Is there a generative AI approach to marketing measurement?
In the world of B2B marketing analytics, most companies are still using rules-based attribution models. These models use rules to determine the impact of each advertising channel on the overall success of a campaign by assigning credit to steps in an individual’s or company’s path to purchase. However, with the advancement of data science techniques, there are more sophisticated models that can now be used.
First, let’s review attribution models. Attribution models are descriptive statistics models used to determine the contribution of each marketing channel to the overall success of a campaign. These models look at historical data and use rules to identify the channels that are most effective in driving conversions. I call them “descriptive statistics models” because all they really provide are backward looking insights.
Attribution models can be divided into three categories: single-touch, multi-touch, and algorithmic. Single-touch attribution models give all the credit to one touchpoint, while multi-touch attribution models give credit to multiple touchpoints. Algorithmic models give credit to touches that lifted the likelihood of success. While this is more sophisticated than rules-based, it still just leaves you with a descriptive statistic. Not a recommendation.
The problem with these types of models is the limited insights. Did a marketing channel get 30% credit because you put 30% effort into it? Is 30% good? What about touches that didn’t generate a record into your CRM? How do you measure those, and should you do more of them?
Enter media mix models! Media mix models are a type of generative AI measurement model. These models use machine learning algorithms to simulate the impact of different marketing channels on the overall success of a campaign.
What makes them generative AI?
- They generate data that didn’t exist.
- They incorporate predictive modeling techniques.
- They simulate complex interactions:
Unlike attribution models, media mix models do not solely report historical data. Instead, they use machine learning algorithms to simulate different scenarios and predict the outcomes of different marketing strategies. This allows marketers to make data-driven decisions based on the predicted performance of their advertising campaigns.
And the best part – they don’t require stitched purchase paths from a CRM. They look for macro relationships between marketing channels and success metrics. So, if your path to purchase is complex or disparate, a media mix model becomes even more ideal.
When should you make the switch from attribution to MMM?
- If your path to purchase is more than a touch or two.
- If you’re spending advertising dollars across more than 3 or 4 channels.
- If your data is disparate.
- If you need to optimize your budget and want to explore different spend scenarios.
- If you have offline media that you want to measure.
- If you want forward looking recommendations instead of backwards looking metrics.
As B2B marketers, we were taught that path-based attribution models were the holy grail. But I think we are finding that the paths are too complex and insights are too limited with rules-based attribution. If decision making is the ultimate goal, generative AI models, like media mix models, will better inform future decision making and performance than any backward looking attribution model will.
And yes, a little bit of generative AI was used to write this. But probably less than a quarter of it.