We currently have fully customizable static attribution models available within the BrightFunnel platform. We are excited to announce that we have also filed a patent (“Dynamic Attribution Models” by Nisheeth Ranjan, Ranjan Bagchi, and Nadim Hossain) on how to create dynamic attribution models. Before we dive into static and dynamic models, lets define the attribution problem.
The Attribution Problem
B2B marketing teams everywhere are familiar with the revenue attribution problem: How do you credit revenue back to the marketing campaigns that influenced a deal? Similarly, the pipeline attribution problem is: How do you credit pipeline dollars back to marketing campaigns that influenced an opportunity?
Both attribution problems are important to solve because we need to measure the revenue and pipeline impact of a campaign before we can calculate an ROI (return on investment) for the campaign.
Static Attribution Models
BrightFunnel’s marketing analytics platform solves the attribution problem by calculating multiple customizable attribution models (or rules for crediting revenue or pipeline dollars back to campaigns) in parallel many times a day. Four examples of attribution models used by our customers are:
- First Touch: Attribute 100% of the revenue to the first campaign that influenced the opportunity.
- Last Touch: Attribute 100% of the revenue to the last campaign that influenced the opportunity.
- Evenly Weighted: Attribute 1/N% of the revenue to all campaigns that influenced the opportunity, where N is the total number of influencing campaigns.
- X-Y-X: Attribute X% of the revenue to the first campaign, X% to the last campaign, and split the remaining Y% over all the influencing campaigns between the first and the last. The constraint that needs to be satisfied in this model is that (X + Y + X) should equal 100.
All four models above are static attribution models because the rules of attribution are spelled out in advance.
In dynamic attribution models, the rules of attribution are learned automatically (via machine learning approaches) and adjusted constantly.
Dynamic Attribution Models
For a BrightFunnel customer, the main steps to follow in order to create a dynamic attribution model are:
- Divide up the customer’s past data into a training set and a test set.
- Start with a default attribution model and use it to credit pipeline/revenue to campaigns in the training set.
- Use the training set (which now links campaigns to pipeline/revenue) to calculate metrics like lead-to-opportunity conversion rate (LTO%), opportunity-to-deal conversion rate (OTD%), lead-to-opportunity velocity (LTOV), opportunity-to-deal velocity (OTDV), average opportunity amount (OA), average deal amount (DA) for each campaign group.
- Calculate actual revenue (AR) generated by the leads in the test set.
- Use the metrics in Step 3 to calculate predicted revenue (PR) generated by the leads in the test set.
- Calculate the test set error as the absolute value of the difference between AR (calculated in step 4) and PR (calculated in Step 5).
- Change the attribution model parameters and repeat steps 2 through 6 above until the test set error is minimized.
- The attribution model that minimizes the test set error is the dynamic attribution model personalized to the customer.
The above process automatically selects an attribution model that best predicts revenue based on the past data of a particular customer. As time passes and more data is generated by the customer, the above process can be re-run periodically to continually optimize the model.
The dynamic attribution model approach creates a custom attribution strategy tailored to a particular customer’s market’s past behavior. The model is automatically fine-tuned and improved as the BrightFunnel platform ingests more data from the customer over time. The test set error is a key metric that shows the customer how the attribution model is improving over time.
There is an exciting roadmap ahead for BrightFunnel’s attribution platform. We already have the capability to calculate multiple complex customizable static attribution models many times a day. Soon, we will make our attribution platform even more powerful and differentiated by adding dynamic attribution modeling.