Summary: User targeting in UX has progressed through 5 stages toward treating each user as a unique individual: target audience, personas, customization, personalization, and (to come) AI-driven individualization.
Designing for the user. That’s the ultimate goal — or possibly ideology — of user experience. We don’t design for ourselves or for everybody in the world, but for the specific individual who has his or her hand on the mouse. (Or finger on the touchscreen, or speaks to our voice interface, or uses whatever mode of interaction we have. The point is that the user is the person driving the user interface and engaging in a dialogue with our design.)
Each user is a unique individual with his or her special skills, goals, desires, and feelings. Even so, in the past, computers treated users as members of a gray, undifferentiated mass. Luckily, there’s a long-term trend toward narrower targeting of users, and upcoming AI-generated user interfaces may finally achieve true individualization where each user is treated as an individual.
Different users have different needs. What differences do you see between these two users? This is a trick question because those differences that are visible in a picture are not likely to be important for user interface design. Instead, we should design based on users’ behaviors, which depend on non-demographic differences such as skills, goals, and motivations. (DallE)
1950s: Target Audience
The oldest step toward differentiation is the concept of a target audience. This was one of the first things I learned about user experience when I started in the field 41 years ago. We cannot design for everybody, because then we’re really designing for nobody. We have to specify our target audience and design for them, while ignoring all the users who are not in the target audience.
The concept of a target audience is even older than my UX career: I first learned about target audiences when I was a teenager taking a course on public speaking. But it’s older than that. I don’t know who conceived of the idea of a target audience, but it was prevalent during the heyday of big-budget advertising in the 1950s.
Designing for the target audience: we identify the users who are likely to be profitable customers and ignore everybody else. (Leonardo)
As a simple example, if we have a website that sells dog food, we only target dog owners. Cat lovers? Who cares! In fact, we would likely define our target audience even more narrowly than people with dogs: we would only design for people who are likely to buy dog food online in sufficient quantities to be profitable customers.
Around the mid-1980s, software developers such as Alan Cooper recognized that the target audience as a whole was too broad to guide user interface design. Some groups of users might use the same application in quite distinct ways from other users. Let’s say you’re building accounting software for small businesses. Some businesses may consist of just the owner and have a very small number of transactions. Because the owner (and sole employee) is also the user of the software, he or she would be very familiar with every transaction. Other small businesses might have, let’s say, 10 employees, one of whom is a dedicated administrator, though not a finance professional. This second set of users is likely to have somewhat different needs in accounting features, for example, processing and searching through a larger body of transactions.
As another example, an auction website like eBay would need to cater to both buyers and sellers. Each category would further include both users dealing in a small set of products and power users selling or buying hundreds or thousands of products. Again, different needs for different sets of users.
Based on this realization, it became common to define a handful of different personas, each of which would be the archetype of a certain group of users with roughly shared behaviors and user-interface needs. Often, one of these personas will be defined as the primary persona. Suppose multiple personas differ in what user interface design would be best. In that case, the conflict is resolved in favor of the primary persona, with the secondary personas having to make do with a less-than-perfect design.
When using personas, the target audience is further subdivided into groups of users (called personas), which share many characteristics in terms of behaviors and needs. (Leonardo)
In the 1990s, we finally arrived at the stage where each individual user could get a different user interface optimized for their needs. With customization, the user can select from various preference settings that alter the behavior of the design.
A classic example is the ability for each user to select their own color scheme.
It sounds tempting to resolve design debates by simply leaving the choice to the user. If the design team can’t decide whether design option A or B is best, what could be more user-friendly than allowing each user to pick whether they want A or B? Just make it a preference setting!
Unfortunately, all experience shows that most users never bother changing the default settings. This means that the design team retains responsibility for picking the best option and making it the default. It also means that customization is a weak method for optimizing the user experience since only a small number of power users will avail themselves of the offer.
A final problem with customization is that users are not designers. They will often pick a poor choice that’s not, in fact, best for them. This often happens when color schemes are made customizable: the resulting choices often deliver poor legibility.
Personalization and customization differ in who is the agent making changes in the UI: in personalization, these changes are made by the computer without the need for the user to take any action, whereas in customization, the user must make the changes manually. This means that personalization can be applied much more widely than customization.
Unfortunately, personalization is also a poor way of optimizing the user experience for each user. The computer is often bad at guessing what users want, meaning that the items it delivers to personalize the UI do not match the user’s actual needs. This is very often the case when e-commerce sites try to recommend additional products to buy.
Personalization works best when it is clear what each user will want. For example, an online banking system can easily personalize screens with information about the current user’s account balance because it is well-defined which accounts each user owns once they have logged in.
TikTok is one of the few services to offer great personalization because it has highly accurate information about each user’s desires based on measuring to a fraction of a second when the user stops watching each video.
Personas are too broad to deliver an optimized user experience since each persona can encompass millions of users for a big website. Customization and personalization hold out the hope of delivering something more suited for each user, but both methods have sufficient weaknesses that this hope remains unfulfilled.
Enter generative AI: with this technology, the computer can generate a completely new user interface for each user. It is no longer a matter of plugging the user’s bank account balance into a specified location on an otherwise predesigned screen. The entire screen will be designed on the spot, based on the user’s needs. (And even though I just said, “screen,” the process would produce an audio interface for a blind user without ever rendering anything onto an unviewable screen.)
With individualization, we have AI generate a completely new and different user interface for each individual user. (Leonardo)
This degree of complete individualization is more of a future expectation than a current reality, because current generative AI is too slow to produce the UI in real time. However, real-time updates should be possible after a few more years of advances in AI. We already have several examples of AI-generated individualization for experiences that can be rendered in advance.
One generative AI system for medical record keeping sends an after-visit summary to the patients that describes any medical issues at a 4th-grade reading level. Since some research suggests that patients forget up to 80% of the information discussed in a clinical visit, having pertinent information in writing is a great advance. And having the AI write the information at a reading level that almost everybody can understand is also a boon. That said, some patients are capable of understanding information at a higher reading level that could transmit more nuance. After-visit medical summaries are a great example of the benefits of individualized user interfaces: it would be easy for the medical record system to include a field with each patient’s preferred reading level. For me, this might be a 16th-grade reading level. (I could cope with higher, being a victim of 21 years in the education system, given my Ph.D., but for everyday use, I prefer not having to work as hard as if I were reading a science paper.)
Another example of AI-driven individualization comes from Carvana which made 1.3 million unique AI-generated videos for their individual customers. Each video included location-specific clips showing each customer’s model of car driving in a scenic location close to their house. The voiceover included both the customer’s name and other tailored elements. Since a video commercial is not an interactive design, this example uses weak individualization relative to what I envision for future AI-generated user interfaces. (The system could generate 300,000 individual videos per hour, in a batch processing mode.) Still, this case study gives you an idea of how much can be done already, with the current weak AI technology.
Maybe this metaphorical example of different outfits from DallE is a bit overdone, but it exemplifies my ideal of having AI generate a completely different user interface for each user. Supergeek, corporateman, flowergirl — they should be treated differently.
Infographic Summarizing This Article
Feel free to copy or reuse this infographic, provided you give this URL as the source.