Emarsys AI Personalization: Turning Customer Data Into Revenue at Scale

Most personalisation claims in email marketing amount to this: you put the customer’s first name in the subject line, maybe show them products from a category they’ve previously browsed, and call it personalised. It’s not nothing — but it’s a long way from what AI-driven personalisation actually looks like.

Emarsys, now part of SAP, is one of the few platforms where the AI layer is genuinely integrated into the campaign and automation engine rather than bolted on as an optional feature. For enterprise retail and e-commerce brands running large databases and complex customer lifecycles, this difference in architecture has significant commercial implications.

This guide explains how Emarsys’s AI capabilities work, how to implement them in your email programme, and how to measure whether they’re delivering actual revenue lift.

Emarsys’s AI Architecture: What It Actually Does

Emarsys’s AI isn’t a single feature — it’s a set of predictive models that operate across the platform, continuously learning from behavioural and transactional data to make predictions at the individual customer level.

The models that matter most for email personalisation:

Predictive Recommendations

Emarsys’s recommendation engine analyses each customer’s purchase history, browse behaviour, and product interaction data to predict which products they’re most likely to buy next. These predictions power personalised product blocks inside email templates.

Unlike simpler recommendation engines that rely primarily on collaborative filtering (“customers who bought X also bought Y”), Emarsys’s model incorporates individual affinity scores by category, brand, price tier, and product attribute. A customer who consistently buys premium products from a specific brand gets recommendations that reflect that preference — not just what’s popular with the broadest audience.

The recommendations update continuously as new purchase and behavioural data arrives. An email sent on Monday may show different recommendations than an email to the same person sent on Thursday, if they browsed or purchased between those dates.

Send Time Optimisation (STO)

Emarsys’s Send Time Optimisation calculates the optimal send time for each individual customer based on their historical email open behaviour. Rather than sending a campaign to your entire list at 10am on Tuesday, Emarsys queues each recipient to receive the email at the time of day they’re most likely to open it — which varies significantly across customers.

The STO model learns from open timestamps across all campaigns the customer has received, identifying patterns like “this customer reliably opens emails between 7am and 8am on weekdays” or “this customer tends to engage with emails on Sunday evenings.” Each customer gets their own send window within your campaign’s delivery period.

In practice, STO typically lifts open rates by 10–20% compared to a fixed send time. The lift is larger for large, diverse customer bases (where the variation in optimal open times is greatest) and smaller for highly homogeneous audiences.

Churn Prediction

Emarsys builds an individual churn probability score for every customer, updated continuously based on their recent engagement and purchase behaviour. This score predicts the likelihood that a customer will not purchase again within a defined future window — typically the next 30 or 90 days.

Churn probability feeds directly into segment creation: you can build a segment of “customers with churn probability > 70% who have not purchased in the last 60 days” and trigger a retention campaign specifically for this group. The targeting is more precise than a simple “hasn’t purchased in X days” rule because the model accounts for that customer’s normal purchase cadence — a customer who typically buys twice a year with a long gap isn’t flagged as a churn risk, while a customer who usually buys monthly and has paused for two months is.

Next Best Product Algorithm

The Next Best Product (NBP) model predicts, for each customer, which specific product they’re most likely to purchase next. Unlike category-level recommendations, NBP operates at the individual SKU level.

NBP is particularly powerful for post-purchase email programmes. Rather than showing a generic “you might also like” block with algorithmically similar products, the email shows the one or two products the model predicts as highest probability next purchases for that specific customer. This typically generates 2–4x higher click-through rates on recommendation blocks compared to category-affinity recommendations.

Implementing AI-Driven Blocks in Email Templates

The Emarsys email editor supports personalisation blocks that connect directly to the AI models. Building these into your templates requires connecting the recommendation engine to your product catalogue — which happens via the product catalogue upload or API sync — and then inserting the recommendation widget into your template.

Product Recommendation Block Setup

In the Emarsys email editor, the Personalisation Token library includes a “Product Recommendations” widget. When inserting it, you configure:

• Recommendation type: Top picks for you, Recently viewed, Frequently bought together, Next best product, or Category bestsellers (as fallback for new customers with limited history)
• Number of products: Typically 3–6 products in a grid or stacked layout
• Fallback logic: What to show if the model has insufficient data for a given customer (new customers, or customers whose purchase history is too sparse). Bestsellers by category or overall is a sensible fallback.
• Product card design: Image size, price display, product name truncation, CTA button text

Once configured, the block renders differently for every recipient at send time. The email template is built once; the content inside the recommendation block is generated individually for each person when the email renders.

Combining AI Blocks With Static Content

The most effective use of AI personalisation in email is combining dynamic AI-driven blocks with manually crafted static content. An email campaign for a seasonal sale might have:

• A static hero banner with the sale headline and key offer (same for everyone)
• A dynamic product recommendation block showing each customer’s personalised picks from the sale (individual)
• A static footer with brand content and social links (same for everyone)

This hybrid approach balances creative control (the campaign message and aesthetic are consistent) with personalisation depth (the products each person sees are individually relevant). It’s also the most maintainable format — the static elements are designed and approved once, while the dynamic block handles itself.

Measuring AI Personalisation Impact vs Control Groups

The only way to know whether the AI personalisation is actually lifting revenue is to run a controlled test. Emarsys supports A/B/n testing within campaigns, which lets you send the same campaign to two groups — one with AI-personalised recommendation blocks, one with manually curated or static product selections — and compare performance.

Setting Up the Test

Split your campaign audience 50/50: Group A receives the AI personalised recommendation block; Group B receives a static curated product selection (your best editorial pick for the season, or your current bestsellers).

Measure the test over at least three to five campaign sends (not just one) to account for natural variability in any single send.

What to Measure

Click-to-open rate (CTOR) — divides clicks by openers, isolating content relevance from subject line performance. This is your primary measure of whether the personalised content is more relevant.

Revenue per email sent — the commercial measure. Total revenue attributed to the email ÷ number of emails sent. Comparing this between Group A and Group B gives you the direct revenue value of AI personalisation.

Products purchased — are customers buying the products that were recommended to them? High recommendation uptake rate (purchases from the recommended products vs other purchases) validates the model’s accuracy.

Most enterprise brands running Emarsys with well-configured product catalogues and sufficient purchase history see CTOR lifts of 15–35% from AI-personalised recommendation blocks versus static selections. Revenue per email typically lifts 20–40% when AI personalisation is active.

Emarsys AI for Lifecycle Email Programmes

The AI models in Emarsys become most powerful when applied to lifecycle email programmes — where the combination of behavioural triggers and AI-driven personalisation creates emails that are both timely and relevant.

Churn Prevention Campaigns

Build a segment: “Churn probability > 65% AND last purchase > 45 days ago AND has not entered win-back campaign in last 60 days.” This segment auto-refreshes as churn scores update. The campaign trigger is segment entry.

The churn prevention email uses the NBP recommendation block to show the specific products most likely to trigger a re-purchase for each customer. This is more targeted than a generic “we miss you” email with random product suggestions — you’re leading with the products the model believes are most likely to convert.

Post-Purchase Upsell Sequences

After a purchase, the “Frequently bought together” recommendation type in Emarsys is powered by a model trained on your specific transaction data. The post-purchase email sent 5–7 days after delivery contains personalised cross-sell recommendations relevant to what that specific customer just bought.

The model improves over time as it accumulates more transaction data. Brands running Emarsys for 12+ months with consistent data feeding see higher recommendation accuracy than brands in their first few months.

Anniversary and Lifecycle Milestone Campaigns

Emarsys can trigger campaigns based on customer lifecycle milestones: one-year customer anniversary, fifth purchase, first time reaching a specific spend tier. These milestone campaigns are enriched with AI-personalised product recommendations.

The combination of lifecycle milestone (emotional relevance) and AI recommendations (product relevance) in a single email creates a touchpoint that feels genuinely personal — acknowledging the customer’s history with the brand while offering something specifically relevant to their buying behaviour.

What the AI Needs to Work Well

Emarsys’s AI personalisation is not plug-and-play for every brand. The models require data to learn from — and the quality and completeness of your data directly determines the quality of the personalisation outputs.

For the AI to perform well, you need:

A complete and current product catalogue — including category hierarchy, product attributes (colour, size, material, brand, price tier), and accurate stock availability. Out-of-stock products in recommendation blocks damage user experience and waste impression value.

Clean, continuous transaction data — order history synced to Emarsys with consistent product identifiers. If your product IDs change between your e-commerce platform and your email platform, the model can’t learn effectively.

Sufficient customer purchase history — for most models, customers need at least two to three purchases before individual predictions become meaningful. For new customers or single-purchase buyers, ensure your fallback logic (category bestsellers, overall bestsellers) is configured.

Consistent event tracking — browse behaviour, add-to-cart events, and purchase events all feed the models. If your onsite tracking has gaps, model accuracy suffers.

Invest in the data infrastructure before investing in complex AI-driven campaign architecture. Strong data with simple campaigns outperforms weak data with sophisticated campaigns every time.

Working With Excelohunt

Emarsys’s AI personalisation delivers genuine commercial impact when it’s implemented correctly — with clean data, well-configured product catalogues, and properly structured test-and-learn campaigns. Excelohunt works with enterprise retail and e-commerce brands to build and optimise Emarsys personalisation programmes, from initial data architecture review through to AI block configuration, testing frameworks, and revenue attribution reporting.

---

Related Excelohunt Services

Looking to implement these strategies with expert support?

• Emarsys — learn how we implement this for clients Get a free audit →