Batch-and-blast email still exists. Plenty of brands send the same message to their entire list every week, measure success by open rate, and wonder why their unsubscribe rate trends upward every quarter. The approach is not just inefficient; it is actively costly.
This piece sets out what batch-and-blast actually costs, how AI segmentation replaces the underlying logic, and what brands need in place before AI tools can deliver on their potential.
The visible cost of mass email is easy to quantify: unsubscribes reduce list size, and list size is a paid-for asset on most email platforms. But the invisible cost is larger.
When you send irrelevant emails to disengaged subscribers, inbox providers register low engagement signals, specifically low open rates, deleted-without-reading behaviour, and spam complaints, against your sending domain. Those signals degrade deliverability for every email you send, including to the engaged subscribers who do want to hear from you. The damage is cumulative and non-linear: once your sender reputation drops below certain thresholds, recovery takes months of disciplined sending.
There is also revenue left unrealised. Sending a promotional email to a subscriber who purchased the featured product last week, or sending a high-discount offer to someone who would have bought at full price, represents both margin erosion and missed opportunity. Undifferentiated messaging produces average results across a heterogeneous audience.
Traditional segmentation divides your list based on attributes you have already observed: geographic location, gender, purchase category, acquisition source. It is better than nothing, but it is backward-looking. It tells you what a subscriber has done, not what they are likely to do next.
Predictive segmentation uses machine learning models trained on your customer data to assign forward-looking scores. Instead of grouping customers by what they bought, it groups them by what they are likely to buy, when they are likely to buy it, whether they are at risk of churning, and how much revenue they are likely to generate over a defined future window.
The practical difference is material. A customer who purchased once six months ago and has not engaged since looks the same as any other lapsed buyer in a traditional segment. A predictive model trained on thousands of similar purchase patterns might identify that customer as having a 70% probability of making a second purchase within 30 days if contacted with the right trigger. Traditional segmentation treats them as a win-back candidate. Predictive segmentation treats them as an active sales opportunity.
The four predictive scores with the most commercial value in e-commerce email marketing are:
Purchase likelihood: The probability a given subscriber will make a purchase within a defined future window (typically 30, 60, or 90 days). This is the core signal for prioritising send budget and personalising offer depth.
Churn risk: The probability a customer will not purchase again. High-churn-risk customers warrant a different intervention than low-churn-risk customers. Sending a discount to a high-value customer who shows no churn signal wastes margin. Sending it to a high-churn-risk customer who responds to incentives can recover the relationship.
Predicted lifetime value (LTV): The total expected revenue from a customer over their remaining relationship with your brand. LTV-based segmentation allows you to invest acquisition and retention spend in proportion to expected return, rather than treating all customers as equivalent.
Next-best product: Based on purchase history, browse behaviour, and cohort modelling, what is the product this specific customer is most likely to buy next? Recommendations driven by this score consistently outperform generic "top sellers" or "you might also like" logic.
Klaviyo's predictive analytics platform surfaces all four of these scores natively for brands on its platform. The Klaviyo customer retention research documents a case where one retail brand grew Klaviyo-attributed revenue by 44.6% year-on-year after integrating predictive segmentation, with predictive analytics accounting for 12.4% of total Klaviyo-attributed revenue.
Predictive models are only as good as the signals they receive. The most valuable behavioural inputs for e-commerce email segmentation are:
Browse behaviour: Which categories and products a subscriber views, how frequently, and for how long. Browse data predicts purchase intent more accurately than demographic data in most categories.
Purchase cadence: How often a customer buys, and how that cadence is changing. A customer whose purchase frequency is accelerating is a candidate for VIP treatment. One whose cadence is slowing is a candidate for intervention.
Engagement decay: Whether a subscriber's email engagement (opens, clicks) is trending up or down over a rolling window. Engagement decay is often a leading indicator of churn before it appears in purchase data.
On-site search queries: What a subscriber searches for on your site signals intent that has not yet translated into a browse or purchase event. Capturing this data and passing it to your email platform requires explicit integration but produces a materially richer signal set.
Klaviyo has the most developed native AI segmentation capability for e-commerce, including predictive LTV, churn risk scoring, and next-best product recommendations built directly into its segmentation builder. For Shopify brands in particular, the native data integration means the model trains on a comprehensive activity record without complex data pipeline work.
Attentive specialises in SMS-first marketing but has expanded its AI personalisation into email, with behavioural segmentation built around mobile engagement signals. It is worth considering for brands where SMS and email are both significant channels and where cross-channel behavioural data is available.
Salesforce Marketing Cloud offers enterprise-grade AI segmentation through its Einstein engine. The sophistication is high, but so is the implementation overhead. Salesforce research indicates that AI-powered email programmes deliver 41% higher revenue than manual campaigns when AI is integrated across the full email workflow rather than applied as a single feature.
This is the part most platform marketing omits. AI segmentation requires a minimum viable data foundation to produce reliable outputs. Specifically:
Sufficient historical purchase data. Predictive models trained on fewer than 500 to 1,000 purchase events per customer segment produce unreliable scores. Brands with smaller catalogues or lower purchase frequency need to aggregate data over a longer window before predictions are statistically meaningful.
Clean, consistent data pipelines. If your e-commerce platform, email tool, and CRM are passing customer identifiers inconsistently, the model cannot build a unified profile. Data hygiene is a prerequisite, not a nice-to-have.
Accurate product attribution. AI needs to know not just that a purchase happened, but what was purchased, at what price point, in which category, and through which channel. Partial attribution degrades the quality of next-best-product recommendations significantly.
A testing framework. AI segmentation is not a set-and-forget capability. The models require ongoing validation through holdout tests (sending a control group the standard segmentation while the AI segment receives the predicted treatment) to confirm the lift is real and not an artefact of selection bias.
Brands that have the data infrastructure in place consistently see open rate lifts of 15% to 25% from send-time and segment optimisation, and revenue-per-email improvements of 20% to 40% when predictive personalisation extends to content and offer selection. The Salesforce benchmark data puts the upper bound at 41% for fully integrated AI programmes.
If you are assessing whether your data stack and email programme are ready for predictive segmentation, the Viaduct Generation Intelligence page covers how we approach this diagnostic. For brands ready to act on the findings, the Optimisation page sets out how we implement iterative improvement frameworks, and the Contact page is the fastest route to a direct conversation.
