Key Insights:
- How to utilize Pardot Einstein Attribution to make operational decisions and qualify leads efficiently
- How much of the value of a deal was actually marketing’s responsibility? ‘Real’ attribution.
- Integrating Pardot data into a full-funnel measurability model
- Statistical background and system architecture of said model
- Data I/O considerations from different platforms
The End of The Marketing Rainbow: ROI
Jaime Lopez of Wartsila explores pushing the boundaries of marketing measurability to get to that ROI pot of gold. This is not a problem that you can attack with your opinion, with your impression, with your bias, with your preconception, it must be done with the data.
How? Utilize tools like Pardot Einstein Attribution — to make very operational decisions about your marketing, ascertain how much of a deal’s value can be attributed to marketing, integrate Pardot data into a full-funnel measurability model, explore data input and output from different platforms.
Data Input and Output From Different Platforms
This is one of the most annoying things to do, but one of the most important things to do. How to take your data out and digest it so that your model can understand it and give you value?
Our chief marketing officer passed by my desk one day and said, “Hey, Jaime, I just told our CEO that I know that roughly for each euro we invest in marketing, we get two insights. But, can you please prove that for me?”
So I thought, okay, wow, interesting — that’s going to be tricky, but I can do it. The first two stages would be to look at our data and try to build a model that understands correlation and causation of our marketing. I need to know more than just marketing and sales correlating together, but as a person with a science background,I need to be able to stand in front of my board and stand behind my numbers. I need to prove causation and that more measurable marketing actually causes more sales.
I needed to show that from roughly 2 million unique visits (visitors to our website), all the way down through the funnel through the 1.6/1.7 billion euros of revenue that have provable and marketing attach points. I need to figure out the correlation and causation. Our Pardot instance is pretty big which means that I’m gonna have plenty of data to create the model, to train the model, to cross validate, which is fantastic and that’s why I really love having sets of examples.
A human would never be able to make heads or tails out of this amount of data. You need machine learning and data to tell the story. I need to emphasize that opinions and facts are two different things. Opinions about what works in marketing are great, and facts about what works in marketing sometimes might make us uncomfortable, but are essential to back up your opinions (or put you on the right track). I’m gonna go with the facts, data, and numbers every time. We went about it in a quite laborious way — we basically wrote our own book and needed to get the help of an analyst to understand our data structures. We then created a model, deployed it, and created objects with it.
Enter Salesforce Pardot Einstein Attribution
But, that is the past and I’m very happy to say that nowadays if you want to recreate this, you don’t have to go through all that pain. Now there is a product by Salesforce that will save you most of the hassle here and that product is Pardot Einstein Attribution. This is available in some editions of Pardot and some editions of Sales Cloud and it’s a product that we’ve been a pilot customer of and we’ve been extremely happy with. Einstein Attribution allows you to move away from manual, legacy, rules-based attribution. You can even forget about a primary campaign source model (first touch, last touch, even touch) and have a data-driven, almost completely hands-off model.
It also identifies the big question that existed with attribution and, in my opinion, solves it: dependence on opportunity contact roles. Call me weird, but in my company, if I ask a sales person to create a contact role for an opportunity, they’re gonna look bad at me and they’re most likely not gonna do it.
So, if I try to attribute marketing spend to sales with legacy contact roles, I get horrible results, nothing. Salesforce put in place some heuristic algorithms that figure out who should have been an opportunity contact role in this opportunity and it creates it in the background. That’s a virtual contact role. Then what they do is utilize Shapley values to correctly split the influence of one opportunity across all the campaigns that influenced it, which is great.
This is modeled as a cooperative game where the outcome, the size of the opportunity, is dependent on the input of each of the comparative players, each of the campaigns. That Shapley value is gonna allow you to see how big a share of the pie this campaign was responsible for.
Brilliant and it’s 100% off-the-shelf. For us, it basically took one dropdown to activate. I’m not gonna go very, very deep into it. If you wanna learn more, Lucy Mazalon has a fantastic article about it in her blog which I encourage you to go check out.
And it was just by pure chance that when we were developing our model, Salesforce was already thinking and planning to produce something like this. So, we fed them what we had learned and that led to some of the best, most profound discussions about marketing that I’ve had in my career and it ended up with Salesforce building something bigger, better, and completely hands-off compared to what we had and I’m very happy to see that our work also helps other people have this available.
So, how does this look in practice and how do you make practical decisions with it? We can see for any campaign, which opportunities has this campaign influenced? And not only which opportunities, but how much of the amount of the opportunity is attributable to the fact that one or many people in the opportunity participated or were touched by this campaign. And it’s very, very granular. The level of detail such as contribution per campaign (to the Euro-cents) is fantastic. It’s something that we couldn’t have ever fathomed doing on our own.
The one minus that I find is that this assumes that 100% of an opportunity’s value will be apportioned to marketing campaigns. This may be true for some faster-moving companies, like software service companies where you’re very driven by marketing, but in a company that is 187 years old like ours that deals with sales cycles that are two- to five-year €100M projects, it’s definitely not true. Not all of our revenue is created or sourced from marketing. A small part of it is and I need to know exactly how big that part is. So yes, I know how much each of the campaigns contributed, but then I need to know as a whole how much marketing contributed.
So, we went about to build our own deeper attribution model that would allow us to get to that real attribution.
Watch the session video to learn how Wartsila created this complex attribution model.
