The author challenges the idea of the average customer: what needs to change in service development and in how organizations work to make a more individual customer experience possible?
Imagine a news service that adapts to what you need.
On a busy morning you’d say you have ten minutes and listen to a news roundup of exactly that length. You could save a story that caught your interest with a single request. In the evening you’d come back to it in a calm reading mode and dig into the details.
Or maybe you’d have the time, but hitting small buttons is fiddly and the text isn’t as easy to make out as it used to be. You could ask for a view with larger text and buttons, and listen to the article while the passage being read is highlighted.
If you were blind, or driving, you could move from story to story by conversation and listen to the news in a voice and at a speed of your choosing. You could ask the service to describe an infographic in more detail or compare the figures in a table. Video news could come with audio description.
A news service is just one example. It could just as well be an online shop, an insurance comparison or planning your studies.
The principles of accessibility and Universal Design are not new. Until now, though, building and maintaining several versions of a user interface has been laborious. AI can make it possible to adapt views and functions during use, without every option being built separately. [1]
People already use AI to describe content and to help them get things done. In WebAIM’s 2026 screen reader survey, 60.1 percent of respondents said they use AI to describe images and 40.6 percent for guidance when doing things online. [2] What if these capabilities were part of the service itself and included in your subscription? You’d get more out of the service you pay for, without copying content or switching from one app to another.
The service could reach new customers and serve existing ones in situations where using it used to be impossible or not worth the effort. And when you get more out of a subscription, keeping it may feel more attractive than before.
But which of these needs go unnoticed if development is steered by assumptions about an average customer?
The cost of average usability
Customer segments and user personas help define who services are designed for and for what purposes. At the same time, they simplify reality. The problem arises when they’re assumed to tell you everything essential about your customers.
Age, life situation and interests don’t yet tell you how a person wants, and is able, to use a service. Losing your sight doesn’t make you lose interest in financial news or cat videos. A service can get its recommendations exactly right and still make using the content difficult, unpleasant or even impossible. According to Eurostat, almost one in four (23.9%) people aged 16 or over living in the EU reported in 2024 long-standing limitations in their usual activities due to health problems. [3]
People’s needs and preferences also vary. Functional ability, load and environment all affect what works at any given moment. [4] Yesterday’s choices shouldn’t box a user in too tightly.
Picture a benefits application that gets submitted with a relative’s help and after two calls to customer service. In the statistics the transaction looks like a success, even though it cost the customer several evenings and the organization staff time spent advising. In an online shop, the same effort may mean the purchase never happens. Not everything needs to be faster, though. Weighing up a purchase is one thing; figuring out how to pay for it is another.
A lack of accessibility can show up in many ways: as a need for help, as extra work, or as a purchase that never gets made. That is the cost of average usability: the service works, as long as the user bends far enough.
What goes unasked, and of whom?
If a customer reads the news, do they prefer reading, or is the service simply missing the option to listen? The usage data may be accurate, but the conclusion drawn from it can be wrong. Usage data can’t tell you how many people would rather listen to the news.
The way you ask also limits the answers. Ask customers how easy the service is to use, and you learn about their experience of the current service. Ask them, too, what would help them use it, and you can identify needs the current solution doesn’t meet.
Sometimes the question itself blocks the information. In the UK’s teacher training application service, applicants first had to state that they were disabled before they could give more detail about any limitations. An applicant with dyslexia, for example, might not recognize themselves in that first question. The first question was removed, and all applicants could choose to answer the question about disabilities and health conditions. After the change, 15 percent of applicants reported a disability or health condition. A year earlier the share had been 4 percent. [5]
It’s also important to recognize whose experiences go unheard. Beyond current users, you need to understand those who find the service hard to use or who have been left out of it entirely. The reason might be price, lack of interest or difficulty using the service. Those reasons won’t necessarily show up in usage data or customer feedback.
Customers’ abilities and situations of use vary, and not every need that comes with them can be identified or put into words in advance. That’s why just listening to customers isn’t enough. A good design process also includes user testing. It can reveal that one person understands complex topics best through a visualization, while another makes sense of things better on the move.
If we don’t broaden our understanding of how people want to use services, and are able to use them at all, we’ll only get halfway in making use of the new possibilities.
Let the service flex, but keep the foundation solid
Researchers at Google compared a traditional and an adaptive user interface in a banking service prototype. In the adaptive version, views and functions changed according to the user’s needs. A user could, for example, ask the service to set a spending limit and add tracking of it to the home screen. The service created the new function at the user’s request.
In the study, which had 72 participants, the adaptive version scored 84 out of 100 for usability, while the traditional version scored 54. The difference was considerable, even though this was only a limited prototype study. [1]
But how does the customer know they can ask for something like that? The service could show examples of what it can do and offer options to choose from. Customers shouldn’t have to work out for themselves what they can ask of the service and how to phrase the request. The service also has to work for customers who can’t, or don’t want to, adapt it.
The design has to define the principles by which the service may create new views and functions. What can it adapt, and what must stay the same? Consistency doesn’t mean the same view for everyone. When paying for a purchase, a user hardly wants a new way of approving the payment every time. Even when the view or the way of using the service changes, the customer still has to understand what they’re doing and what follows from their choices. Accessibility has to be part of the design principles so that views generated during use also work across different ways of using the service and with assistive technologies.
Adaptation has to stay under the customer’s control. Users must be able to correct the service’s wrong assumptions and go back to the familiar way of doing things. Content, in turn, has to stay reliable: a summary or description must not distort the matter or leave out information needed to make a decision. The same principles must guide every team developing the service.
From insight to impact
You can’t build everything at once. Start by choosing a development target that matters to both the customer and the service provider.
The target could be, for example, a demanding application process. The customer could proceed independently and, at tricky points, ask for a more detailed explanation, an example or step-by-step guidance. A pilot could track where customers ask for help, what kind of help they ask for, and whether it helps them move forward. At the same time, it would show whether the process gets easier and whether the need to contact customer service drops.
A pilot can reveal problem spots, but looking more closely at customers’ experiences helps you understand what causes them. On that basis you can judge what’s worth changing and whether the benefit of a solution is worth the cost of building, quality-assuring and maintaining it. Learning shouldn’t end with the pilot, though. Once the service is in use, you can learn more about customers’ needs and find new ways to make use of adaptation.
What you learn must also change what you do. If filling in a form is hard, the problem won’t necessarily be solved by better instructions or interface changes. The cause may lie in the process itself, and changing it takes collaboration across team boundaries.
Learning also has to reach beyond the individual service. You need a shared understanding of what you want to improve, what level of quality you’re aiming for, and who decides on the necessary changes. When lessons learned feed into shared decisions and ways of working, one insight can benefit the whole organization.
Finally, back to the news service from the beginning. A broader understanding of customers’ needs can make the service more valuable to subscribers, strengthen their wish to keep subscribing, and help reach new customers.
Which brings us back to the average customer. Who do we want to design services for? If the answer is still an assumed average customer, AI’s potential for more individual service will largely go untapped.
Sources
- Arora et al. (2026): Self-Evolving Systems: Moving Beyond Deterministic Interfaces to Adaptive Generative Interfaces. Google Research. A limited prototype study that illustrates the possibilities.
- WebAIM (2026): Screen Reader User Survey #11 Results. 1,780 valid responses, collected in July–August 2026.
- Eurostat: Population with disability. EU-SILC, 2024 data. Covers people living in private households; the figure does not measure difficulties using digital services or the need for assistance.
- W3C WAI: Diverse Abilities and Barriers.
- Department for Education (2022): Changing how we ask about disabilities and health conditions. Becoming a teacher design history, 9 November 2022. The comparison covered the first 28 days of each recruitment cycle.





