AI and design in banking: customer experience, chatbots and trust

In banking, AI most often shows up in four parts of the customer experience: the support chatbot, personal tips in the app, fraud protection and simpler product applications. But customers judge the feeling of safety and control, not the model. So the key design task is to make AI transparent: show clearly where a bot is answering, offer a quick route to a person, require confirmation for any action involving money and explain decisions in plain language.

Financial services in Uzbekistan are moving quickly into mobile apps, and the number of automated interactions is growing with them. Here is how to design AI features that help customers rather than irritate or alarm them.

Where AI meets the customer

• Support chatbot. Answers frequent questions: limits, fees, transfer status, blocking a card, branch addresses. It takes load off the call centre and works around the clock.
• Personal tips. Spending categories, payment reminders, alerts about unusual charges and suggestions for suitable products.
• Fraud protection. Systems that notice atypical transactions, ask for confirmation and warn about suspicious recipients.
• Product applications. Document recognition, form autofill and preliminary application checks: anything that reduces the number of steps.
• Voice and text assistants inside the app that help people find the function they need.

In all these scenarios the technology is invisible to the customer. They see only the screen, the text and the buttons, so those decide whether AI feels like a helpful assistant or an incomprehensible black box.

Designing for trust: five principles

In banking the cost of error is higher than in most other services. Customers trust the app with their money, so any confusing behaviour is perceived as a threat. A few principles help avoid that.

1. Honesty about who is answering. Users should understand they are talking to a bot. Passing a bot off as a person destroys trust once the pretence is discovered.
2. A quick way to a human. A "Contact a specialist" button should be visible at every step, not appear after the fifth failed attempt.
3. Confirmation of money actions. A bot may prepare a transfer or change settings, but the customer confirms the final action personally, seeing the amount, recipient and fee.
4. Clear explanations. If a transaction is blocked or an application declined, the customer needs a reason in plain language and a next step, not an error code.
5. Predictability. The same actions should produce the same result. AI features must not change the interface or settings unexpectedly.

In a financial product customers forgive a slow answer, but not the feeling that the system is doing something with their money without their knowledge.

How to design a banking chatbot

• Buttons plus free text. Common scenarios as quick buttons, complex questions as text. This reduces misunderstandings.
• Clear boundaries. The bot says plainly what it can and cannot help with, and immediately offers an alternative.
• Answers linked to actions. Not just "you can change your limit in settings" but a button that goes straight to the right screen.
• Never ask for secrets. The bot never asks for a PIN, full card number or SMS code, and the interface itself should remind users of this.
• Context carried over at handoff. When the conversation moves to a specialist, the customer should not have to explain everything again.
• Two languages. Uzbek and Russian on equal terms, with the option to switch at any time.

Tone and copy

A bot's copy is as much a part of the design as colours and buttons. A good banking bot speaks briefly, calmly and to the point, does not joke in situations involving lost money and does not overload answers with conditions. Prepare a vocabulary in advance: how the bot greets, apologises, reports a delay and hands over to a specialist. Then answers sound consistent even when an AI model generates them.

We described the general principles of business chatbots in detail in our article on AI chatbots for business in Uzbekistan.

Explaining decisions

When an automated system takes part in decisions that matter to the customer, such as loan approval, a limit or a blocked transaction, the interface should help them understand what happened. Good design practice:

• show the main factors behind the decision in simple words, without technical detail;
• suggest what the customer can do: upload a document, confirm the transaction, visit a branch;
• offer a way to request a review involving a staff member;
• avoid wording that sounds like an accusation.

An example of a good message

Instead of "Transaction declined. Code 403", write: "We have paused this transfer because the recipient is new and the amount is higher than usual. If the transfer is yours, confirm it in one step. If you did not make this transaction, tap 'This wasn't me' and we will block the card." The customer understands the reason, sees two clear actions and feels the bank is on their side.

Decisions that significantly affect customers should remain open to human review. This is a matter not only of fairness but of reputation.

Security as part of the interface

Fraudsters increasingly rely on social engineering: calls from a "security department", fake messages and now synthetic voices and video. App design can become a line of defence:

• warnings written in human language before a transfer to a new recipient;
• a pause and extra confirmation for large or unusual transactions;
• a clear "How to spot a scammer" section inside the app;
• a consistent, recognisable style for official messages so fakes are easier to notice.

We wrote about how organisations can protect themselves from fake video and voices in our piece on deepfakes and brand protection.

Accessibility and different audiences

Banking apps are used by people of all ages and levels of digital experience. AI can help, for example by explaining a complex feature in simple words, but only if the core interface is clear too.

• large, readable text and sufficient contrast;
• plain wording without banking jargon;
• clear icons with labels;
• the ability to complete key scenarios without the chatbot.

For older users in particular, important buttons should stay in the same place, actions should be easy to undo, and when a mistake happens the system should guide them calmly and clearly.

Common mistakes

• A bot you cannot leave. The most common source of customer frustration.
• Pushy "personal" offers. If tips feel like advertising, customers stop trusting useful notifications too.
• Opaque rejections. "Transaction declined" with no reason or next step generates support calls and negativity.
• Different tone in different channels. The bot speaks one way, the app another, text messages a third. This weakens recognition and makes it harder to tell official messages from fakes.
• Launching without testing on real users. What is obvious to the team is often unclear to the customer.

Checklist for an AI feature in a financial product

• Users understand they are dealing with an automated system.
• A route to a person is available in one step.
• The customer personally confirms every money action.
• Decisions are explained in plain language with a clear next step.
• The interface reminds users which data the bank never asks for.
• The feature works equally well in Uzbek and Russian.
• The scenario has been tested with users of different ages and experience.

If you are designing a financial app or website and want AI features to strengthen trust, the Dasta Creative team can help with research and interface design as part of our web design service.