Introducing The Prediction Era for Digital Ads

What if you sent your click data into your own algorithm? Lydia Fern, VP of Product, introduces a new paradigm for digital ads.

Recorded at Masterclassing's Retail and eCommerce Masterclass in Los Angeles on January 28, 2026.

To read the talk, see below:

Lydia Fern, VP of Product at Bonsai

Bonsai is a first-party marketing platform that automates profitable growth. I'm extremely excited to be here with you all today, especially after hearing some of the reasons that you decided to come to this conference. Questions like:

• "How are retailers dealing with a crowded ad marketplace?"
• "How can I attract new consumers across both retail and e-commerce?"
• "If I wanted to double conversions without doubling spend, where would I start?"
• "Where is the one place in the customer journey where most retailers are not currently using data but they really should be?"

These are just a few of the very important and nuanced questions that you all brought today. I'm excited to dig in more at our roundtables later and I hope to start to address some of these right now in my talk. 

Before I jump into nuance and complexity, I want to start by going back to the basics: the few pillars of modern digital advertising that I'm sure many of you who work in marketing are very familiar with. Digital advertising in the modern era is bought and sold through an auction, and that auction is automated through very well-known products like Google Smart Bidding and Facebook Bid Strategies. These products require a feed of training data to make them work. Most commonly, that training data is conversions which are captured and tracked through a pixel on your website. It's well established that the more training data that you give these products, the better the outcome. And the more recent or latent this data is, the better.

This is the system that powers digital advertising in the modern era. It's worked incredibly well for many years. But lately, it is faltering. Some of you might be feeling the effects of this. You're spending more than ever on ads. Your platforms are telling you that you're winning on conversions, and yet your CEO and CFO are asking you, "Why is our business down year-over-year and what can marketing do about it?" There is a problem with the system. It's just not growing our businesses the way that it used to.

I would argue the issue is not with the system itself. It's with that training data that we are giving it.

There are three critical problems with this data, with conversions, which are represented by these beautiful green dots on the screen.

1. Conversion data represents a privacy violation.

Every piece of tracked conversion data represents information about your customers that you're sharing with a third party. Interestingly enough, the effectiveness of your ad buying is actually not negatively impacted by this privacy violation. If it were, then we might actually see Google make good on their threat to eliminate third-party cookies altogether.

2. Conversion data is an inaccurate prediction of incrementality.

It's well understood that many people click on ads after they've already decided to purchase. What this means is that all those conversions that you saw yesterday, some portion of them were not actually driven by your ads. The data itself is inaccurate. As marketers, sometimes it's easy for us to forget that when we put out digital marketing content, we're not actually buying conversions, we're buying clicks. A vast majority of people who click on your ads do not convert.

3. The data is sparse.

If your website has a standard conversion rate of maybe 1%, that means that for every one conversion that is influencing your ad buying, there are 99 clicks that could be valuable pieces of information but are being completely ignored by your bidding.

Just to go back to point number two for a second, the inaccuracy or the lack of purity of this conversion data can actually negatively impact your business outcomes. In a case where conversions are more non-incremental than incremental, you might be better off having no bid management at all.

These are the three problems that we have with the data powering our digital advertising. What is being done about it? There are a few solutions out on the market available on ad platforms. I will argue that these solutions are faulty fixes.

• Using attribution models to address the data sparsity issue. You may have implemented this using Google's Data-Driven Attribution (DDA) or perhaps by increasing your attribution windows in Meta ads. These strategies do improve data sparsity, but just not enough to make a difference. Even if by using DDA you can improve your data volume by 5x, you're still leaving 95% of clicks on the table unused.
• Using first-party business data to enhance conversions. You may have done this using Google's Enhanced Conversions or Meta's Advanced Matching. These strategies can be a good way to identify and then remove those non-incremental conversions. But then we're back to square one with data sparsity.

These solutions are not enough. What we need is a new paradigm. Instead of ad platforms training on a few perfect conversion signals, they should train on many good enough click value signals.

Imagine a world in which you took not just yesterday's conversions but the last 12 months of conversion data, applied an attribution model, and enhanced it with your first-party business data. Instead of sending that directly back into Google Ads or Meta ads, you send it into your own algorithm—an algorithm that can then predict the incremental business value of every single click that came to your site yesterday, not just the ones that immediately led to a conversion. In this paradigm, every click to your site will be assigned a predicted click value and then sent back into your ad platform the very next day to influence bidding. Data sparsity is solved for and data purity is good enough.

Does this predictive click value new paradigm actually work? We have two findings.

One is a simulated finding and one is a real-world finding that shows that yes, this does improve business outcomes.

First, the simulation. The details of this simulation will be available in a white paper as a free download after this talk. In essence, the simulation seeks to compare three different training data strategies: traditional conversions, enhanced conversions, and then this new paradigm, predicted click value. It seeks to measure the impact on incremental business outcomes over a period of four weeks. Here's what we found: After 27 days, enhanced conversions did drive about a 7% incremental lift on business outcomes over the traditional conversions. Improving the purity of your conversion data does drive lift. But if you compare that to a predicted click value strategy where every click is scored and sent in, we saw that after the same 27 days, that strategy drove even more: 18% lift over the enhanced conversion strategy.

It also works in the real world. There is a 2025 Algorithms Report Card that's available for download via the QR code on the screen. We found that businesses who used a predicted click value training data strategy beat their category growth by 40x in 2025. Real companies who used a traditional conversion strategy lagged behind, only achieving about 60% of their categories' growth.

Two final thoughts to leave you with.

Number one, this strategy of predicted click value doesn't just work for Google ads. This generalizes to any and all click-based second-price auction ad platforms: Meta ads, Microsoft ads, TikTok ads. There will be many more coming down the road.

Number two, this strategy will grow your business while still protecting consumer privacy. There is zero consumer data being shared with any third-party platform and you could run your ads without having a single third-party cookie, pixel, or floodlight on your website at all.

That concludes my talk. Thank you.

‍