Establishing “Data Readiness” to Drive the Most Value from an MTA Capability
By Robert Cardarelli, SVP at Ipsos MMA
With all the changes in the digital and TV data landscape, the promise of standing up a robust multi- touch attribution (MTA) model has left marketers with several big questions.
- How do I get an attribution model up and running?
- Should I just use “out of the box” software and not worry about what’s behind it?
- What data do I need?
- How do I stitch the data together?
- How do I know the data are accurate?
- What are the hidden truths to make it work?
All valid questions that will directly impact on the buy-in and usage of the solution.
It is an unquestionable fact that user-level data is NOT perfect. Standing up an attribution model requires EXPERT knowledge of the data landscape to understand the limitations, impacts on results, and creative work-arounds. A true partner will identify the latest industry trends and practices, what data are available, what are the strengths and weaknesses, who are the new and relevant players, and how does it all fit into my technology stack. The desire to measure and optimize in-market campaigns on a real-time basis is not going away. To do this right, however, you must get the data right – this has never been truer given the proliferation of fragmented and in some cases, questionable sources, varying levels of granularity, privacy requirements, and sharing issues.
Data Strategy: The single most important (and overlooked) element of MTA
Before embarking on any multi-touch attribution implementation, it is critical to have a data strategy led by someone who 100% understands your business and media ecosystem, tech stack, rules, and sources required to successfully implement it. This requires a comprehensive understanding of ALL the components of user-level data, how to source it, what are the limitations, which proxies to use where needed, and what are the privacy requirements.
At the end of the day, a clear data strategy needs to be in place that links key business questions to supporting data. This strategy defines what sources need to be GENERATED vs. which data touchpoints need to be sourced, how to connect the data through leading technologies, where to use individual vs aggregate data, and what media partnerships are required. The strategy must deliver tangible business value as a unified attribution program requires significant collaboration, transparency and upfront investment to get it right.
First, map out the measurement objectives – what really drives value?
A guided discovery process is core to any attribution engagement – this helps brands become “data ready” to reach their goals. At this phase, key stakeholders align on what is required to be successful. The data strategist matches these required success criteria with their knowledge of the data marketplace – industry trends, key partners, regulations, and what technology platforms are best utilized in the attribution process. What comes out of this process is a clear understanding of what measurement objectives can be met – currently and in a future state. It is not as easy as connecting into an ad-server, getting placement, creative, site, and audience details and measuring attributed revenue. This serves its purpose: a basic, incomplete and inaccurate read on media performance. A holistic plan needs to be architected to ensure measurement objectives are met with leading edge technologies, partnerships, both TV and digital alike.
A recent example where we executed a data strategy was to help a large national retailer measure and balance brand vs. performance media in real time. This required exposure-level TV and brand digital media to be categorized and collected at the same level of granularity as all lower funnel “performance” tactics, such as search, affiliate marketing, and retargeting campaigns. This allowed us to truly understand their interactions, providing budget optimizations that support brand building while still meeting weekly sales goals. This data strategy supported attribution models that optimized these different tactics to answer a specific business objective: how to drive the most brand awareness while removing waste (through frequency optimization) across the lower funnel “performance” channels.
Understand projectability, and control for bias
The journey doesn’t stop after you collect and stitch the data together. MTA data is messy – projectability and measurement bias is a huge issue. Just having a massive dataset, with 100s of terabytes of information doesn’t mean in any way you will have complete or accurate results. Nowadays, more than ever, it is critical to put the right controls in place. Using data from identity platforms, for example, with varying match rate issues needs to be addressed. Working with different levels of granularity, missing data, bad actors, data source providers, etc. all lead to biased data. It is essential to conduct rigorous statistical tests on unified data to ensure it can be used before any modeling takes place. Techniques such as raking, weighting, stratified sampling, and outlier detection should be utilized to ensure campaigns, media types, and audiences are projectable to the intended universe.
Validate, validate, and validate continuously
Once the analytical dataset has been built, and initial models are up and running, tests need to be conducted to validate model results against in-market performance on a continuous basis. Different model-based spending scenarios and simulations need to be tested at the initial model estimate stage against real business outcomes such as site conversions, omnichannel revenue and profit, and other KPIs to ensure that any data calibrations and transformations have been applied correctly. All limitations have to be clearly defined to the client organization before moving into model production.
A good data strategy never ends
MTA requires continuous care and feeding, or it becomes stale and loses its predictive accuracy quickly. To do so, a unified attribution roadmap should be established, tracked, and refined on a quarterly basis with the key stakeholders. A true partner has a data strategist who is constantly collaborating with the media team and looking for ways to improve the model by introducing new sources, refining algorithms, designing experiments, and applying new techniques that drive value. Continuously educating clients on what the data and supporting technologies really are, source by source, is critical for attribution adoption and acceleration. “Plug and play” MTA software alone is not the answer – it is crucial to ensure there is a full understanding of the data beneath it, making sure that your data and technology partners are working seamlessly together, and that value is tracked over time – both on the investments in your data as well as the in the program itself.