CRM personalization usually runs on what you already know about a customer. Their last purchase, their favorite categories, their RFM score (Recency, Frequency, and Monetary Value). The approach works. But it hits one plain limit: it only works for customers you have recent data on.
And in some industries, only 5% of a customer base is active in a given month. For the other 95%, the data is too old, or simply doesn't exist. That's exactly where predictive AI comes in, and it's what this article is about.
Batch just launched Batch AI Predict, a library of 15 predictive scores available natively inside the platform. Let's break down what a predictive score is, how it differs from classic personalization, and how these scores turn into campaigns that drive revenue.
Rules-based personalization: useful, but quick to hit a wall
The first way to personalize a campaign is to write rules from past data. If a customer bought in the Photo category less than 3 months ago, send them the Photo offer. If they haven't opened in 60 days, move them into a win-back sequence.
This rules-based approach has two strengths. It's readable, and it's under control. But it rests entirely on one condition: having recent, reliable data for each customer.
That's where it breaks. When a customer hasn't interacted in a long time, the rule has nothing left to stand on. When the target behavior has never happened, say a first-time buyer you want to push toward a second purchase, the data doesn't exist at all. In both cases, rules-based personalization has no answer. And with 5% of customers active in a month in some industries, that's not an edge case. That's most of the base.
The predictive score: inventing the data that's missing
When the data doesn't exist, you have to invent it. That's exactly what a predictive score does.
The principle: look at customers who already showed the target behavior. Those who churned, those who made a second purchase, those who became high-value customers. A Machine Learning or Deep Learning model identifies what they have in common, then estimates how likely other customers are to do the same.
The output is a score: the probability a customer churns, the likely date of their next purchase, the discount level that maximizes their margin. Where rules-based personalization waits for the customer to act before reacting, predictive AI anticipates before it happens.
It's a shift in method. You no longer describe a customer's past, you estimate their future, including for the 95% with no recent history.
A score is a step, not an end goal
A predictive score is the product of serious work: algorithmic models trained on large volumes of data. It's a high-value raw material.
But a raw material is still a raw material, and a score left sitting in a customer profile changes nothing about CRM results. It produces value only once it's wired into a marketing action: a campaign, a sequence, a journey. The score tells you what to do. The activation does it.
That's why Batch AI Predict doesn't present its 15 scores as a technical catalog, but through the lens of their use cases. The useful question for a CRM Manager isn't "which scores exist" but "what can I do with them." And the clearest way to answer that is to sort the scores by how they're used.
Three uses: target, orchestrate, personalize
The 15 scores split across three broad operational uses. Three questions every CRM Manager asks when building a campaign.
Targeting: the WHO. Who to contact. These scores flag high-potential customers, at-risk ones, and those most likely to engage with a given product theme. They focus CRM effort on the right targets, and avoid hitting customers the campaign makes no sense for.
Orchestration: the WHEN. When to contact them. These scores anchor a message trigger in the customer's actual lifecycle: likely second-purchase date, repurchase date, churn date. They turn a calendar-driven plan into orchestration driven by behavior.
Content: the WHAT. What to send. These scores determine the product to recommend, the discount level to apply, the channel to favor. They tailor the message beyond the segment.
Who I target, when I reach them, what I send. The library follows that exact reasoning. The 15 scores and their use cases will be detailed one by one in the next articles of this series.
Business value comes from the use case
Having the richest possible customer profile, packed with predictive attributes, is not a business goal, it's a means. The goal is to ship as many high-impact marketing sequences and campaigns as possible, all built on these scores.
A use case ties together three elements: a score, an activation process, and a test to measure impact.
Take the churn date. The score estimates a customer is likely to go inactive on a given date. The use case triggers a retention sequence a few weeks before that date, with a fitting offer, then compares that sequence to the usual win-back flow in an A/B test. That wiring is what turns a probability into measurable revenue.
A/B testing is central. A score's ROI always depends on the use case and the brand's context. You don't declare it, you measure it. That's also what makes the approach reassuring: you test on a controlled scope, you compare to what you already do, and you scale only what works.
Worth knowing: the measured results from Batch customers, on conversion, margin or reactivation, will be detailed in every use case article of this series. Each case will cover the pain point addressed, the score used, the process set up and the test attached to it.
A predictive layer built to be easy to switch on
Three things make Batch AI Predict accessible, even for teams that have never worked with predictive AI.
It's plug and play. The scores are computed by Batch from data already in the platform. No data infrastructure to build, no model to train yourself.
It's simple to use. The scores live right inside the Batch platform, in the customer profile, next to the attributes teams handle every day. No new tool, no new interface.
It's fast to test. Count 2 to 4 weeks between data scoping and your first live campaign: data mapping and score selection, model configuration, scores pushed onto profiles, then first campaigns. ROI validation happens from the first tests.
For teams that want to go further, Batch runs qualification calls. The goal: pin down the exact need, then prioritize use cases based on data availability, internal resources and the expected impact of each case. A roadmap built on the brand's real context.
The use cases in the Batch AI Predict series
Here are the first use cases identified, sorted by the three uses presented above. This list isn't exhaustive: it will grow as articles are published.
Target: optimize the campaign's WHO
Maximize the performance of a themed campaign: automatically build an optimized audience for a range or theme, instead of a slow, approximate manual target;
Reduce marketing pressure on themed campaigns: narrow the audience to genuinely interested customers to protect engagement and deliverability;
Anticipate the slowdown of your best customers: detect a top customer drifting away before it's done, to trigger a targeted sales action;
Identify the right customers for exclusive perks: prioritize VIP perks on customers with high future value, not just past revenue;
Maximize recruitment on a specific product range: identify high-potential customers for a premium or iconic range without pushing it to the whole base.
Orchestrate: optimize the campaign's WHEN
Trigger your retention campaigns at the right moment: activate retention at the first weak signals of churn, before full disengagement;
Identify the optimal repurchase moment for each customer: trigger the message exactly when the customer is ready to rebuy a recurring-consumption product;
Anticipate unsubscription before renewal: spot at-risk subscribers early enough to act before their decision.
Personalize: optimize the campaign's WHAT
Personalize the product content of your campaigns: dynamically feed product blocks so each customer sees a selection of their own;
Optimize the discount level per customer: assign each customer the promotional mechanic that maximizes both conversion and margin.
Batch AI Predict, one layer of a bigger system
Batch AI Predict is one of three AI layers in the Batch platform.
Batch AI Assist brings together 12 AI agents that augment teams in their daily work: campaign analysis, content generation, journey optimization.
Batch AI Decide runs an autonomous CRM, where agents decide, launch and optimize campaigns against the goals you set. In between, Batch AI Predict supplies the intelligence on future customer behavior.
That predictive intelligence feeds both the augmented work of teams and automated decisions. It's the layer that takes a CRM from reacting to anticipating.
Predictive AI is no longer reserved for organizations with in-house data teams. By building it natively into its platform, Batch puts it within reach of CRM teams that want to decide better, without changing how they work.
Andréa Massimi
Content Marketing Manager @ Batch
Follow us
The CRM Newsletter
Subscribe to get the latest news in your inbox!
