It’s Time to Say Goodbye to Segment Averaging

Segment averaging is on borrowed time. For decades, marketers have used this imprecise method to calculate how broad customer groups might respond to a given campaign—with mixed results. Until recently, however, there was little alternative. Before AI, teams didn’t have the means to turn customer-level behaviors into scalable campaign predictions. So, they did what they had always done: assumed most customers behaved like an “average customer”.

However, statisticians have always been wary of this “flaw of averages.” In fact, a 2012 book of the same name details how future assumptions based on present and historic averages are typically wrong. In it, author Sam Savage explains how statistical uncertainties can easily derail even seemingly robust predictions, impacting everything from government and economic policies to business-level decision-making.

The average customer is a myth

Segment averaging is no exception to the flaw of averages. If anything, it strengthens Savage’s argument. There’s a reason marketing forecasts remain structurally wrong quarter after quarter, and it’s not because the data is bad. It’s because models built on segment averages predict a fictional average customer.

Price sensitivity, channel preference, timing, and messaging triggers all vary at the individual level. A segment of ten thousand people does not respond as one. Assuming otherwise is wishful at best (and self-sabotaging at worst).

Where segment averaging fails 

“Approaching your market as a group of averages can lead you down a path of mediocrity and eventual decline,” explains Rich Edwards, CEO of Mindspan Systems, in the technical consultancy and software company’s blog. “There is no average customer. And making average offers and services leads to bland, undifferentiated experiences that no one loves.”

And if there’s one thing customers expect today, it’s differentiation. Hyperpersonalized experiences, with tailored customer messaging and curated product recommendations, have become the standard. But personalizing a customer’s online shopping experience is different than personalizing an entire ad campaign. Or is it?

AI rewrites the marketing formula

AI made hyperpersonalized CX possible. Using data from searches, transactions, and real-time browsing behavior, AI can not only create highly accurate customer profiles; it can orchestrate personalized experiences based on those profiles. While AI has completely rewired how businesses approach CX, the same can’t be said for marketing. At least not yet.

The reason: customer profiles, which describe who someone is, cannot predict what they will do next. The task of translating past behavior into future action has traditionally fallen on marketing teams. Marketing “intelligence,” in other words, is largely a reflection of human intelligence.

However, humans are humans, not computers. To make sense of the vast amounts of customer data available, human marketers leverage the tools and formulas they have, including, you guessed it, segment averaging.

That was before Marketing AI.

Marketing AI bridges the gap between customer data and intelligent action. It gives teams the power to learn each customer’s behavior from their actual signals, mimic how they think and respond, and predict the outcome of every marketing decision against that individual before a dollar is spent. Not for a segment. For each person.

Introducing the digital twin

The cornerstone of Marketing AI is the digital twin: a continuously updated predictive model of how each person behaves, responds, and decides. It learns from every click, every non-response, every purchase, every service call, and it gets sharper with every cycle.

Until recently, the concept of assigning an AI model to thousands (if not millions) of customers would have been unthinkable. The token cost of running a frontier large language model (LLM) at the depth required to predict individual-level behavior would have been prohibitive at any scale.

That’s not the case, however, for small language models (SLMs).

SLMs represent a different architectural philosophy from LLMs: depth over breadth, precision over generalization. Instead of trying to know everything about everything, SLMs excel at knowing everything about something specific.

In Marketing AI, purpose-built SLMs can provide specialized behavioral intelligence that, combined with each customer’s unique representation and signals, produces individual-level predictions rather than relying on segment averages.

This approach enables customer-specific predictive intelligence without requiring a separately trained and deployed language model for every customer. Conceptually, each customer can have individualized predictive intelligence; technically, the customer representation and predictive outputs are unique to the individual. This is what makes the digital twin practical at enterprise scale.

The prediction only gets more precise

As new customer signals and outcomes become available, the intelligence behind the digital twin can improve, sharpening predictions about how that person is likely to respond to a specific message, which journey path they may take, or what might convert them in this cycle rather than the last.

That’s a major departure from segment averaging. For one, it solves the problem of individual-level analysis. But, more importantly, it minimizes future uncertainty. Thanks to SLMs, marketing teams can accurately predict individual customer behavior. And they can do it at enterprise scale without burning through budget.

That’s more than a revolutionary concept. It’s a reality that’s redefining how modern marketing operates. Those that deploy digital twins early on will undoubtedly have a significant advantage over those that wait. But make no mistake: marketing is undergoing a sea change. Soon, every enterprise will leverage customer-level intelligence to make more accurate predictions and better marketing decisions. And when that happens, imprecise practices like segment averaging will be a thing of the past.