ru EN DE
Запланируйте демо
Exponea стала частью Bloomreach. Узнать больше

How To Use Predictive Audiences to Target Better for Increased ROAS and Higher ROI

Capabilities used: Exponea Core, Omni‑Channel Orchestration, Real‑Time Customer Predictions

Setup time: Novice 60 min / Advanced user 25 min

Reading time: 15 min


Brands have successfully used remarketing campaigns to target customers based on their online browsing patterns. While this has contributed to improved ROIs, the limitation is that it only works if a customer has already taken a single, specific action (like adding a product to cart). Exponea’s Predictive Audiences on the other hand, rely on multiple sources of information, in real-time, to predict future intent and the likelihood of a customer taking an action. This allows marketers to proactively anticipate a customer’s needs and target the right people at just the right time leading to improved reach, increased ROAS, and lower advertising costs.


This guide will help you understand:

  1. 1.Who should care about this use case?
  2. 2.Challenges that this use case solves
  3. 3.Direct results and business impact of this use case
  4. 4.How does it work?
  5. 5.How to easily deploy a pre-built prediction model
  6. 6.Real-world example
  7. 7.Results
Part 1

Who should care about this use case?

Ally oversees the in-house performance marketing team at a mid-sized company that runs paid media across several channels like Facebook, Instagram, Google, and Adform. The team has a monthly variable spend budget which is allocated across campaigns running on these channels based on the team’s analysis of past purchases, browsing behavior and buyer personas.

Part 2

Challenges that this use case solves

Performance marketers like Ally have several channels available to acquire and retarget customers. However,

These channels are expensive and there’s little that can be done to lower costs

All customers who take actions on the website are retargeted irrespective of intent

Customers react negatively to ads that are not relevant to them

It is hard to predict if a channel or campaign is likely to perform, especially for products with a long lifecycle

Part 3

Direct results and business impact of this use case

Increased ROAS

Predict which customer segments are most likely to convert so you’re spending your advertising budget wisely.

Lower CPAs

Target only high-intent prospects who are currently in the market for your product. This also keeps your ads relevant and customers happy.

Lower costs

Quickly determine the likelihood of success of a campaign with a certain audience to help change course and save on resources.

Higher ROI

In addition to granular targeting and cost optimization, predictive audiences help realize greater returns from your marketing investments by closing the gap on lost opportunities. Anticipate customer behavior and proactively target instead of reacting to past data to maximise on your investments.

Actual numbers realized by a real-world implementation of these features by Exponea clients:

-38% CPA

+150% ROI

+63% purchases

Part 4

How does it work?

The above results can be achieved using Exponea’s AI-based Predictions module. We predict future customer behavior, intent, and likelihood of action to a high degree of certainty based on data collected from multiple sources and in real-time.

Real-time computation means that once set, your data is automatically kept up to date and your advertising content is ready to target micro-moments, which Google defines as intent-rich moments when a customer goes from awareness to conversion reflexively. Your customers are not always thinking about your product, but it is vital to be front and center when they are thinking of you. Real-time also means that it is continually recalculated and updated as the customer interacts more with your website. For example, the probability of purchase will change and be updated for a customer as they make more purchases on your website.

Exponea offers several pre-built models as well as the option to create custom prediction models.

Compared to most other CDPs in the market, Exponea’s pre-built models are easy plug-n-play solutions that put power in the hands of the marketer. Marketers can easily deploy predictions themselves without the involvement of engineers or data scientists.

Part 5

How to easily deploy a pre-built prediction model

Consider a team like Ally’s that wants to get more efficient with their ad spend. Instead of retargeting broadly, they need to identify and spend on customer segments that are more likely to convert. Using Exponea Predictions, they can calculate the probability of purchase for every customer. The team can then create separate customer segments based on their likelihood of purchase and spend accordingly.

The team can easily set up the pre-built Purchase Prediction model which calculates a purchase probability value for each customer.

This value is stored as an attribute in the customer profile which can be pulled into any campaign or marketing activation on Exponea.

Then, a Segmentation classifying customers into likely, moderately likely, and highly likely to purchase is set up. These customers can then be retargeted on Facebook or Google or any other advertising channel.


Armed with this information, the team can now decide how and on who they want to invest for maximum ROI.

This is one of the easiest ways to predict if a customer is likely to convert or not. Exponea has more pre-built templates and even a custom prediction model.

For example: instead of just determining whether someone is likely to convert or not, say you want to determine the likelihood of someone becoming a high-value loyalist over time and invest more in this segment. This can be done by building a custom prediction where you start by defining what a high-value loyalist looks like to you (eg: someone who has made at least 4 purchases in the last 90 days) and then define lookback window for data so the model can learn, filter by customer or event attributes and then run the prediction which can then be used as explained above.

Part 6

Real-world example

Online retailer more than doubles ROAS and decreases CPA by using Exponea’s Predictive Audiences for Facebook prospecting

Business need

Estonia’s largest online retailer wanted to reach new customers and maximize revenue. The company also wanted to evaluate the efficiency of the investment.

Marketing need

The marketing team was using Facebook Lookalike Audiences to prospect for new customers. Facebook Lookalike Audiences works by analyzing a seed audience — typically made up of people who have made purchases on a site — and creating a new, unreached audience of Facebook users. This audience can then be targeted with ads on Facebook. However, as powerful as Lookalike Audiences are, they are based on singular actions taken in the past.

The marketing team wanted to use Exponea’s AI-driven Predictions to build an audience with forward-looking data. They wanted to create an audience likely to purchase within the next 7 days, and use that audience as a seed for Facebook’s Lookalike Audiences.

Marketing challenge:

The marketing team will now have to create a new email every week and source new content for each send, as well as test each new email. This is time consuming and repetitive work. In addition to this, the team now has to get creative and find ways to personalize the content. By broadly classifying their audience and rule-setting they can personalize content to some degree. But this means an additional workload to try and create different versions of the same newsletter every week.

How Exponea can help

Using Exponea Predictions, the company created a purchase prediction model to determine which customers were likely to purchase within the next 7 days. Then, they created a Scenario with an A/B test to evaluate the results. 20% of the customers were kept as a control group and used to create a standard lookalike audience on Facebook. The rest 80% were checked for their probability of purchase, estimated by Exponea’s Purchase Prediction, and if the probability was greater than 10%, they were imported to Facebook to build Lookalike Audiences.

Part 7


By adding an AI-edge to the already powerful Lookalike Audiences tool, the company was able to see significantly improved results. By using an audience that had a high probability of purchase in the future instead of relying on a single source of past purchase data, the company doubled revenue and ROAS and saw a 62% increase in conversions as well as a 38% reduction in CPA. Setting up a Scenario and A/B testing within Exponea made these results easy to evaluate as well.

To take advantage of any variant of this use case in your company, please reach out to us at

Мы используем cookies,

чтобы сделать нашу коммуникацию удобной для вас. Нажав на кнопку "Принять", вы соглашаетесь с нашей Политикой конфиденциальности. Вы также можете управлять настройками, нажав на "Управлять cookies".

Управлять cookies

Настройки cookies

We use cookies to optimize our communication and to enhance your customer experience. We also share information about how you use our website with our third parties including social plugins and analytics. You consent to our use of cookies if you continue to browse our website. You can opt out of our cookie use on the Do not Sell my Personal Information page. For more information please see our Privacy Policy.