How to Use AI in UX/UI Design: VOYA’s Approach

A few lines in a design tool can now produce an interface in seconds. Polished screens, visual alternatives, and working prototypes are easier to create than ever.

But do those outputs become digital products that actually meet users’ needs?

That is where the real design work begins. There is a significant difference between drawing a payment screen and helping someone complete a payment with confidence. AI can speed up production. Understanding which problem the output solves, where it needs to work, and how to validate it remains part of the design team’s responsibility.

At VOYA, our approach to AI in UX/UI design brings user needs, business goals, and real product constraints into the same conversation. We see AI as a tool that supports thinking, helps us explore alternatives, and creates opportunities to learn.

What is AI-assisted design?

AI-assisted design means using artificial intelligence to support activities such as making sense of research, developing ideas, mapping user flows, exploring interface options, and building prototypes.

The term is often associated with image generation. In digital product design, its applications are much broader. Planning a SaaS onboarding experience, reviewing a complex transaction flow in a finance app, or improving form error messages can all involve AI-assisted design.

The important distinction is between generating an output and making a decision. AI can suggest an approach. A design process is still needed to assess whether that approach fits the user, the business model, and the technical environment.

How to use AI in UX/UI design

An effective AI-assisted UX/UI design process can be organized into six steps:

1. Define the user problem and the business goal.
2. Build context from research, the existing product, and technical constraints.
3. Use AI to explore alternative flows, content, and interface drafts.
4. Adapt the selected approach to the design system and product rules.
5. Evaluate the prototype with real users.
6. Measure product performance after implementation.

Each step will carry a different weight depending on the project. Designing a new product and improving an existing payment flow create different needs. Those needs should determine how and where AI is used.

VOYA’s approach: Understand, structure, design, and develop

We describe VOYA’s design approach in four stages: understand, structure, design, and develop. AI can support different activities throughout this process. At every stage, we need to know which question we are using it to answer.

1. Understand: Define the right problem first

“Let’s make the app look more modern” expresses an expectation. To turn that expectation into useful design direction, we need to understand the need behind it.

Are users struggling to complete a task? Are they missing the value the product offers? Or has the product outgrown a structure that can no longer support its needs?

Answering these questions early can matter more than the screens we produce. User interviews, analytics, support requests, and stakeholder conversations help us examine the problem from different perspectives.

AI can help organize this material. For example, appropriately prepared interview notes that are approved for use with the chosen tool can be reviewed for recurring themes. AI can also help identify open questions from a meeting or bring structure to a scattered brief.

An AI-generated summary, however, is not a research finding on its own. It needs to be checked against the source material. Conflicting statements should remain visible, and assumptions should be clearly distinguished from evidence. A model’s suggestion should never be treated as proof of a user need that the research has not established.

2. Structure: Map the flows before designing the screens

A digital product needs to support the paths people take to reach their goals. That means designing the relationships between steps as well as the appearance of individual screens.

In a money transfer flow, selecting a recipient, entering an amount, and confirming the transaction might initially seem sufficient. Yet insufficient funds, transaction limits, and a lost connection are also part of the experience.

AI can help outline alternative flows and potential scenarios, giving the team more situations to discuss. The necessary steps can then be clarified using user expectations, product rules, and input from the engineering team.

For fintech and SaaS products in particular, a shorter flow is not always a better flow. Giving users the information and control they need to understand the consequences of an action is also a design responsibility.

3. Design: Bring alternatives into the product’s design language

AI can quickly make different interface interpretations of an idea visible. Drafts that explore information hierarchy, content layout, or interaction patterns can make team discussions more concrete.

At this stage, we evaluate how clearly the interface works in use. Can the user identify the primary action? Does the information help them make a decision? Can they comfortably complete the same task on mobile?

The selected direction also needs to work within the product’s design system. Colors, typography, components, spacing, and interaction behavior should be considered together. A single AI-generated screen cannot be assumed to work consistently with the rest of the product.

We also consider loading, empty, error, success, and access-denied states as part of the design. The care given to these states contributes to the quality of the overall experience.

4. Develop: Validate the design and support implementation

A working prototype does not, by itself, establish that a solution is right for its users. Evaluating it requires realistic tasks and clear research questions.

Can users find the intended action? Do they misunderstand a term? Can they tell when a task is complete? Can they move forward without help?

AI can help draft tasks and checklists for these evaluations. Observing real users provides a different kind of evidence. Asking a model to behave like a user does not produce the evidence that usability testing provides.

The handoff to engineering should explain component behavior, exceptions, content rules, and interaction details alongside the screens. Once the product is live, it can be evaluated against the goals established at the start. Design decisions can then be revisited using evidence from actual use.

How to choose AI design tools

Tool selection should start with the task. There is no requirement to use the same tool throughout the process.

Tool capabilities will change. The underlying needs remain: being able to edit the output, trace decisions, and work within the existing workflow.

How to write an effective AI design prompt

A prompt such as “Design a modern finance app” offers little context for product-specific decisions. A more useful starting point describes the user, task, problem, constraints, and expected output.

The following example is for a hypothetical SaaS product:

We are designing a flow in a reporting product for multi-location businesses. A branch manager needs to review daily sales performance. Users should see their own branch’s data first, then compare it with the previous period. The current problem is that too many metrics are presented with equal priority. Prioritize mobile use and preserve the existing component structure. First, list missing information and assumptions. Then propose two alternative information hierarchies and explain which user need each addresses. Include states for no data, loading failures, and missing comparison data. Do not invent product rules that have not been provided.

The purpose of this prompt is to create a draft the team can evaluate, with visible reasoning and open questions. A perfect screen on the first attempt is not a realistic measure of its value.

Example: Using AI to improve a payment experience

The following is an illustrative scenario, rather than a client case study.

Imagine users are leaving a payment screen before completing their purchase. The immediate response might be to redesign the screen. But the cause could be that the total appears too late, delivery terms are unclear, or entered information disappears after an error.

We would first examine potential causes through existing data and observations of users. AI could help group approved, de-identified support records by theme or organize questions for further research.

The next step would be to develop possible solutions: showing the total earlier, placing an explanatory error message beside the relevant field, or preserving form data when a payment fails.

After evaluating these alternatives through a prototype, we would implement the selected solution and measure its impact. Alongside payment completion, we could examine error frequency, repeat attempts, and the need for support.

AI’s contribution here is to help the team make alternatives tangible more quickly. Evaluation and measurement establish which approach works.

How to measure the success of AI-assisted design

The number of screens produced is not a useful success metric on its own. More output can also create more review and rework.

When evaluating AI use, we ask two questions together: Has the process become more efficient? And does the resulting experience address the intended need?

On the process side, a team can track the time needed to prepare alternatives, the amount of revision required, and ambiguities that arise during development. On the product side, relevant measures might include task completion, error rates, abandonment at critical steps, and support requests.

The right metrics depend on the project. A baseline and a measurement method are also needed to compare results. We want to understand where and how an improvement happened, rather than simply attribute it to the use of AI.

What to watch for when designing with AI

One of the main risks is accepting the first output as a finished design. A convincing interface can conceal missing business rules and unsupported assumptions.

Research findings and AI suggestions need to remain distinguishable. A visually simple screen also needs to be assessed for accessibility and clarity. Keyboard interaction, focus order, contrast, and error feedback all belong in the design review.

Data handling should be considered from the outset. Appropriate access controls and tool terms need to be checked before using client information, personal data, or restricted documents. Building useful design context does not require uploading every available piece of information to a tool.

It is also worth considering the maintenance effort created by alternatives generated outside the design system. Producing a screen quickly does not guarantee that it will integrate smoothly into the wider product.

Using AI to design products and designing AI product experiences

Using AI within a design process and designing a product that includes AI raise different questions.

In the first case, AI is a tool used by the team. In the second, users interact with an AI system inside the product. Its capabilities, the information behind its outputs, and the control it gives users become central to the experience.

For example, when a report includes an AI-generated summary, can the user access the source data? Can they correct an inaccurate suggestion? Can they recognize when the system is uncertain?

In VOYA’s approach to AI product transformation, these questions deserve as much attention as the interface. An AI feature needs to provide a tangible user benefit and work as part of the overall product experience.

Frequently asked questions

Can AI be used for UX/UI design?

Yes. AI can support research organization, user flow exploration, interface drafts, copy, and prototypes. Product decisions still need to be assessed against source evidence, user needs, and technical constraints.

Do you need design skills to use AI design tools?

You may not need extensive design knowledge to generate an initial draft. Assessing its usability, consistency, and suitability for the product requires design expertise.

Will AI replace designers?

AI can automate parts of design work and change how teams work. Defining the problem, making trade-offs between competing needs, and validating the outcome remain responsibilities that someone needs to own.

How much time can AI save in the design process?

There is no single percentage that applies to every project. Time savings depend on the task, the quality of the inputs, and the revisions the output requires. Total effort should account for both production and review.

How does VOYA approach AI-assisted design?

VOYA approaches AI use through user needs, business goals, and product constraints. The aim is to evaluate alternatives more quickly across research and prototyping while giving design decisions a sound foundation.

Find the right starting point for your product

The value of AI-assisted design depends on the problem the output solves. Our starting question is which product need deserves attention, before deciding which tool to use.

At VOYA Digital, we bring research, UX/UI design, design systems, and prototyping together for fintech and SaaS products. Get in touch to evaluate your existing product experience or explore a new AI-powered experience.