It’s 2023. Big tech is going through its biggest round of layoffs since that sock puppet company went down. App Store revenues have shrunk for the first year ever and every mobile gaming company, the biggest driver of mobile app revenues, is struggling to stay afloat. Companies that made exits in 2021 and 2022 are patting themselves on the back (or at least their investors are), and those who didn’t are praying for a brighter 2024.
There are many factors involved in this state of affairs, but a large one is the App Tracking Transparency framework introduced by Apple in June of 2021 and the continuous march of privacy changes across that industry that will continue into the future. This rapid degradation in the effectiveness of advertising has affected all businesses, but especially those focused on growth over profitability.
As long as internet advertising has existed, marketers have relied on attribution and attribution models as a way to measure performance on marketing campaigns and improve returns over time. It is a valuable tool, but quickly became a major crutch for digital companies and traditional companies advertising in the digital space. More and more organizations are struggling with the need to move their marketing budget decisions away from an attribution only approach, but many are addicted to this method because of the consistency and the ease with which executive teams can understand the deterministic approach. This leads to underinvestment in channels that are more upper funnel like awareness campaigns, or can’t be tracked well, like linear TV and influencer campaigns.This creates major inefficiencies in marketing investments, which will only get worse over time.
Older businesses, that grew their marketing functions before the internet, solved this problem by having separate brand marketing and direct response marketing budgets with different goals. This sidesteps the problem, but makes for a poorly integrated marketing strategy. To be truly efficient in marketing investment, businesses need a fully integrated and coherent measurement approach.
The Risks of Relying on Attribution
One of the main risks of relying on attribution is that it fails to take into account the incremental impact of marketing channels. Attribution models assign credit to touchpoints based on their position in the conversion path, but they do not account for the fact that some touchpoints may have been unnecessary or redundant. This means that marketers may be overvaluing certain touchpoints and undervaluing others.
The most egregious situations for overvaluing are brand keywords in paid search and reengagement campaigns.
In the case of brand keywords in paid search, it is inconceivable that the vast majority of users who searched for your brand name were not going to continue to your site regardless of whether you bid on those placements. The gap between measurements based on attribution and measurement of incrementality is often huge.
In the case of re-engagement or retargeting campaigns, working with the assumption that no customers would have returned without the exposure to advertising, despite having already interacted with your site, or purchased products from you, is clearly faulty. Customers return to a product spontaneously all of the time, and the tracked touchpoints are often just incidental.
Another risk of relying on attribution is that it may encourage marketers to focus on short-term metrics rather than long-term value. For example, if a marketer is solely focused on the last click, they may be tempted to invest heavily in channels that drive immediate conversions rather than channels that have a longer-term impact on brand awareness and customer loyalty. This is more apparent the more friction there is in the conversion funnel. For a free app-install product, where the user is prompted to install, with no paywall, and slowly introduced to the product before being offered microtransactions or a subscription, this may be a small issue. For products with high upfront costs, long registration processes, or long commitments, this can result in a huge disconnect between the ideal media mix and what measurements based on attribution would suggest.
Attribution on the web vs. mobile
Attribution on the web and mobile can be very different due to the different tracking mechanisms available. On the web, cookies can be used to track user behavior across multiple touchpoints, making it easier to assign credit to each touchpoint. However, cookies are not as effective on mobile devices, where much interaction and most ad views happen within apps rather than on a browser. For tracking between mobile apps, advertisers have historically used unique device identifiers, but that tracking has been subverted on iOS and now the primary cross-app tracking solution on iOS is SKAdNetwork.
This increasingly fragmented world of attribution increases the risk of failing to accurately assign value to different marketing campaigns as most users work across multiple devices, browsers, and apps, and attribution methods are incapable of creating a coherent picture of the users journey from learning about your product or brand to converting to a paying customer.
A brief introduction to incrementality
Incrementality is a measurement technique that aims to measure the true impact of marketing efforts by comparing the behavior of a test group that is exposed to a marketing campaign to a control group that is not exposed. By comparing the behavior of these two groups, marketers can determine the incremental lift that the campaign provided.
In practical application, this can be done by implementing control groups on a specific ad platform, exposing potential customers to an ad campaign in one geographic area (geo-split), or activating and deactivating a campaign to observe metrics pre and post.
Incrementality is valuable because it takes into account the impact of all touchpoints, including those that may not have been captured by traditional attribution models. It also allows marketers to measure the long-term impact of their campaigns, rather than just focusing on short-term metrics.
A brief introduction to marketing mix modeling
Marketing mix modeling is a statistical technique that allows marketers to analyze the impact of different marketing channels and other factors on overall business outcomes. This analysis can be used to optimize marketing budgets by identifying the channels that provide the highest return on investment.
Marketing mix modeling takes into account a wide range of factors, including the impact of offline channels, external factors such as the economy or weather, and the interactions between different marketing channels. This allows marketers to gain a more holistic view of the impact of their marketing efforts. A marketing mix model can be as simple or as complex as is needed to accurately model your business. The more factors your include that affect sales, the better your model will be at predicting the future and evaluating different investment scenarios.
Where attribution has long-term value
While there are limitations to attribution, it still has value in certain contexts. For example, attribution can be a superior measurement tool where data sets are small and the context of delivery is constant. In the case of creative testing, attribution data can be a more valuable measurement tool, since it does not require the same volume of data that is needed to measure incremental uplift, and the test is run in a controlled environment.
Where to start
If you are currently using last click or multi-touch attribution models, the first step to integrating incrementality and media mix modeling into your business is to conduct a comprehensive audit. Look at your areas of largest investment and generate hypotheses regarding where you think you may be over or underestimating impact in your attribution measurements and plan incrementality tests where you think you will find the largest discrepancy. Next, make a list of what factors you think have the greatest impact on sales and start working with historical data to come to a first version of a marketing mix model. Once you have a few tests under your belt and a first version of a model, you can start piecing together how you can use these tools on an ongoing basis.