A Glimpse on New Advances in Segmentation

Segmentation is a fundamental topic in marketing strategy. Appreciation of its importance has continuously grown since the 1950s. Segmentation is crucial to targeting messages, fitting preferences of different consumer groups, or creating other types of differentiated programmes. In consumer marketing, segmentation may be based on demographics, perceptions and attitudes (mental constructs more generally), and behavioural measures (e.g., overt actions tracked or recorded). A recurring challenge in segmentation is the integration of constructs or variables from different content areas or of different measuring scales in a more comprehensive model. It should be kept in mind that segmentation is the marketing-driven interpretation and application of analytic results from marketing research studies, often using quantitative analytical methods (e.g., clustering).

The American marketing research association for practitioners Greenbook recently (July 2024) dedicated its newsletter to new advances in segmentation, linking to three articles on the subject in marketing practice (from most recent to older): (1) “The Future of Segmentation: AI-Driven Approaches to Understanding Your Market”; (2) “The Evolution of B2C Consumer Segmentation in the Digital Age”; (3) “The Shifting Sands of Demand for Segmentation”. This post will address the recent developments in applied segmentation through reference to those informative and instructive articles for raising key issues and suggesting some useful notions. Unsurprisingly, methodological advances in an area such as segmentation are related these days to technological advances, and lately it implies in particular analytic capabilities and modelling methods of artificial intelligence (AI).

In his article on the future of segmentation (24 July 2024), Hakan Yurdakul (CEO at Bolt Insight) describes areas where AI is making considerable impact (“revolutionizing market segmentation”): (a) Data Integration and Analysis; (b) Predictive Analytics; (c) Personalisation at Scale; and (d) Real-Time Segmentation.

The strength of AI algorithms in Data Integration and Analysis is in allowing the integration of data from a variety of sources and of different types. At start, Yurdakul explains how from earlier days segmentation was based mainly on demographics (e.g., gender, age, education, location), which too often has provided a limited and inadequate account of differences in consumer behaviours and mindsets. However, it should be recalled, that since the 1980s more types of variables have already been combined and integrated in clustering and thereof segmentation models, such as aspects of consumer lifestyles and standards of living; marketing researchers already started to integrate data from surveys and records of actual (customer) behaviour. Traditional clustering methods (K-means, Hierarchical) made a more-or-less good, robust job in handling different types of data, and later on two-step clustering analysis offered some improvement in data integration. Nonetheless, marketers now have access to an even greater variety of data, as Yurdakul suggests, such as social media activity, browsing histories and purchase patterns. AI algorithms can provide stronger, much needed techniques (entailing statistics & machine learning – ML) for combining and analysing the data of greater volume and variety. They allow brand marketers an opportunity for unveiling “a direct path into consumers’ thoughts and patterns”.

Personalisation has become a crucial, ‘hot’ topic of practice in marketing; it embodies the most granular form of segmentation, down to the individual level, a key to implementing ‘1to1 marketing’ (cf. Martha Rogers & Don Peppers). Yurdakul relates to the significant contribution (‘breakthrough’) of AI in enhancing abilities of content customisation, and tailoring recommendations and promotions to the characteristics and preferences of individual consumers.

Yet, personalisation is an activity that is usually derived from capabilities of more detailed segmentation, but it is not the operation of segmentation per se (i.e., using data available per individual does not constitute segmentation). Yurdakul does not clarify in his article where the line is drawn between implementing models of segmentation and acts of personalisation, or how to combine them. Furthermore, personalisation does not necessarily involve the idiosyncratic data of individuals but rather profiles of many small and detailed segments to any of which individual consumers can be assigned. Oftentimes, the personalisation initiative will be based on a combination of personal data augmented by segment-level profiles (e.g., used as indicative ‘flags’).

Predictive Analytics is definitely another vibrant and progressing area in marketing. It involves increasingly advanced models (e.g., generalised linear models, neural networks) for predicting future behaviours, membership of consumers in various classes, customer journeys, and more. Yurdakul notes how AI can help brand marketers, by analysing historical data, to “anticipate trends, adapt strategies and proactively meet consumer needs”. Of course, predictive modelling may use other types of data on consumers (e.g., their social connections, interests) to predict future behaviours. Real-Time Segmentation is apparently the less familiar area, but its relevance is growing for enabling faster responses to consumer requests and actions. AI is called upon for the ability of its models to thrive on dynamic data, continuously updating and re-assigning consumers as needed. A key capability noted by Yurdakul in this context is “agility” which enables brand marketers to “maintain relevance and keep a competitive advantage”.

The use of individual-level data directly (e.g., for personalised offerings), or as part of a clustering model, raises ethical issues. It concerns also matching individually identified data with segment-level data; concern holds even when involving actual customers of a company. Two main questions that arise are (1) for what purposes is the data going to be used? and (2) how is data practically going to be utilised? The marketing or brand managers in companies are required to provide adequate answers and take measures to protect the privacy of consumers; they have to act in compliance of regulations such as GDPR (Europe) and CCPA (California). Yurdakul suggests measures, including ‘fencing’ consumer interactions so that personal and interaction information of participants in a platform is not ‘leaked’ for training public large language models (LLMs), or basically keeping consumer interactions anonymous. The first recommendation may be applicable to other types of AI models as in artificial neural networks (ANNs) more broadly, and the second could imply banning attachment to other sources of personal data.

Ed Lorenzini (CEO at Analyse Corporation) sheds light on more issues in segmentation in the digital age in his article (23 September 2023) on the evolution of segmentation in consumer marketing (B2C). He advocates, similarly to Yurdakul, the need to break-away from demographic-based segmentation, and to “combine demographic data with psychographic and behavioral insights to fully understand consumers”. The digital age is signified by the abundance and variety of data that marketers can access and utilise, especially online behaviours but also with other implications (e.g., psychographic or lifestyle).

Lorenzini characterises advanced segmentation by the incorporation of greater variety of information on consumers beyond demographics. A particular topic of interest in his article is the alignment of customer journey mapping with advanced segmentation — that is, considering the behavioural information on customer interactions during their journeys as an integral dimension of advanced segmentation. Behavioural segmentation on customer journeys with brands can help marketers to “reveal consumer habits, allowing the businesses to tailor interactions to meet specific needs”. He proposes, as an example, that providing comprehensive product information would be specifically helpful to consumers who research extensively before purchasing. Lorenzini further describes how the segmentation with respect to customer journeys links to predictive modelling and personalisation. In the future, Lorenzini foresees an interlinkage of advanced segmentation with technologies of augmented reality (AR), virtual reality (VR) and Internet-of-Things (IoT) devices in providing novel methods for consumer interaction and data collection.

It is conspicuous, however, that Lorenzini refers to the simultaneous use of different segmentation variables (e.g., needs or benefits, values, interests, behaviours), but he does not seem to suggest using them conjunctively to form multi-facet clusters or segments of consumers or customers. He suggests that AI and ML methods and models can analyse large data sets and uncover complex patterns in a way that implies a somewhat different approach of ML algorithms to segmentation: identifying “correlations between consumer behaviours, interests, and demographics, creating detailed consumer segments”. It remains unclear what form these segments take (see further comment in conclusion below).

Lorenzini notes the crucial challenges that advanced segmentation presents, including data privacy concerns, resource allocation, and ethical data usage. While application of AI techniques tends to be resource-intensive (“requiring sophisticated software and skilled personnel”), he proposes that businesses can start with simpler techniques and gradually upgrade, and also train their existing staff, to manage costs. With regard to ethical data usage, he emphasises the need to respect consumer preferences and avoid discriminatory practices, while putting forward mutual benefits for the brand and its customers.

In the third article on shifts in demand for segmentation (12 April 2023), Chris Crook (managing partner at Nature) contributes primarily by stressing the gaps between simpler and more complex models of segmentation. The insights of Crook are based on his professional experience in the Australian market. He also points to a distinction between segmentation for strategic purposes, aiding businesses in contexts such as service marketing, retail and FMCG, and segmentation aimed at tactical purposes, “where the segmentation itself is a means to an end (e.g., personalisation)” (recall the earlier comment relatedly). Crook focuses on strategic segmentation.

The very basic segmentation applications represent simple representations of demographic variables, whereas the more complex models are based on attitudinal variables. Crook argues that the attitudinal-based models tend to be overly complex and esoteric, and hence have been justifiably criticised for not being actionable enough in practice, and thereby were soon shelved and abandoned. Crook suggests that the early 2000s marked a turning point in Australia in balancing between stakeholders’ needs for these types of segmentation, and furthermore to deliver on both. That happened, according to Crook, with the adoption of a new approach of appending databases to uncover consumer segments, but unfortunately Crook does too little to explain how that actually works (e.g., what is the “attribution rate of 70%” in “appending consumer segments”?). He notes that segmentation involves science and creativity, where success is driven by stakeholder engagement, survey design, framing, and nuanced trade-offs at the analytics stage between segmentation meaningfulness and database appending accuracy.

• Note: From the descriptions and commentary of Crook it may be inferred that the methodology is based on appending a number of data sources with different types of consumer information (e.g., demographics, needs in the product category, attitudes, {personality} traits). Importantly, the data sources (databases) to be combined may be internal or external to the organisation. The attribution appending rate seems to correspond to the extent of being successful in matching between consumers across databases (this aspect could be subject to ethical data issues).

There are two confusing issues apparent in the articles reviewed above. In relation to the impact of advanced technologies and methods, associated mainly with AI or ML, the authors refer to areas such as predictive analytics and personalisation that are not strictly or specifically about segmentation — these areas are adjacent to segmentation in the sense that may use the results or products of segmentation models and may complement segmentation analyses. Additionally, the approaches to and types of segmentation modelling advocated with the aid of AI remain somewhat ambiguous. In particular, it is unclear if methods are used to construct discernible and internally homogeneous groups of consumers (on the basis of ‘clusters’) or a fuzzy form of classes or segments is constructed; possibly the analyses are focused on highlighting associations between specific constructs within the data pool (e.g., pairs or triads of attitudinal and behavioural constructs). The nature of segmentation applied is not properly explained.

Advanced segmentation, especially when aided by AI capabilities and methods, offers beneficial opportunities to companies and brands for better understanding the consumers and addressing their needs and wishes. Yet, they need to keep certain guards in protecting consumers’ privacy, fairness and ethical use of data — these are keys to gaining customer trust. Yurdakul strongly advises companies to adopt AI-driven segmentation methods for gaining more actionable insights but doing so while taking care in how they use the data to maintain customer trust and not to breach data ethics. Prior to Yurdakul, Lorenzini advised that with “careful planning and responsible practices, businesses can enjoy the benefits of advanced segmentation while building customer trust”.

Segmentation becomes especially more effective and influential when it is data-driven and based on application of appropriate analytic techniques and modelling. However, it should be observed that ‘advanced segmentation’ is not a matter only of the variety of data employed (e.g., demographic, attitudinal, behavioural) but also of the advanced analytic methods that are applied, and the models they produce (i.e., multivariate or multi-faceted). Segmentation can be a powerful instrument for companies and brands, for strategic and tactical purposes, but they need to follow certain methodological and ethical acceptable guidelines to win their golden grail in the market competition.