The Marketing ROI Goal: Real Attribution with Einstein and Machine Learning

 

This session, led by Jaime Lopez, General Manager for Marketing Operations at Wärtsilä, explores the ambitious journey to prove the direct correlation and causation between marketing spend and sales revenue. The focus is on leveraging Pardot data depth and machine learning to move beyond simplistic attribution models and calculate true marketing ROI (Return on Investment).

Key Takeaways

 

• Marketing Must Prove Causation: It’s no longer enough to show that marketing and sales correlate; marketers must prove that measurable marketing activities cause sales to justify their investment.

• Einstein is a Game Changer: Pardot Einstein Attribution is a powerful tool that automates complex attribution calculations, largely eliminating the pain points of manual legacy models.

• The Contact Role Problem is Solved: Einstein Attribution solves the biggest attribution gap—sales reps failing to create Opportunity Contact Roles—by using heuristic algorithms to create virtual contact roles.

• Model Validation is Key: For large enterprises with long sales cycles, a deeper, custom attribution model is necessary to isolate marketing’s true contribution to profit (EBIT) versus sales contribution.

• Webinars Deliver Highest ROI: For Wärtsilä, webinars were found to be the channel with the highest ROI due to low cost and high pipeline sourcing.

1. Leveraging Pardot Einstein Attribution

 

Einstein Attribution is a core Salesforce product (available in top-tier Pardot/Sales Cloud editions) that replaces legacy attribution models (First Touch, Last Touch, Even Touch) with a data-driven approach.

Einstein’s Calculation Method

 

1. Virtual Contact Roles: Heuristic algorithms determine which individuals should have been on the Opportunity Contact Role and virtually create those connections.

2. Shapley Values: Concepts from game theory are utilized to model the outcome (Opportunity Size) as a cooperative game. The Shapley value correctly splits the value of one Opportunity across all the marketing campaigns that influenced it.

3. Outcome: Provides granular attribution (to the euro/cent) per campaign, eliminating the dependency on manual sales actions.

The Limitation

 

Einstein Attribution assumes that 100% of the value of the Opportunity is apportioned to marketing campaigns. For B2B enterprises with multi-year, multi-million-euro sales cycles, this is not accurate, necessitating a deeper, custom model to isolate marketing’s true share.

2. Real Attribution: The Custom Machine Learning Model

 

Wärtsilä built a custom, deeper attribution model to calculate the exact percentage of sales value and profit (EBIT) that is directly attributable to marketing.

Model Architecture and Methodology

 

• Data Input: The model fetches rich data sets including Prospect Activity (the most valuable data from the Pardot API), Campaign data, Opportunities, Accounts, and Contacts.

• Data Prep (Fuzzy Matching): The model performs fuzzy matching to connect related accounts (e.g., Carnival Maritime, Carnival Corporation) that are marketed to separately but belong to the same parent company, ensuring complex relationships are tracked.

• Modeling: A Supported Vector Machine (a classification algorithm adapted for regression) is used to find nonlinear functions that fit the data, understanding complex relationships better than simple linear regression.

• Post-Production (Isolating Effect): The model can simulate the cancellation of marketing (i.e., running the model as if there were zero marketing activities in a given year) to precisely isolate and quantify marketing’s impact on revenue.

• Validation: Each scenario is run thousands of times to establish a probability distribution and state results with high confidence (e.g., 95% confidence).

Key Findings and Operational Decisions

 

• Marketing’s Share: Wärtsilä determined that 10.3% of its revenue last year was due to marketing activities (with 95% confidence).

• Marketing ROI (EBIT): For every €1 Wärtsilä spent on marketing, it produced €3.45 in profit (EBIT). This moves the conversation from cost to profit generation.

• Operational Insights (Shapley Values): The model identified specific factors that influence purchasing likelihood and marketing effectiveness:

  • A high Pardot Score of an account is strongly correlated with a higher likelihood of buying.

  • Companies that buy more types of products are more responsive to marketing.

  • Companies that buy high volumes of field services (maintenance) are less responsive to marketing.

3. Technology and Strategic Advice

 

Data Access and the Pardot API

 

• API Value: The Pardot API is essential for deep attribution because it allows extraction of the rich and complex Prospect Activity data.

• API Caution: Due to organizational politics, Wärtsilä pulls data from an internal enterprise data warehouse (EDW) instead of connecting directly to Salesforce, but strongly advises other organizations to connect the API directly to Pardot/Salesforce if possible. Always ensure APIs are set to read-only to avoid data corruption.

Strategic Channel Focus

 

• Webinar ROI: Wärtsilä’s biggest ROI came from webinars. After scaling up from one per quarter to 15-20 per quarter, webinars sourced a significant portion of pipeline at a very low cost.

• Be Ambitious: Marketers must be ambitious and leverage the depth of Pardot data with machine learning to gain operational insights and claim their seat at the revenue table.