Training the Meta AI for High Quality Leads

I’ve been obsessed with the idea of using B2C marketing strategies at B2B SaaS companies since I worked for Intercom four years ago.

I met Guillaume Cabane, VP of Growth at Drift last year. Turns out he had the same “crazy” idea.

Just a few months later, we figured out how to hack the Facebook B2C pixel to optimize towards the highest quality users. Drift now has a better return on investment from Facebook prospecting ads than Google search.

This is a game changer for B2B SaaS companies. This is the future, and the future is AI.

Some background info

If you don’t already know, the Facebook pixel is a piece of code that you place on your website to collect important data that helps you track conversions from Facebook ads. The pixel helps you optimize ads, build targeted audiences for ads, and retarget people who have already visited your website. It’s often considered the gold standard of paid acquisition due to its powerful targeting AI.

For example, retailers feed transactional data into Facebook’s AI to train its bidding engine and then, Facebook optimizes bidding for consumers who are likely to buy from that retailer. The moral of the story is simple — you should never be bidding more on a lead than what they are worth for your business. The pixel’s capabilities are pretty in-depth and have been assisting companies for over a decade, however there are still some limitations.

Due to a few limitations, SaaS companies and other B2B businesses sometimes have a hard time fully leveraging Facebook ads. Facebook only keeps data from the past 28 days, which means purchase data from sales cycles that take longer than 28 days can’t be fed back into Facebook’s AI. This is typically a roadblock for many SaaS companies. Facebook’s AI learns faster when events happen sooner, so there is a huge incentive for SaaS companies to optimize towards an event higher in the funnel.

AI on AI on AI

As mentioned above, SaaS companies are running into limitations when bidding on Facebook ads. This is because they typically bid low on a large audience because they’re unable to properly identify and optimize for high-quality leads.

However, fast-growing SaaS companies such as Drift are heading in a different direction with their ad strategy. Instead of just bidding on ads with no real gameplan, they’re feeding MadKudu data into Facebook’s AI. This is helping them optimize bidding against leads which MadKudu would score high. In brief, Madkudu’s AI is now training Facebook’s AI for better results as the end-goal. And with Lightning AI — we’re layering another AI on top of Facebook. This is the future; computer systems talking to each other to make the best optimizations possible.

How to translate lead score data for Facebook

The main goal in mind when using Madkudu’s AI is simple. The goal is to feed transactional data to Facebook that it can use to optimize bidding against leads that we want, because MadKudu’s predictive score identifies a lead’s value at the top of the funnel. During this process, we just need to capture the lead data as early as possible and send it to Facebook in a way they understand.

Facebook has two main attributes it looks for to train its AI: an individual and its “value.” For example, in eCommerce world this would mean feeding a purchase back to Facebook, however we still need to change our value a bit. MadKudu is excellent at predicting the amount that a lead will spend based on historical deal data, which helps us differentiate between self-serve and enterprise leads. However, no matter what, not all the leads obtained will convert.

The formula below can be used as a reference when determining if a lead will convert:

Lead Value = % likelihood to convert x Predicted Spend

So, if a lead has a 10% chance to convert to $30,000 in ARR, we can send Facebook a “transaction” worth $3000 as soon as the lead gets generated. Now we can send data to Facebook nearly immediately to train its model. So in turn, we’re actually training Facebook’s AI to value the same type of leads that we value internally using MadKudu.

MadKudu’s FastLane is the easiest way to capture lead data. FastLane is a basic line of javascript that turns any lead form into a customer fit-driven lead capture device. The same mechanism that helps Drift convert more leads into demo calls is training Facebook’s AI.

The overall impact

Throughout this process, Drift (while working with Lightning AI) experienced a 300% increase on conversion from Facebook spend and was therefore able to hit a larger audience for cheaper. The impact was nearly instantaneous. And now Drift is only enabling Facebook to spend money on leads that Madkudu will score well — which Drift already knows to work well in predicting conversion to customers.

Drift is able to now extend MadKudu to its paid acquisition leveraging our API and our many integrations. And the best news of all, connecting the Facebook Pixel and Madkudu to Segment only takes a few minutes. After that’s completed, Drift can easily pipe MadKudu data to the Facebook Pixel in real-time.

‍