The future of UX: design that knows you better than you know yourself | Joana Cerejo

What does it mean to truly anticipate a user and not just what they'll click next...but what they're trying to become?

In Episode 19 of Human × Intelligent, Madalena Costa is joined by Joana Cerejo, design lead, AI product designer and author of the Anticipatory Design Playbook. Together, they explore the real depth of anticipatory design, how behavioral science fits into modern AI product work and why most systems fail not because of bad technology but because of a fundamental misunderstanding of human intent.

In this episode:
- The three layers of anticipation: needs, behavior and outcomes
- Why designing for agency can't be an afterthought
- Behavioral science frameworks every AI designer should know
- The filter bubble problem and collective manipulation
- What the Nest Thermostat gets wrong about resilient design
- Why transparency is the foundation of everything

Connect with Joana Cerejo:
→ LinkedIn: https://www.linkedin.com/in/jcerejo/
→ Website: https://jcerejo.com/
→ The Anticipatory Design Playbook (Amazon): https://www.amazon.es/-/pt/dp/1041079109
→ Watch Why Personas Fail AI (And What Works): https://www.youtube.com/watch?v=_7dSuJB6M1o&t=897s

Human × Intelligent is a podcast at the intersection of design, AI and human agency. Hosted by Madalena Costa.
→ humanxintelligent.com
→ https://www.instagram.com/humanxintelligent/
→ https://www.linkedin.com/company/human-x-intelligent/
→ https://www.instagram.com/designwithmaddie/
→ https://www.linkedin.com/in/madalenafigueirasdacosta/

Guest bio
Joana Cerejo is a design lead and AI product designer working at the intersection of user experience, behavioral science, and intelligent systems. With nearly a decade of experience designing AI-powered products across fintech, e-learning, and manufacturing, she specializes in making systems that are human-centered, trustworthy, and ethically grounded. She is the author of the Anticipatory Design Playbook, exploring how AI can move beyond predicting behavior to genuinely supporting people in meaningful, long-term ways.

Resources & tools section
Frameworks mentioned in this episode:
→ Prochaska Transtheoretical Model - stages of behavioral change; helps design systems that meet users where they actually are
→ Fogg Behavior Model - behavior happens when motivation, ability, and prompt align at the same time
→ Nudge Theory - the right intervention at the right moment can make or break a service

Book:
The Anticipatory Design Playbook by Joana Cerejo - available on Amazon

Concepts to explore further:
→ Filter bubble effect
→ Human-in-the-loop design
→ Foresight/futures thinking methodology
→ AI literacy and explainability

Transcript

SPEAKER_00 0:07

Today we are joined by Joanna Sterejo, a design lead and AI product designer, working at the intersection of user experience, behavioral science, and intelligence systems. She spent nearly a decade designing AI products across industries like FinTech, e-learning, and manufacturing, always with a strong focus on making these systems more human-centered, trustworthy, and ethically grounded. Joanne is also the author of the Anticipatory Design Playbook, where she explores how we can design AI systems that don't just predict behavior but truly support people in meaningful ways. In this conversation, we are diving into one of the most complex questions in AI product design today. What it really means to anticipate users. So welcome to HumanX Intelligent.

SPEAKER_01 0:51

Thank you, Madalena, for your warm welcome.

SPEAKER_00 0:54

Thank you. So when we talk about anticipating users in AI products today, what are we actually anticipating? Needs, behaviors, or outcomes?

SPEAKER_01 1:05

Well, in fact, in fact, all of the three, but they operate in uh different layers. And most of the products only address the surface level per se. For example, NEET, um anticipating the NEET is the most calming approach. It means predicting what information or action the user might want to do next. This is where most of recommendation engines live. And then you have the behavior anticipation layer. This uh goes a little bit deeper. It means try to understand the patterns in how people interact with the system, including implicit clues like browsing history, timing, all kinds of uh historical data, not just explit comments uh between the user and the system. And then you have the outcomes, like you mentioned. This is the most ambitious layer under anticipation and also the least common because it's really hard to achieve this properly. Um and it means helping the users to to try to help the user reach their desired future state, like uh wanting a new skill or improve uh um or improve something in their life or gaining a new habit, something like that. And this is where the anticipatory design shift from reactive to proactive, because the system will try to help the user move towards something to move the user toward um a specific outcome. This is the real shift from designing for tasks to designing outcomes. Because as designers we shift from asking how can the user sign up for a gym, for example, to how can we help the users build a more healthy habit or be more fitness, something like that. So we switch from the task to the outcome. Today most products they claim to be anticipated, but mostly what they are, they are good at optimizing for engagement metrics. But through anticipation, it means understanding the user intent, what the user is trying to accomplish, and then designing this call behavior, what kind of actions the system needs to do to help the user achieve that future goal.

SPEAKER_00 3:45

Yeah, for sure. And basically, when teams say we are anticipating users, it sounds com kind of or almost magical, right? Like the system is or knows what you want before you do, which it's kind of weird, but that raises a deeper tension. Are we predicting what users will do, what they need, or what we want them to become? And like this ambiguity is where everything starts to unravel, right? I would say, in this sense, and if anticipation isn't just prediction, then it's something much more powerful in my opinion, and more dangerous, or it can be more dangerous, right? Because it shapes the experience and not just response to the to the user, right? Which leads, I think, we to a critical threshold, which is at what point does anticipation stop being support and start becoming control?

SPEAKER_01 4:35

That's a very good uh question. In my perspective, when the system acts without the user understanding or explicit control of the user, that's when we start creating systems that um can cross the fragile line between uh useful and becoming uh cultural uh controlled. And I think the the designer responsibility is to be very aware of this and design these control levels um in the system design itself. Be very conscious where we are removing the user to control in the the service we are trying to automate, but even so design guardrails so the user can overcome the the service automation anytime they they feel they need to be in control. Yeah. If we remove the user from the from this control level, that's when most of the system starts to felt creepy and uh that we don't control the the solution. And this level of control will be critical depending on the industry where you are designing from, because it will depend on the level of the risk trade-off. If it's entertainment, things can can be okay, you remove control, but the impact that can have on the human being is not that dramatic. But when you start designing for service in financial and health, the trade-off is really, really important. So we need to learn how to design this human in the loop mechanism in our systems. We can automate to a really high level with anticipation, but we need to design these mechanisms into the system so the user feels in control. Otherwise, the system will feel creepy, and in the end you can have the best algorithm in the world, the best user interface, but if your user don't trust your system, they will not use it, and in the end of the day, you don't have a business. So the human in the loop design is extremely important for this highly automated uh solution.

SPEAKER_00 6:58

It's like uh it's like at first anticipation is very helpful or feels very helpful, like reducing friction, saving time, and so on. But it at the end or gradually it can actually become like narrow choices or like a guided behavior. Like it's not uh we are not doing the behavior, we are being set up for something in that sense, and even like make the decisions for the user. So I think the the real design challenge would be or it becomes like how do the how do you design anticipation in a way that still preserves this user agency.

SPEAKER_01 7:35

For example, there are many studies and reports now saying that most of the AI solutions they fail when they reach the market. Mostly they failed because they are not looking to the actual user problems. They are trying to put AI as a sophistic sophisticated layer over already fundamental broken systems. And this will touch in the agency because it cannot be thinked as an afterthought or look to agency like one more feature that we need to put on top of the system. If we do so, the probability of the system fell will be higher. So we have to design for agency since the beginning of the process. And the way I do that is looking to um behavior science frameworks. They are they exist for many decades. I think it's the missing link between what already exists and to what we need to bring to our uh design process for the the AI era. And these frameworks are already proven proven to be very, very useful. The three that I like the most is the Praska Transtheoretical model. They say that people progress through states of changes, meaning every user that signs in in a service, they're ready for the change or their behavior pattern will be in these different uh levels. So we have to prepare the system to automate the anticipate the experience to meet where these users are. And most of today's AI don't do that. They just think in the moment where all the users are in action mode, interacting with the system. They don't think uh about the other stages where the user can come from. Another framework is the frog uh frog behavior uh model. He explained that behavior only occurs if you align three things at the same time, they cannot happen separately, which is the motivation, ability, and prompt at the right time. Um and if these thing elements don't converge at the same time, behavior will not happen. And many times what I see in many AI systems um is that they lack the understanding of the motivation of the user to use the system. They focus too much on the ability, make the technology possible for the user to interact with the system, and most of the time they forget how timely interventions are so important for the success of the solution because it can annoy the user or make them feel generic so the user don't feel that hype uh hyperpersonalization they are hoping to have with the AI. So the wrong prompt at the wrong time easily breaks the expectations and the user mental model have over the an AI system. And the last one that touched to this last part of the behavior model is the nudge theory that highlights that the right intervention at the right moment can break or make your um uh service successful. So this practically means that um we need to start designing for these behaviors, how the user expects to interact with a supposed intelligent uh system. And in my book, I I put this thought into a framework in a way that intent is the key element, what triggers everything in our design process. So the the intent is the core without understanding the user intent and the behavior. Workflows, I'm sorry, workflows in algorithms cannot meet the user needs at the end of the day.

SPEAKER_00 11:55

Or um how would how do you how would you say professionals can add these frameworks or these small nudges in their products in a roadmap? How would they justify, for example, adding to a roadmap and how would you they frame it for a road?

SPEAKER_01 12:12

Like we start this conversation. Uh intent and agency they need to be designing from the beginning, cannot be an afterthought. And like agency, intent can isn't a feature to to add uh to a broken system. It has to live across the entire user journey. If you treat anticipation like a feature, you are over relying on the data to make predictions. And human beings are everything but not predictable. This is the the layer that I see where most of the AI solutions today broke. And the way to incorporate these thoughts is to look to your design uh process and try to break it into three big moments. First, look as uh with an anticipation lens, what changings uh the user uh is trying to achieve, create his uh their context. Then you move to a lens when you start to imagine how many possible features the user can have that will impact their journey towards that future oriented uh goal. And in the end you look to the shape lens, starting to use those behavior science frameworks and other technologies from foresight and apply them on how to build a system that can adapt and preserve the user agency over time. And if you break your process between anticipation and imagination, possible futures, and then starting to converge and shaping your solutions towards the disalignment, you will have a more resilient user experience and service in the end. A good example for us is uh for example the the Figma make. AI can make things visible very, very easy. They will do exactly what you ask. You will prompt Figma make, he will generate what you do, what you want. But Figma Mate cannot design user intent. That understanding and that research need to come from you. Otherwise you can have a very polished design and uh prototype, but empty in context for your user.

SPEAKER_00 14:41

It's like uh it's like we are building something with them, so it's like a colleague, but even though in that I really enjoyed when you said a future this is not a future, it's something to add it it would be something adding to a broken system. I find that very intriguing. You would say that it's something that everyone or every UX designer, product designer should have in the back of their head and should always implement to whatever feature, whatever job they're taking in that moment, right?

SPEAKER_01 15:11

Yes, because today most of the the companies they overrely on the data. Everything is about data. You use historical data and present data to predict the behavior of your user. But the problem is we are living in such a fast pace that the social aspect, economy, um technology itself will impact your user in ways that your system needs to be prepared to scale and adapt to evolving needs of the user. Most of the systems don't do that today. A very good example that I like to to share often is the NES thermostat. When he was launched, it was the best example of anticipatory design. Everyone mentioned this as the the gold standard for an anticipatory design experience. This was so good that even Google acquired them for a bunch of money. But the problem with these systems, although they they overrely on the data, they can learn your habits for a year, but they have a really difficult time in scaling for different scenarios. So they work on probability, not what's profitable for the user. And that's why we need to start using more methods and tools from foresight and start to think of possible future scenarios. And a very easy example is if you have a newborn, he joins the house with the house, in that day the system will break because it is not ready to scale for a different scenario. He was trained for a year of usage data that no longer meets the context of the user. I think if the team starts to think more about possible future scenarios and incorporate them in the algorithm, saying if this happened, if in creating several um possible uh scenarios, we can create more resilient user experience that can scale with the user context and intent.

SPEAKER_00 17:32

What would you say to those people that say that humans are very predictable and they don't change and we are not that that complex? What would you say to them? Why do you think that it's important to go with evolvement and scalability and growth with humans? Because we are always changing, you know, there's not enough patterns or yeah.

SPEAKER_01 17:54

If you look to the latest reports, most of the solutions that fail is because of that. And for example, in the GI space, more than 90% of solutions they fail because of that. The technology is really cool, but people are not understanding the use cases where the humans would want to use that. And I think that's the main factor why most of the solutions fail. We are failing to understand the the behavior, the relationship of the human with something that is so high so intelligent and has so much high automation capabilities. And the studies are supporting that uh that vision that there is a disconnection. Um datum is really good for complex but stable systems where things can be highly automated because they don't change, so AI has been very, very successful in B um B2C context. But moving to B2B, things starts to getting more uh more comp not complex but complicated because we are missing the the human nuance. And this leads to another interesting point. You can design systems where people overtrust or distrust completely your uh your solution. So it's not just designing for transparency and control, but also for um yes, uh it's also about designing for transparency and control because the risks of a person overtrust your systems could have a really big impact on their well-being. And the same of the opposite if they don't trust and neglect some um some signals of the system also could lead to catastrophic um outcomes. So I I cannot say how much important it is to start incorporating more behavior science in our design process when we are designing for intelligent systems. It will be really the key to make them more ethical and more proactive. And if we want the systems to be long-term successful, the key is to bring the foresight methodology to try to make them more resilient over time.

SPEAKER_00 20:27

And considering what you just said, where would you draw the line between the prediction and manipulation in artificial intelligence systems?

SPEAKER_01 20:36

Like I mentioned before, the line is in the design in the agency, the control, and all come together to the intent alignment with um the user intent alignment. Because prediction serves the the standard user goals and manipulations they will serve the system hidden objectives. And if you don't design um a transparent system, the user will not see how much they can be uh manipulated. Many times they overshare data um and that it becomes the um coin for the business for other purpose that the user has no idea. I can add another thought on the on the manipulation level, which is the the filter bubble problem. Um is a concept that the algorithms are conditioning the vision and the experience of the users. They will be fit what they want to see or seen in the past. So the over reliance on the the the past data. They are looking for what's probable because based on what the user saw in the past and not what is prefable, what the user might want to see in the future. And this is currently this is a really big issue, and especially in the social media space. And entertainment, people are not making no longer big literature and drama because the content has to be clickbait. So this is putting in question even the quality of humanity sense of art and philosophy, because everything has to be optimized for an algorithm. Recently I read even an article about that in Spotify that the albums are not done anymore for the test of time because songs have to be clickbait and hook you in the Spotify loop. So this is a very big problem and manipulation to a system level that I think most of people are not aware of.

SPEAKER_00 22:57

It's like the different the it's like the prediction informs, but the manipulation like pushes people or pushes the outcome in a preferred direction, and that is not very very good sometimes.

SPEAKER_01 23:10

And it's doing that collectivically, not just to an individual level, which is extremely dark.

SPEAKER_00 23:17

Yeah. But so you but that begs the question, like when that line is crossed, like where does the accountability go, right? Who is responsible when this anticipatory system influences user behavior, groups of user behavior negatively, right?

SPEAKER_01 23:36

We didn't mention accountability yet. It's so critical. Is one of the seven dimensions I uh identify for measuring AI user experience along with the trust, transparency, agency, fairness, privacy, and uh, this goes on. But the short answer is the I think organizations that deploy systems, they are responsible for the accountability of the system. But this question goes very, very deep. But at least what we as designers can do is to try to create uh safety mechanisms or feedback mechanisms in our systems that uh at least uh improve the this accountability uh layer. Users need to know that they are they are people, not just a number to to the systems. And this means creating, like I mentioned, uh the feedback loops, try to bring uh ethic patterns and responsibility structures to our user experience. And the way to do that is for the users to gain some machine learning, deep learning and AI literacy, otherwise they will not know how to influence the development toward these ethical and accountability and explainability aspects of uh It's like uh because these systems aren't neutral, they like embed decisions made by designers, companies, and incentives, right?

SPEAKER_00 25:12

So ultimately, I guess all of this kind of leads to one defining question is like if teams want to design anticipatory systems, they're actually human-centered. What is the one principle they should not compromise on?

SPEAKER_01 25:28

For me is transparency 100%, because transparency will hold everything. Transparency is the base to build trust. You you can't have a trustworthy solution without uh transparency. And transparency will be the open for creating user agency and interport user intent. For me, it's for sure the foundation of everything where we take on. And uh I think we need more, maybe I don't know, heuristics or uh structured ways to outdate transparency in our design process.

SPEAKER_00 26:09

I guess after everything like prediction, control, adaptation, transparency, visibility, responsibility, the core principle isn't about like the intelligence or the futures or whatever it is. It's actually about respecting the user's ability to choose, understand, and change. At least this is what I gathered from everything that you shared, right?

SPEAKER_01 26:29

And because um if you get transparency right, everything else will come. The trust, the user agency, the fairness, the accountability, because you have a solid foundation to to do the rest well.

SPEAKER_00 26:43

Thank you so much. I have one last question before you before you go. And I would say for everyone that is listening, how would you say that they could start learning more about anticip anticipatory systems?

SPEAKER_01 26:55

Of course, getting your book number one, but besides that Yes, be curious and look for what behavior science already discovered decades ago. We are not uh inventing the wheel, we are just figuring out that our old design process has flaws and gaps for this new area, so we need to be curious and look to other areas of knowledge. Behavior science and foresight, also known as uh feature thinking, are the best ways to do that. In their methodology experiment. I'm suggesting to use this tool or use that, but what's today work is working, maybe tomorrow is not. So everything is new, everything is evolving. So what we need to do is experiment and uh feel faster so we have feedback of what is working and what is not. So that's my best advice today is to don't have a box mindset. If um we always done the design uh process in this way, that's why we need to do no being open-minded, try to grab knowledge from other fields of knowledge and bring them to your design process.

SPEAKER_00 28:09

It's basically if anticipation is about shaping what comes next, then designers aren't just like building products, they're shaping behavior, choices, and futures. And I love that you said go grab from other places, because we forget that there's so much knowledge in other people, other people's brains that develop from construction, whatever it is. So it's very good. Thank you for that. Thank you, Joanet. Thank you so much for for challenging us to think more critically and specifically about responsibility in what we are doing today, because we are doing for users, but users are people.

SPEAKER_01 28:43

So yes, and we are building systems that are influence so much people's life that we need to be more mindful and design for accountability and fairness. So there's a lot to do in these times. It's exciting.

SPEAKER_00 29:01

I think it's exciting. So thank you so much. Thank you to you, and thank you to everyone that is listening. And please contact Joanna on LinkedIn. But yeah, thank you. Thank you so much for the invite. It was a pleasure.