In a highly competitive fintech environment, user and customer experience are becoming key drivers of business growth. For microfinance organizations and banks, the transparency of conditions, the convenience of digital channels, and intuitive product application scenarios directly impact audience loyalty and final conversion rates.
A standard audit of complex banking ecosystems requires significant time for routine screen analysis. To optimize this process and focus on finding non-obvious insights, we at Rocket Tech applied a hybrid approach: we combined the deep expertise of a UX designer and a UX researcher with the analytical power of a large language model (LLM). In this case study, we will share how this method helped us conduct an audit for KMF.
Client and Business Objective
KMF is a major player in the microcredit market. As part of the project, our goal was to improve the efficiency of the kmf.kz website as an acquisition and conversion channel by identifying and prioritizing UX/UI issues.
Key areas for improvement included:
- Increasing applications for products (loans and deposits).
- Reducing drop-offs in forms and calculators.
- Improving the understanding of product terms and the transparency of offers.
- Enhancing navigation convenience and the mobile user experience (mobile-first).
- Strengthening brand trust and the perception of the bank in the digital channel.
Approach: Synergy of UX Specialists and LLM
To work on the project, we took the path of targeted team reinforcement. The UX researcher and UX designer integrated a dedicated LLM, entrusting it with the role of a powerful assistant for initial scoring and interface comparison.
The audit methodology was built on the following steps:
- Compiling a list of heuristic rules and criteria for evaluating UX/CX.
- Taking screenshots of the necessary KMF and competitor interfaces.
- Checking KMF screenshots against the list of UX/CX evaluation criteria using the LLM.
- Checking competitor interface screenshots against the list of UX/CX evaluation criteria using the LLM.
- Highlighting KMF's issues and competitors' good UX/CX solutions using the neural network.
- Manual re-verification of interfaces in dynamics and the LLM results by our specialists to filter out false positives and account for context.
- Compiling final recommendations for KMF.
Key Research Insights
The combination of LLM analytics and manual expert validation allowed us to abstract from specifics and identify systemic patterns in user scenarios. We classified the identified growth areas by their degree of impact on the final conversion.
Critical Barriers in Funnels (High Criticality)
These are scenarios where the user faces an insurmountable obstacle and cannot complete the target action independently. Our hybrid approach helped identify:
- "Blind" validation zones: Situations where the system responds to user actions with an error message but does not explain the reasons for their appearance or provide recommendations for fixing them. We proposed a framework for clear feedback that reduces the percentage of abandoned applications.
- Issues with the Kazakh-language version of the interface: The audit highlighted local gaps in the consistency of interface translations, which could negatively affect the trust of the audience that prefers the state language.
- Technical blockers: Failures were discovered in the loading logic of key CTA elements, which interrupted the user's path to installing the app and applying for a product.
User Experience Gaps (Medium Criticality)
Shortcomings that do not completely block the process but increase cognitive load and force the user to expend extra effort:
- Omnichannel experience gaps: We found points where the client's transition from the web to the mobile app looked like a dead end and offered recommendations for correction.
- Non-obvious UI patterns: The neural network and experts drew attention to interface elements (for example, in calculators) whose visual state was misleading and did not correspond to the actual functionality.
- Reduced control in complex scenarios: In multi-step processes (for example, when applying for a loan), users lacked transparency and an understanding of their progress. We recommended implementing contextual tooltips to reduce anxiety when applying for financial products.
Business Outcomes
Using a dedicated LLM as a scoring tool allowed the Rocket Tech team to multiply the speed of analyzing dozens of screens of the KMF website and app, as well as competitor interfaces. Because the routine checking of heuristics fell on the shoulders of the neural network, the UX researcher and UX designer were able to devote maximum attention to high-quality manual verification of hypotheses in dynamics.
As a result, the client received not just a list of visual edits, but a structured framework of recommendations based on data and advanced fintech UX practices. Implementing these changes will allow KMF to eliminate barriers in conversion funnels, make financial products clearer for the user, and strengthen the company's digital leadership in the market.
