Why Marketers Should Replace Traditional Segmentation Models with Predictive AI



Head of Marketing at Relay42

Anthony Botibol
Head of Marketing 

Despite the wide availability of artificial intelligence, machine learning, and predictive analytics, marketers are still resorting to conventional methods of customer segmentation when attempting to personalize their marketing campaigns. These traditional segmentation strategies are typically rules-based models or demographic targeting tactics, which can only get you so far and are increasingly being overshadowed by the power and prediction capabilities of AI and predictive analytics. The ability for marketers to use AI to harness vast amounts of behavioral data patterns from prospects and customers represents a significant shift in how they can approach audience creation, personalization, and deciding on the next-best action (NBA) for an individual customer.

This blog will explain why traditional segmentation tactics should be replaced with predictive AI, and how marketing teams can make the switch.


Ultimately, segmentation of a campaign audience is better than none at all because it ensures that a brand reaches the most relevant audience, making the campaign more effective. Historically, these segmentation strategies have leaned on rules-based models to categorize customers into groups to receive specific types of offers or messages. For example, a brand might decide to target females in their database who are between the ages of 30 and 39, who purchased 1 or more items within the last 6 months above a total lifetime value of €50. This segment would then be used to target that audience with a specific offer or product that is deemed suitable for that cohort of customers. It’s perhaps the simplest form of marketing segmentation but is also a process born from offline strategies in the 1990s, which pre-dates social media, super-fast broadband connections, mobile apps, and always-on devices. Furthermore, it is based on known customers only.

Recency, Frequency, Monetary (RFM) segmentation is one of the most popular customer segmentation techniques that goes beyond demographic data by focusing on the recency, frequency, and financial value of a customer’s transactions and has been a powerful tool in the marketer’s toolkit for some time. Again, it’s a process that evolved out of the 1990s to optimize the costs of sending expensive direct mail and catalogs to customers. Ultimately, the desire was to ensure that expensive mailers were not sent to customers who were unlikely to make a purchase and to treat groups of customers differently based on their recent (or not) purchase behaviors. The tactic was also then adopted and applied to email marketing for outbound campaigns and is still utilized today as a way to target customers in more tailored ways for a variety of campaigns and channels.

Examples of recency, frequency, monetary (RFM) segments

However, in the context of today’s rapidly evolving digital landscape where shopping and buying behaviors are moving further towards digital channels there is a huge treasure trove of behavioral patterns occurring all the time that older segmentation tactics like RFM are not effective at dealing with, but which predictive AI can handle.

Here are some of the limitations of traditional marketing segmentation and RFM:

  • Limited scope of data
    RFM relies primarily on historical transaction data, which will not accurately reflect customers’ current interests, such as browsing behavior which could indicate an immediate interest or disinterest in purchasing. Customer preferences, social media engagement, customer service interactions, or brand sentiment are examples of important signals that are not part of an RFM model.
  • Reactive, not proactive
    Ultimately, traditional rules-based segmentation strategies are a way of looking backward by measuring past behavior to infer future actions.

    For example, a popular RFM segment is commonly labeled as ‘At Risk’. Ultimately, these are customers with low Recency (i.e. have not purchased in a while) but had normally a relatively high frequency of purchasing and a high monetary value to the business. It’s a rudimentary way to categorize someone who we want to try and win back but while this puts some customers into a segment of presumed churn, what it does not do is predict churn before it becomes apparent through their buying behaviors.
  • Known customers only
    RFM and rules-based segmentation models only apply to known customers. Therefore, they have limited use for nett new customer acquisition goals, but can be applied for outbound and direct marketing campaigns to improve upsell, cross-sell, and retention marketing programs. For real-time inbound digital marketing, social or paid media these strategies are limited at best to using RFM segments to create some lookalike targeting audiences.
  • Lack of individual personalization
    Finally, and most importantly, what rules-based segmentation does not do is consider any individual customer inside an RFM cell or segment. It doesn’t take account of potential purchase behaviors that are happening on the website in the moment, which is where a conversion can be won or lost through effective personalized experiences.

So while RFM, demographic targeting, or simple lifecycle-stage segmentation strategies have been effective in the past, these methods all operate under a significant limitation: they rely on static data, known customers, and predefined rules that cannot adapt to new information.


Predictive AI, in contrast, offers a dynamic and continually evolving approach to customer data analysis and allows for the personalization of a message or offer to an individual rather than a group. By leveraging self-learning algorithms, predictive AI can analyze vast amounts of digital event data (impressions, clicks, views, likes, purchases, etc) to identify patterns in the moment and, in real time, process that data alongside historical CRM and purchase data to predict a future behavior at an individual level. This allows for highly personalized inbound and outbound marketing strategies that can adapt rapidly to changes in customer behavior.

Unlike RFM's static and backward-looking analysis, predictive AI is inherently forward-looking and retraining itself to become more accurate over time. These models can predict future customer behaviors with a high degree of accuracy, identify emerging trends, and segment customers into highly personalized groups (or a ‘Segment of One’) based on their predicted future actions. This proactivity ensures marketers can better anticipate customer needs, tailor communications more precisely, and provide relevant offers to customers at the optimal time and through the most effective channels.

Here are 4 specific ways Predictive AI will improve your customer segmentation strategies to improve personalization:

  • Benefit from enhanced prediction capabilities
    The advanced and deep-learning algorithms can predict future customer behaviors with a higher degree of accuracy than, for example, RFM, and continuously learn and adapt when new data comes in. Marketers, or humans in general, cannot decipher the sheer volume of data or identify complex, non-linear patterns in the data, and so predictive AI does the heavy lifting to make prediction scores at an individual level or create predictive customer segments.
  • Leverage a broader scope of data
    Predictive AI can handle unstructured data such as text, images, and social media activity, in addition to structured data, enabling it to consider a much broader and richer view of a customer’s needs. Better still, the ability to process and react to digital events through real-time data processing is an advantage that is critical for converting customers in the moment.
  • Activate individual customer experiences
    We’ve already talked about how predictive AI can enable individual personalization (or the ‘Segment of One’), but AI can also predict future customer actions to power dynamic customer journeys and follow-on experiences, or to deliver next-best actions, next-best experiences, and decide on the optimal channel mix. Rather than follow the path that the marketing team lays out for a buying process, predictive AI can move a prospect or customer between journeys or adapt the journey mid-flight based on real-time behaviors and/or model scores.
  • Improve efficiency and ROI
    While no one can say that building, training, and deploying predictive AI models is an easy task, manual segmentation is not a quick process either. However, when marketing teams can replace traditional segmentation with AI they optimize the resources that were previously needed to continuously rebuild conventional segments, can better predict at-risk customers before they churn, and can forecast lifetime value too.


The use of predictive AI in marketing is a higher-order capability that marketers should strive for, but with the understanding that a fully AI-driven and cross-channel personalization strategy doesn’t happen overnight. Knowledge, skills, data, technologies, and a whole host of processes within the business need to adapt, and in cases where data is incomplete or limited it can yield inaccurate results. However, while the process of building, training, and deploying a predictive AI model would previously have taken many months of dedicated engineering and data work, pre-built AI models that can be customized and trained on your own data sets can turn months into days – and when incorporated with a Customer Data Platform it, more importantly, uses trustworthy data and puts predictive AI into the hands of marketing teams. The accelerated developments in AI modeling across the Martech and Adtech space is therefore making the ability to deploy predictions more accessible.

Examples of pre-built models could include:

Examples of predictive AI models used in marketing including likeliness to convert, search recommendations, behavioral clustering, next best channel, next best product and many more


Predictions inside marketing channel tools usually underwhelm or are limited by the data limitations and bias of the tool itself. For example, while an email marketing solution has the right data to predict when to send an email to a recipient, it does not hold the data required to predict which offer, product, or content to give them, when to offer it, or whether to divert them away from sending an email entirely.

For predictive AI to be used effectively therefore, to make unbiased decisions across channels and strategies, marketers need to zoom out from their channels entirely and apply centralized AI decisioning that can select the optimal experience using the whole breadth of data and behaviors regardless of how or where a customer chooses to interact with their brand.


AI models also need time to train themselves to a degree of accuracy that is trusted, and then monitored and arbitrated by the business to continuously refine the predictions being made, or even pause models during unpredictable times (for example, during a Black Friday period where irregular buying patterns may not correlate, or perhaps where outside influences can impact consumer behaviors.)

Example of predictive AI leveraged through a Customer Data PlatformJourney mapping and the ability to orchestrate and arbitrate cross-channel journeys is therefore a critical requirement for the governance of AI models because it provides a composable way to drop AI models into journeys, ensures that conflicting AI predictions don’t overlap, and enables you to inject models at lifecycle stages, improve audience selections, and even combine offer, channel, recommendation, and conversion/churn models towards a fully-automated Next Best Action marketing strategy.

Relay42 provides out-of-the-box predictive AI models that can be fully trained and customized by marketing teams for a variety of needs and will inject AI scores into existing channel tools, drive the personalized content or recommendations on the website, alter an individual’s journey, or even halt an existing message from landing with the customer entirely. Relay42 utilizes the full breadth of behavioral event data from paid and owned digital channels, alongside all other online and offline data including anonymous website traffic, to make unbiased predictions that can then be applied to every campaign, channel, or always-on journey. Combined with built-in Journey Orchestration it enables the user to arbitrate their models, analyze performance with journey analytics, and have full helicopter control of the omnichannel experience.

This shift towards predictive AI through Relay42 is not just a trend but a fundamental evolution in how customer data is analyzed and acted upon by digital marketers. It changes the entire way that teams think about marketing and enables an always-on approach to Next Best Action where models are deployed to affect a variety of actions and insights such as conversion, churn, lifetime value, and real-time offers.