NatWest and NVIDIA have worked together to develop speech AI designed to better understand how the bank’s customers across the UK speak - from different accents and pronunciations, to the words, phrases and everyday language they use when talking about their money.
The collaboration combined NVIDIA's cutting-edge AI speech recognition technology with NatWest's leading AI research and its understanding of the millions of customers it serves across the UK.
Together, researchers and engineers from both organisations set out to address a challenge facing the next generation of conversational banking services: improving how speech AI understands the wide range of accents, words, phrases and financial language used across the UK.
The work underpins NatWest's recently announced Spending Insights tool - a conversational AI service developed by the bank that allows customers to explore their finances using natural voice and text conversations – and showed the system to mishear 20% fewer words, with an overall reduction in Word Error Rate, the standard measure of speech-recognition accuracy, across the accents tested.
Spending Insights, which is now available to 75% of Royal Bank of Scotland customers as part of an ongoing rollout, allows customers to talk naturally about their finances using voice and text, asking questions about their spending, exploring trends and receiving personalised insights. Customers can interrupt, ask follow-up questions and provide additional context, with the experience combining conversational responses with interactive charts and visualisations.
Maja Pantic, Chief AI Research Officer, NatWest Group said: “NatWest has built trusted relationships with customers across every nation and region of the UK, and we know they are increasingly using AI in their everyday lives. By understanding how their expectations are changing, we are using AI to strengthen our relationships and better support customer needs.
“That means customers being able to interact with us in a way that feels natural to them. If they choose to speak to AI, the technology needs to understand the different accents, words and everyday language used across the UK.
“We are continually learning and improving how we use AI to make banking simpler, more personal and more responsive, while maintaining the trust customers place in us. Working closely with NVIDIA has been an important part of helping us advance that work.”
Aser Blanco, Global Head of Banking from NVIDIA said: “Speech recognition becomes much more challenging when it has to work in real time and understand the accents, terminology and ways of speaking people use in everyday conversations. Working with NatWest has allowed us to apply NVIDIA Nemotron open-source models to those real-world banking challenges and identify where the technology can be improved. The results show how adapting speech applications to real-world language can make voice experiences more accurate and natural.”
NatWest selected NVIDIA Nemotron automatic speech recognition (ASR), which converts spoken words into text, for the speech-recognition element of Spending Insights. The resulting text is then interpreted by NatWest’s proprietary Small Language Models, helping the experience understand what a customer is asking and generate a relevant response.
The speech AI project began with in-depth testing with 90 NatWest colleagues from different parts of the UK, whose natural accents included Glaswegian, Northern Irish, Welsh, Yorkshire, Mancunian, Midlands, Southern English, non-British accents and more. Together, they recorded around 3,000 realistic banking questions.
Colleagues were asked not only to read scripted questions aloud, but also to ask them in their own words, using the phrases, pronunciations and more informal or colloquial language they would naturally use when talking about their finances.
Using these insights, NatWest researchers worked closely with NVIDIA specialists to identify where speech-recognition models performed well and where they found particular accents, pronunciations, words or ways of phrasing a question more difficult to understand.
The research also highlighted how seemingly small transcription errors can change the meaning of a banking question. For example, during testing, a Scottish colleague saying "last quarter" became "last water".
Researchers found that merchant names, dates and financial terminology could be particularly challenging for speech-recognition technology, despite being critical to understanding a customer's request correctly.
NatWest then expanded the work using anonymised contact-centre transcripts, allowing the model to learn from a much broader range of real-world banking language. This helped further improve how the model understands merchant names, financial terminology and the different ways customers describe their finances in everyday conversation.
The transcripts were anonymised and remained within NatWest’s secure environment. No NatWest customer data was shared with NVIDIA, and any adaptations made using NatWest data are ringfenced to the bank.
NatWest and NVIDIA will continue working together to optimise the speech-recognition technology as conversational banking experiences evolve, helping ensure customers can interact with services in the most natural and intuitive way possible.
Notes to editors
- Further information on NatWest’s voice to voice Spending Insights tool is available here: Spending Insights. Video available here: Spending Insights video.
- NatWest’s development and use of AI is governed by its AI and Data Ethics Code of Conduct, which sets out the principles for a responsible and ethical approach to innovation and ensuring customers and their data are protected. Further information is available here: AI and Data Ethics Code of Conduct.
- NatWest published its AI Adoption Report in September 2026, providing a picture of AI adoption across the UK, drawing on NatWest's experience as a major adopter of AI alongside proprietary research among consumers and businesses. The full report is available here: AI Adoption Report (PDF, 15 MB)
