Modern ad platforms like Google and Meta use signals to decide which users within their overall audience to to show your ads. They do this by using a continuous feedback loop where the network tries to show your ads to a subset of users, wait for results to come in, and then figure out which other subset of the users to target next based on the results.

In other words, they use the success signals you provide them (i.e. events you send them) to determine how to best bid on the next set of impressions and clicks.

Picking the best success signal is one of the major levers in advertising these days. We have examples of CAC reducing by 50% by picking a better event to optimize towards.
‍

➿ Marketers as Evaluation Providers

Marketing is becoming a field of providing the best signals for AI to optimize towards, and you’ll want to build a good process for sending them signals.

Use the following best practices to choose the best event for paid social and paid search:

⏲️ Use an event that happens within 1-3 days of click.

These machine learning models are constantly learning and adjusting their targeting based on the feedback they receive. Think of them as incredibly fast learners, but they require clear, consistent signals. If you feed them fuzzy or delayed data, they'll optimize for the wrong things, burning through your budget on irrelevant users. Ideally, pick an event that happens within the first 1-3 days.
‍

📈 Use an event which correlates with high-value users.

It's not enough to just optimize towards any early action. A free download or a simple page view might be easy to get, but do they lead to paying customers with high lifetime value (LTV)?

Your event must be a strong predictor of future revenue. We recommend a correlation of at least 0.6 or higher.

5️⃣ 0️⃣ Send an event that fires at least 50 times / week.

You need an event that happens frequently enough to give the AI sufficient data to learn from, but also one that is a strong indicator of a high-value conversions.

For instance, if you're selling a SaaS product, optimizing for “MQL demo request“ might be better than “demo “request because it's further down the funnel and indicates higher intent, even if the volume is lower. The goal is to find the earliest, highest-volume event that reliably predicts a profitable customer, allowing the AI to scale acquisition effectively.

However, if the signal volume is low (< 50 events per week) then the AI may not be able to accurately fine-tune your campaigns.

❓ FAQs

What if I don’t have enough volume on my high correlating events?

If you’re early in your ad-spend journey and don’t have enough event volume from lower-funnel events then you’ll have to get creative.

The most common method is to add questions to your sign-up flow and qualify the users coming from paid traffic before you let them sign-up. You can ask them a set of questions that helps you qualify that they are well wihtin your ideal customer profile, and that your product / service is going to valuable towards.

Use the answers of these questions to send events to ad networks when people answer the questions in the right combination. Use this event for optimization.

Does event optimization work in B2B?

Yes! We’ve had a lot of success on Meta and Google using event optimization as a lever to reduce CAC and bring on higher value pipeline.
Most common event in B2B is a qualified demo request, sign up, or lead. However, the trick is to have a solid MQL definition. Use tools like Clearbit to enrich leads as they come and send events to the ad networks when you have conviction that the lead falls within MQL.

Some ICP examples:

• People who work at X, Y, and Z industries with 200-1000 employees which have physical locations in NY and California
• Companies which have at least $100k in their bank account (ask them in a quiz!) in the U.S.

What if I have power laws in my revenue from users?

If you’re in gaming, social networks, or other business models where a small percentage of your user base is going to account for a large portion of your revenue (i.e. “whales”) then event optimization becomes much more complicated.
The main heuristic we’ve seen work here is to build your own ML models to predict the likelihood of any user becoming a “whale”. If your internal model is predicting that a user is likely to become a whale, then send an event to ad networks for optimization.

‍