Conversion Engineering (aka AI-Powered Attribution)

Welcome to the 6th edition of my Marketing Mix Newsletter! Today, we'll talk about Conversion Engineering, which I started to discuss in my previous newsletter.

In my earlier discussions on Marketing Attribution, I explained why it no longer works effectively for analytics or for optimizing smart-bidding ad platforms. If you missed these articles, I highly encourage you to read them before continuing. Especially the article about Customer Journeys and why it's impossible to track them.

There are two important phases for each business selling online, both of which are very hard to track today using legacy attribution approaches:

1. Research Phase: Everything that happens before acquiring a lead or conversion.
2. Sales Cycle Phase: Everything that happens after acquiring the lead (this is mainly relevant for lead-generation businesses with a longer offline component in CRM, which we will cover in the next newsletter).

Research Phase

Let's start with the Research Phase.

The biggest problem in the Research Phase is that most of the clicks and ad impressions you pay for don't end up with a conversion, even though they are quite valuable for the business and contribute significantly, but it's almost impossible to track this contribution.

Here is a simple example: someone clicks on your Facebook Ad on mobile, opens your website in an in-app browser, browses the website but doesn't buy anything. Later, the same visitor might open your website from different browsers and devices (usually directly) and finally make a purchase.

Problems:

• Facebook Platform Credit: Facebook doesn't receive credit, and smart-bidding doesn't learn.
• Analytics Credit: Analytics show that the purchase actually comes from direct/none, while the Facebook Ad was a primary contributor to the purchase.

Broken Attribution

For many businesses, this is a huge problem because upper-funnel campaigns don't receive any credit, and most of the credit goes either to <direct/none>, <organic/search>, or brand campaigns.

Here's how marketers can fix both analytics and optimization using Conversion Engineering. But how can you do this if it's not possible to stitch the customer journey between multiple devices?

The Key Shift:

Conversion attribution doesn't matter as much as the contribution of each visit. Marketers need to find leading indicators within each visit that help identify valuable (incremental) visits and immediately assign value to them.

In the past, marketers already used such an approach. For example, you can check whether there were some valuable micro-conversions during the visit, like:

• Spending some time on the website (e.g., >3 minutes)
• Viewing a product page and interacting with it
• Adding a product to the cart

Example:

Imagine you have an e-commerce website, and every month you have:

• 1,000 Purchase events (total value of $2M)
• 3,000 Add to Cart events
• 15,000 Viewed Product Page events

Let's try to understand the approximate value of Add to Cart and Viewed Product Page events in this particular case:

• Purchase: $2M/1000 = $2000 (average)
• Add to Cart: $2M/3000 = $666
• Viewed Product Page: $2M/15,000 = $133

This way, when we look at the same exact journey from the perspective of leading indicators (contribution), our attribution changes dramatically:

You might say that this will add a bias because there was only one purchase for $2000, while in total we attributed $1331 + $1598 + $2799 = $5728!

This is easy to fix using a normalization technique.

In this particular case, we've tracked a value of $5728 instead of $2000, which is 2.864 times more than was actually received. All we need to do is normalize each visit value by the coefficient of 2.864:

Normalized Attribution:

Before:

• Facebook Ads: $0
• <direct/none>: $2000

Now:

• Facebook Ads: $465
• <direct/none>: $1535

Now we can clearly see that Facebook Ads received the credit it deserved, and a conversion with a value is being sent back to the Facebook Ads platform so that smart-bidding algorithms can see that the click brought value to the business, and Facebook Ads will try to find more clients like this.

This is a very simplified example of Conversion Engineering. In reality, it can be much more complex and sophisticated when using the power of AI!

Advanced Techniques

For example, at SegmentStream we score each website visit using a very complex AI model that takes into consideration hundreds of behvioural and contextual events to define the actual contribution of the visit.

Engineering conversions for each visit today is the most accurate approach for marketing measurement and ad optimization. It helps overcome the problem of complex customer journeys and the evaluation and optimization of upper-funnel campaigns, which generally don't receive the value they deserve.

While in reality SegmentStream uses Conversion Engineering under the hood, we rarely use this term to avoid confusing conservative marketers. Instead, we call it AI-Powered Attribution. For those who are more technical and want to read some tech-savvy material about this, I encourage you to read this article in our documentation.

Conclusion

When you modify your attribution so that it is based on Conversion Engineering and reflects the real contribution of each visit, you can:

• Better reflect the value of each channel in your marketing reports
• Properly optimize ad platforms by sending more accurate feedback signals to their ML (Smart-Bidding), which I covered in my previous newsletter .

I understand this material might be difficult to comprehend, so feel free to ask your questions in the comments! Once this technique is mastered, your Marketing Mix could be hugely optimized, potentially bringing up to 50% additional revenue with the same budget!

P.S. As usual, if you find this newsletter useful, please like and share it. Your support helps keep this newsletter alive and continue delivering value to our community!