Angélique Wynants is Senior Director, End User at EliseAI, where she’s building the company’s first dedicated end-user experience for renters on a platform designed primarily for property management companies. Her career has centered on designing for people navigating high-stakes, unfamiliar systems. At Nubank, she built for people getting a bank account for the first time in Mexico City, and at Comun, she led operations for a digital bank built specifically for undocumented immigrants. She studied economics at the London School of Economics.
In this conversation, Angélique shares the three principles she returns to when designing for the underserved and never-before-served: education, simplicity, and support. She talks through the assumptions that trip up well-intentioned product teams, the community events that replaced traditional user research, and a security feature that tanked adoption the moment her team stopped thinking from the user’s point of view. She also dives into what it’s like building EliseAI’s first B2C product inside a B2B company and where AI belongs in a support experience built for people with the least room for error.
Building for the underserved
Who are the end users you’ve built for at Nubank, Comun, and now EliseAI?
In the first part of my tech career, at Nubank and Comun, I was building for end users who were new to financial services and digital financial services. These were people who were getting a credit card or a bank account for the first time, or sending a remittance to a different country for the first time, or who were unfamiliar with using apps for banking needs. At Elise, I look after the experience of those going through the experience of finding a home, which is, I’d say, equally critical as financial services.
Here, it’s making this experience intuitive for the full spectrum of end users. This could be affordable housing applicants who are going through the Elise flow for the first time trying to apply for a home, or who have never used technology for any of their housing needs. Their needs are similar — both are going through a highly stressful moment in their life where they’re sending money to someone, getting a loan, applying for an apartment. It’s been a nice thread in my career, building for that subset.
Comun was specifically focused on undocumented immigrants. It was a digital bank for undocumented immigrants. It served everyone, really, but because the product enables people to get a bank account without a Social Security Number, it particularly catered to undocumented immigrants. The app was also fully in English and Spanish, so it catered even more to Latino undocumented immigrants. At Nubank, back in Mexico City, they were predominantly people from Mexico but who had never had a bank account before and previously were maybe saving under their mattress, and for the first time wanted to use more formal financial services.
Three guiding principles: Education, simplicity, and support
What principles matter most when you’re designing for users with limited experience or comfort with technology?
Typically, I always keep three principles in mind when I build for this type of end user.
The first one is education. I don’t assume any prior knowledge of the product. This could be explaining how interest rates work on a credit card, which you and I would know, but that’s not at all intuitive to someone who’s never owned a credit card before. Or being transparent about how income requirements work in housing, like someone submitting their income and not understanding why they’re below certain thresholds. Things like linking FAQs, educational videos, a UI wording that is educative is very, very important.
The second one is simplicity. Less is more. The UI needs to be really simple, straight to the point. I always try to make sure we minimize the amount of screens to complete a flow, that the browsing is simple, that every step is over-explained in simple terms — again, not assuming that this person has ever gone through a flow like this before.
The third one, which I think is really important and often overlooked, is support. I’ve done a lot of support work also in my career, so it’s very important to always give them a way to reach out for help. The first layer of support is increasingly AI, obviously, but they should never be stuck in the loop of either the technology not being able to help them, or not knowing where to look for help. This is very important, that there is an option for support, AI or human.
Where product teams go wrong
What do product teams tend to get wrong when they design for users whose financial circumstances, language skills, or digital literacy differ from their own?
This is one of the most surprising aspects of building for low-income or low-tech-sophistication users. Across teams that I’ve led and worked with, I always see two assumptions that people repeatedly make. The first: we tend to assume a lot of previous knowledge about a product or process, and someone who’s never used technology before, or is new to a system, a country or a process, may not necessarily know. One example is when applying for an apartment, what is a guarantor? When I moved to the U.S., I wasn’t familiar with what that meant, why I needed it, what they needed to show — whereas that’s a very intuitive concept for anyone who’s applied for an apartment previously. Never assume previous knowledge.
The second: teams assume a level of trust in technology that isn’t actually adapted to the realities of the end user. There’s a political scientist I really like called Francis Fukuyama, and he writes about low-trust and high-trust economies — how certain countries or communities operate in a fundamentally low-trust environment. Their trust circle is smaller; they tend to distrust larger organizations, institutions, new technologies. But people working in tech inherently trust tech. They don’t necessarily think that if you’re asking someone to input their bank information on a mobile, this person is going to think they’re being scammed, because that’s not going to be their first intuition. But that is the first intuition of a lot of people living in lower-trust communities, in which they often get scammed, for instance. The credit card is an obvious one — there’s a lot of social proof that we need to build to increase product adoption. In FinTech there’s a lot of things we can do. I think in housing as well, there’s a lot of things we can do to build trust, but that’s very often overlooked.
Can you walk us through a product decision that changed once your team better understood how the user was actually navigating their experience?
One example from one of my previous jobs is that we wanted to make the bank account connection more secure. We wanted to enable people to connect their primary bank account so that they could pull the money from the other bank account and then make payments. One way of doing that, which is quite common in the U.S. banking sector, is what we call micro verification. It says, “Hey, I sent you two cents on your primary bank account.” Now you go to the primary bank account, you see the two cents, and then you have to enter the amount. Most people working in tech probably have gone through that experience before, using digital banking services. We implemented that, and our product adoption completely dropped. People were very confused. They didn’t know how to use it.
They distrusted the fact that they had to enter their other bank account information. They weren’t sure how to go back and forth from one app to the other. So we had to quickly find another solution: building a very sophisticated risk score, so we’d only add this friction for very high-risk users who may be fraudulent, and find other ways to do verification for everyone else. Most of the undocumented immigrants we were working with — this was their first time using a digital service, and now we were asking them to open three apps at once and do something called micro verification. It just wasn’t viable. For the ones who did have to continue with that process, we also built educational videos — step-by-step, why it’s important, and how long it will take.
Getting inside the user’s world
Is there anything you do in particular to help your teams get in the mindset of the user?
I think user research is really important. There is traditional user research, but there is also non-traditional user research — going to the places where the target user lives and works or applies for an apartment, talking to them, seeing them using apps and technology, understanding what they’re afraid of. It also, frankly, requires a level of empathy that is really important. When building teams that build for the end user, it requires a different and, I think, higher level of empathy to be able to put yourself in someone else’s shoes, in a life that you have never lived and probably will never live.
What have you done to get meaningful feedback when traditional user research doesn’t work?
This is interesting because when I started at Comun, which is the digital bank that builds for undocumented immigrants, I tried conducting regular user research calls. I would reach out to our users and try to schedule a call with them. Friday, 2:00 PM — they would say, “That’s great.” I would send a calendar invite, and then I would connect at Friday, 2:00 PM, and no one would connect. This would happen repeatedly. At first, I didn’t understand why. So then what I tried to do is just call their number, a few different times throughout the day — obviously, only when I had their approval to reach out to them.
I realized that they ended up picking up at different times of the day, and were very happy to talk to me, but they just happened to be in their car commuting, or they’d just gotten home. A lot of them don’t have traditional working hours — different shifts, night shifts, construction sites, cleaning services — and they don’t know where they’re going to work the next day, or what time. So even though they wanted to help and wanted to talk to us, we couldn’t stick to a very specific schedule. What I realized is that what we really had to do is go and meet them where they are, in the spaces where they live, work, or hang out.
At Comun, because a lot of our users were from the Latino population, we would go to local community events and have stands, and talk to people there — “Hey, have you heard of Comun? Tell us more about how you send money back to Colombia. Tell us more, what bank do you use? Do you have a bank, or do you use your husband’s bank?” Because a lot of the time we also realized they share the accounts, for instance. How do they get paid? Those informal conversations, in person and at community events, ended up being the most relevant for knowing what to build and what the pain points were — a lot more than phone conversations, which were very difficult to get.
Designing for high-stakes decisions: Money and housing
EliseAI is a B2B platform for property management companies and you’ve been brought in to build the B2C end user experience for the renter. How do you balance the needs of both, especially when they have different priorities?
This is a very interesting question because there are very, very tough trade-offs between prioritizing B2C and B2B — something I’ve experienced for the first time joining Elise, because it’s the first time I’m joining a company that’s B2B. When working at a company that’s purely B2C, the end user is the client; the revenue of the company depends on whether the end user likes the product or not. That’s very different from B2B, in which we’re selling to businesses, and the client is not the end user. Elise’s platform is built for B2B, so they remain the main client. So when there are very urgent B2C issues, this is always prioritized, because in a way it can also impact our client. If someone misses out on a unit — Elise works with some of the largest property managers in the country, some with buildings in very remote areas — that unit can be vacant for a couple of months. In that way, their issues overlap.
But a lot of the property managers also have feature requests that have no impact whatsoever on the end user. That’s when it sometimes conflicts: should we improve the end user-facing platform, making the flow more educational or simple, or improve a feature within the B2B-facing app so property managers can more easily understand what’s pending on their side? That’s difficult, because the property managers are the clients — that weighs heavily in the prioritization.
AI, education, and trust
When users need to understand things like credit scores, income requirements, or why an application was denied, where should education live within the product experience? And how does AI adoption differ across your end users?
This has changed quite a bit with AI and is still changing. Previous to AI, when I was at Nubank, there was a typical help section, which would have the FAQs, educational videos, and we would use LLMs to maybe customize what FAQ would show up first. We would also link FAQs and videos throughout the UI, which is still something that I try to push teams to do now as well.
But now with AI, the new and better way to do it, in my opinion, is to have an embedded AI assistant that is constantly visible throughout, desktop or mobile. Instead of going to search the help section and having to press a back button and then coming back to the step where they’re struggling, there’s always an AI bubble that’s there, like, “Hey, I’m not sure what that step is, or what ID can I upload? or why is this failing?” And so they have a personal assistant that’s helping them throughout the process. Another interesting thing to explore too is other channels outside of the product — WhatsApp works very well for some communities. If they submit a support ticket in the app, or they’re talking to the AI and they exit the app, they just won’t go back to it. On WhatsApp, they actually respond immediately.
It’s a mix. The trust issue is definitely there, but what’s interesting is that even when we have humans responding, we are now being asked if it’s AI — the humans also have a certain tone of voice. It’s hard now to differentiate when it’s a human or when it’s AI. I would say for lower income populations, or even immigrants who don’t necessarily speak English well, long texts don’t work at all. So it’s either a short text, or some of them want to call. What I’ve done in the past is making sure we send videos — a series of how-to videos on YouTube — so when someone’s like, “I still don’t understand, that’s too long,” we just send a link, and then they get it that way.
My philosophy when it comes to support is that they should never be stuck in a loop of no help. So we have a support team — now at Elise it’s 7:00 AM to midnight — where if the first layer of AI is not able to help solve the issue, it escalates to a team that is trained to respond to those questions.
How do you apply your guiding principles—education, simplicity, and support—when deciding between using an AI assistant versus keeping a human in the loop?
Education is educating the AI to use an educational tone. I actually feed the AI a tone of voice, which is the same one that I use to train the support team: empathetic, straight to the point, avoiding extra words so that the text is not too long. I literally have dos and don’ts, such as do not say, “Hi, thank you so much for contacting us.” No, it’s, “Hi, I saw that this is your issue. This is how we’re going to solve it. This is what you can do next,” in a very empathetic tone, but we don’t want three paragraphs. So it’s training the AI to be educational, and even the prompt assumes no prior knowledge, and explains in a simple, straight-to-the-point way. That also works for simplicity and support. The AI needs to be escalating when it doesn’t understand or it’s not able to help the user, instead of hallucinating or giving answers that aren’t there.
Even though it’s the same underlying technology, the prompts we use for AI responses to B2C users are completely different from the prompts we use for our B2B customers or our internal tech team. The knowledge is different, the guidance is different, the prompts are different.
As AI moves into essential services, how do you make sure it removes barriers rather than creating new ones?
I think the knowledge feeding the AI is really crucial to make sure AI is an enabler rather than a barrier. This technology is very, very powerful and we can model it however we’d like. The type of knowledge that educates the user is truly important, so that it becomes an educational tool, rather than something that’s confusing or making it even more complex for them to understand.
I also think about enabling AI to decision, as opposed to just a chatbot. AI can make faster decisions. So once it takes action, it contributes to removing barriers, which are actually often human bottlenecks — like decisioning on an application or on a credit card, or making personal information changes to a bank account that a bank clerk would otherwise make, or a few hours waiting at a bank branch. The speed of AI itself can remove barriers rather than creating new ones. And, obviously, we need guardrails in place. Having a very strong product QA in place is really important — I always have a team doing product QA and human QA. And we always have a way for the end user to escalate any issues, so that if the AI isn’t being an enabler and rather is being a barrier, we can catch that and iterate on it.
What does LogRocket do?
LogRocket’s Galileo AI watches user sessions for you and surfaces the technical and usability issues holding back your web and mobile apps. Understand where your users are struggling by trying it for free at LogRocket.com.