Sustainable AI for data quality in finance
Exploring sustainable AI use to improve data quality for Triodos Bank.UX Research • UX DesignScenario
Financial institutions increasingly rely on AI to handle large volumes of customer and business data. While this improves efficiency, it also introduces serious risks due to its black box nature. I collaborated with Triodos Bank to explore how AI can support data quality without undermining key values in sustainability, transparency, and human control. Through qualitative research, iterative prototyping and literature research, we collaborated on a proof-of-concept leading to more sustainable data processing, combined with strengthened human agency. This is presented in a high-fidelity prototype built on a scalable design system.
Discoveries
During the process, we discovered that traditional OCR methods actually consume more energy than a Visual Learning Model. Using Google Gemma 3, we've developed a method that reduces energy consumption by 90% while retaining accuracy, simultaneously building on human agency by creating an "AI-in-the-loop" workflow, where the AI reviews the user's input. Not only did Gemma 3 use the least amount of energy, it was also the most consistent model during testing.
Overreliance
A key concern of the project was the overreliance on AI. During usability testing with real users, we discovered that participants were more likely to trust the AI's results without reviewing them. This can be a major risk when dealing with sensitive data, so we changed the "human-in-the-loop" workflow, instead making the AI a literal co-pilot, reviewing the user's input and providing feedback. This made our test subjects more critical, while feeling more in control.
Results
Efficiency shouldn't be a key priority when handling sensitive data, but introducing this step into the workflow can reduce the amount of indivduals needed to resolve data quality issues, as the Visual Learning Model can reduce the amount of reviews required. Reducing the amount of pressure on data stewards can free up time for them to focus on more important tasks, while not making them feel like they're being replaced by AI.
