Propensity scoring: unlock +20% AOV on thematic campaigns

Building the audience for a themed campaign takes a serious amount of time. And too often, despite all that work, engagement still falls short. If that sounds familiar, this use case is for you.

It's one of the most common use cases for predictive AI in CRM, and the first in a series that breaks down, one by one, the concrete uses of Batch AI Predict. To understand why a predictive score only creates value once it's activated in a campaign, start with our framing article.

The problem: lots of time, little engagement

The scenario is familiar. A new collection drops, a seasonal push is coming, a product line needs promoting. You need to build the audience. So you dig into purchase history, cross-reference categories, stack rules:

• 'bought in this category in the last twelve months',

• 'clicked on that campaign',

• 'belongs to this segment'.

The targeting eventually comes together, after hours of work.

And the results disappoint. Engagement doesn't match the effort. That's expected: rule-based targeting is built on what customers have already done, not what they're about to do. It misses customers who've never bought the theme but would respond to it, and it includes customers who bought once without any real appetite.

Before going further, let's ask the right question: how do you know the problem is solved? The real success metric here is campaign conversion rate as a proof that you're reaching the right customers. Close behind: average basket and additional revenue, which measure business impact, and targeting time, because half the problem is exactly how long all of this takes. We'll come back to the numbers at the end.

The context: a packed content calendar and approximate targeting

Picture a brand with a broad catalog, running its base at the pace of a heavy commercial calendar: several themed campaigns a week, product lines to push, seasonal peaks back to back. Each campaign needs its own targeting, built by hand by the CRM team from purchase history.

Two problems stack up. Manual targeting is slow, which caps how many campaigns you can run. And it stays approximate, because history doesn't tell you who wants to buy this theme right now. The result: high effort for average return, on a volume of campaigns you'd want to scale up.

The process in Batch: building the audience from the propensity score

With Batch AI Predict, themed campaign targeting is built directly inside the platform, without third-party tool or export. The process runs in three steps.

1. First, identify the target theme: a product, a category, a line, a family.

2. Then, the propensity score for that theme is available on each customer profile, like any standard attribute.

3. Finally, build the audience by selecting customers whose score exceeds a given threshold (say, the top most-likely customers) exactly as you'd filter on any CRM attribute.

Where manual targeting took hours, the selection takes minutes. And above all, it no longer rests on an approximation: it ranks customers by their real probability of engaging with the theme. The CRM team spends less time building, and launches the campaign on a far more relevant audience.

The score behind the use case: product propensity

No black box. The score powering this use case is the product propensity score. Its principle is simple to state: for each customer, it predicts the probability they'll buy within a given product target over the next X days.

That probability, between 0 and 1, is computed by a Machine Learning model trained on all the data available about customers:

• behavioral data (purchases, browsing, engagement — opens, clicks, unsubscribes),

• product catalog,

• socio-demographic data, and more.

The model looks at who bought the theme in the past, identifies the signals that characterize those buyers, and applies that reasoning across all customers to estimate their propensity, including those who never bought the theme but show the right signals. That's precisely what rule-based targeting can't do.

Worth noting: this same propensity score also powers another use case. Here we use it to maximize campaign performance; you can also use it to preserve customer engagement by narrowing the audience to relevant customers only, reducing marketing pressure. One score, several use cases: that's exactly the logic of Batch AI Predict.

The performance: what we observe on this type of use case

Here are the performance levels you can expect on this type of use case, measured via A/B test against a target built by CRM experts.

• The conversion rate is multiplied by roughly 1.5, typically rising from around 0.9% to 1.4%.

• Average basket grows by about 20%.

• Put together, and in the context of a specific client whose send volume allowed it, that amounted to additional revenue on the order of €46,000 per tested campaign, a figure to read against the volume of each campaign.

• And on the operational side, the time spent targeting each campaign drops by about 70%.

These numbers say two things, answering the starting problem point by point. Conversion and additional revenue prove you're finally reaching the right customers: engagement is no longer disappointing. The slashed targeting time solves the other half: the CRM team can launch more campaigns, test more, without spending all day on it.

In short

Targeting themed campaigns is one of the most time-consuming jobs in CRM, for a result often below the effort. Batch AI Predict's product propensity score builds the optimal audience in minutes, right inside the platform, and reaches the customers who are genuinely receptive. More conversion, more basket, far less time spent.