAFA (Google, Amazon, Facebook & Apple) place their limitations on cross-platform tagging.
Time will tell if there’s any future for pixel-based solutions, such as MTA.
Marketing (or media) mix modeling (MMM) on the other hand doesn’t rely on consumer-specific tags or pixels. Instead, MMM leverages statistical analysis such as multivariate regression or Bayesian inference in attributing sales uplifts to different marketing tactics, as well as seasonality, pricing, promotions and macroeconomic factors.
As the building blocks in MMM consist of sales and marketing time series data, the scope of the modeling can cover both online and offline marketing activities, assuming the (marketing) data is in proper shape. These solutions also evade the upcoming cookie apocalypse, as they don’t depend on identifiable tags or pixels.
Also different to MTA, MMM breaks the sales down into two components, Base sales & Incremental sales.
Base sales represents the part of the sales that would’ve happened even if none of the marketing activities wouldn’t have happened. Base sales usually grows over time due to various tactical and branding activities, and it’s often utilized as an indicator for the brand strength. Base sales is also affected by external factors such as seasonality.
Incremental sales is the proportion of sales that’s been generated by marketing and promotional activities (as in this part of sales happened solely due to the marketing activities). MMM usually decomposes incremental sales into subsets that are each driven by different marketing components, i.e. media channels and promotions.
Usually MMM requires rigorous model validations, as landing on a robust model with the first tries Is highly unlikely. As the results of MMM include a lot more explanatory power compared to MTA results, they are better suited for simulating different scenarios (whereas in MTA the results need to be extrapolated, if the scenario is different to existing dataset).
The downside of MMM is that it’s traditionally been quite heavy initiative to implement. The models require substantial amount of data in order to provide reliable results, and the datasets should have consistent formatting over time to pass the bar. Another limitation of MMM is that the results tend to favor time-specific media, and short-term uplifts while neglecting part of the long-term impact.
Multi-channel Attribution (MCA) is a mixture of the above mentioned MTA & MMM. While continuing on the individual-level consumer data, MCA expands the scope to wider set of tactics, including both online and offline touchpoints.
How can you have offline touchpoints when in MTA it was impossible, you might ask. It’s not as easy as following the footsteps in digital environments, but it can be done.
Connecting the dots can be achieved by for example the following techniques:
Foot Traffic. A lot of the applications found from smartphones nowadays request location while using the application, which has paved the way for marketers to utilize beacon technology in locating the consumer – or more specifically the device ID –physically. This feature requires that the device has its Bluetooth on (which many of us do to use our wireless tools), as well as consumer’s consent naturally.
Point-of-Sale (POS) Data. Shopping at physical stores yield traceable location data, if the purchase is made with a credit card, and if the finance company is working with some of the data companies gathering this sort of data.
Consumer Panels. Companies can also work directly with the consumers by asking them to share their offline behavior by opting-in via an app or implementing surveys to better understand specific buyer journeys.
Multi-Source Matching. Connecting the dots between the POS Data, Foot Traffic and Consumer Panel results can reveal additional information about possible offline touchpoints. The difficult part here is to evaluate whether your assumption matches what happened, and there can never be full certainty. But as in MMM, having highly probable information to guide the decisions is often more than enough to build the hypothesis and test things, which is the desired outcome in any modeling initiative.
In terms of the sales attribution, MCA utilizes similar models as in MTA (e.g. linear model, U-shaped model etc.).
The final judgement comes down to one of the most clichéd statements there are – it depends.
When picking a marketing attribution solution, whether it’s an in-house approach, SaaS-tool or third-party provider, one should consider three things: Data amount & quality, Media Mix and Future roadmap.
Especially MMM requires a boatload of data, and this approach doesn’t work with either insufficient data amount or quality. If you haven’t maintained your data on a decent level, chances are you have to resort to simpler models, such as First- and Last-Touch Models.
Companies with online-only/online-dominant media mix benefit more from the MTA, as it’s been developed for this environment. Considering the ease of use of MTA, marketers that favor online marketing should (and in most cases already do) prefer MTA over MMM & MCA. The only pitfall here is to select a model that captures the customers acquisition process the best.
One of the use cases of marketing attribution is to build hypotheses for future tests. If you’re looking to try out new scenarios, MMM provides the most reliable insights due to the fact that it includes various crucial factors, such as promotional activities, seasonality, base sales development and macro economic drivers.
Learn more about Marketing Attribution from this Guide >>