It’s interpersonal: Personas and live segmentation in customer service conversation

Professor Elizabeth Stokoe
6 min readOct 15, 2021


by Marie Flinkfeldt, Sophie Parslow, and Elizabeth Stokoe

Zamurovic Brothers on Noun Project

Who’s the client? According to marketers and designers, knowing the answer to this question is crucial to unlocking and increasing a customer base. Companies can leverage what they know about the characteristics and profile of their ‘typical’ client to shape the way that their products and services are communicated — and, in theory, this will attract more (and more satisfied) clients.

However, our new research challenges the oversimplification of customer ‘types’ — often referred to as ‘personas’ or ‘segments’ — and reveals how ‘segmentation’ unfolds off-script in real-life customer encounters.

Let’s start with some background.

What are ‘personas’ and ‘segments’ — and how are they different from persons?

While the terms ‘persona’ and ‘segment’ can be used interchangeably, they refer to quite different practices. Segments are data-driven groups of (potential) customers who share a need or set of characteristics. They’re often deduced from demographic categories like age, gender, education, or income. In user-centred design, segments are built by analyzing data trails — where people spend time and attention, what they ‘like’, what they buy, and so on. In contrast, personas — “one of the most popular design devices there is” — are fictitious characters created “to mimic a real customer” and are given ‘personalities’ and names such as ‘Nerdy Nina’ or ‘Facilities Manager Fred’.

Personas are widely used to craft communication strategies across different areas of business:

We need to know who we are speaking to so we meet their needs and create an experience that resonates with each of them. (…) Customer support teams can use personas to serve your customers better. When they understand their problems better, your team can empathise with them. You can create scripts and dialogues around common issues.”

But, of course, developing a script to fit a pre-crafted persona implies using it repeatedly — and this is inconsistent with the goal of appealing to clients at a personal level. This tension is apparent in much of popular writing on personas in user-centred design and marketing, which tends to overlook or downplay the fact that scripted conversations unavoidably diverge from ideals of personalized service.

As concepts, both personas and segments have been criticized for pragmatic and more philosophical reasons. While their usefulness in actually driving customer success has been much discussed, others question that our postmodern ‘fleeting’ selves are capturable in this way at all. Most importantly, the use of personas has been associated with reproducing inequalities, since they are “inherently reductive and, if used carelessly, can easily lead to the creation and reinforcing of stereotypes.”

It’s also easy to imagine how following a persona-based script when interacting with an actual client could go wrong. For instance, if the persona is designed as an older person, it may lead to assumptions that produce what has been termed ‘elderspeak’ which includes speaking more loudly or slowly. This might not create friction if such adjustments are warranted, but can be experienced as “the baldest form of ageism” if not.

While personas are fictional generalizations, customer service representatives deal with real people. So how do they design what they say for a presumed client segment in real conversation?

People already do ‘user-centred design’: Some lessons from conversation analysis

We collected and transcribed hundreds of customer service calls to analyze the performance of a deceptively simple and repetitive task: requesting a client’s email address.

A long line of research in the field of conversation analysis (CA) has laid out how requests for information can be accomplished in a variety of ways. The precise words and grammatical forms we use matter a great deal. For example, we found that customer service staff asked for someone’s email in two main ways — either taking for granted the existence and active use of email, or conveying the possibility that the client doesn’t have an address:

(1) Can have your email address please?

(2) Do you have an email address at all?

Requests are shaped in a process called recipient design — a conversation analytic term going back fifty years that captures how people design what they say with an eye to who the other person is and what they might already know. For example, when talking about a third party — say, a colleague — people adapt the way they refer to this person depending on the co-participant’s presumed relationship with the person in question. If you know my colleague well, I can just use her first name (‘Anna’), but if you’ve never met her (and I know this), I’d be better off using some sort of description (‘Anna who’s one of the programmers at work’).

Our examples above are both ‘yes/no’ questions but are the vehicles for quite different conversational possibilities. Example 1 (“Can have your email address please?”) is designed for a recipient who is an email user (“what’s your email” carries the same assumptions but through different grammar) whereas Example 2 (“Do you have an email address at all?”) is designed for a recipient who might not be. In other words, what we see in our data is a form of non-scripted segmentation unfolding live in encounters with clients.

The two core forms for requesting a client’s email also mapped onto another category: age. In short, those asked to provide their “email address please” were ‘hearably’ young; those asked whether they had an “email address at all” were ‘hearably’ seniors. By ‘hearably’, remember that a great deal of customer service happens on the phone and voice quality is an important source of information for both parties.

While there were other differences as well (e.g., the business sector from which the calls were drawn), our data support the claim that the clients’ presumed age informed the design of the requests. This was also reflected in the way older clients were presumed to prefer analog methods of contact in the future, such as receiving a printed and posted confirmation of services by ‘snail mail’. Notably, this was the case even in instances where the client had already provided an email address. So, although they had self-identified as an email user, older clients were sometimes treated as belonging to a group who would prefer more traditional forms of contact if given the choice, as in the following example:

Customer service: So, what I’ll do now for you, I’ll get your confirmation sent straight through. Now, I have got your email address here, but would you like it sent via the post instead? What would be your preference?

The routine but non-scripted character of how customer service representatives in our data asked for clients’ email addresses and preferred mode of communication suggests a careful in-the-moment design. By such means, customer service representatives balance the fine line between age stereotyping and effective, personalized service provision. User-centred design is therefore bound up with recipient design.

Conversation analysis and more responsive segmentation

Without doubt, being good at analyzing what customers, clients, and users say in the moment — and using that to craft service to fit them — is key to customer success.

The ‘email question’ appears in almost all customer service interactions but is not necessarily scripted. Although our research identified two main ways in which the information was elicited, there was further considerable variation in the finer details of language production (including aspects of delivery, such as intonation, restarts, silences, ums and uhs). These details are a crucial part of recipient design and can affect whether a question meant to be helpful comes off as condescending or not. But trying to capture such linguistic variation in pre-written scripts would be counterproductive as the sheer number of alternatives would make them impossible to use in practice.

Regarding personas, some suggest that one way to develop them is imagine how people talk and then write scripts based on those ponderings and wonderings. Others ask: “Why on earth would a marketer create a fake person when they have so many they could talk to in real life?” Although conversation analysts would probably support the latter position, our argument is not necessarily that personas are a bad idea. Rather, research findings can nuance our understanding and development of personas and their use.

The fact is, customer service representatives already work to ‘segment’ clients in their live encounters, and so we can learn much more by analysing inter-personas in interaction.

Marie Flinkfeldt is Assistant Professor at the Centre for Social Work in the Department of Sociology at Uppsala University; Sophie Parslow is an ESRC-funded PhD student at Loughborough University and Conversation Design Lead at, and Elizabeth Stokoe is Professor of Social Interaction at Loughborough University and a consultant at



Professor Elizabeth Stokoe

The London School of Economics and Political Science, specializes in conversation analysis, communication training, & science communication.