THE FINTECH INTERVIEW
“ Until about a year ago, most of them weren’ t very good and were quite frustrating,” Dave explains.“ A number of institutions are only now starting to bring agentic AI into that space – the reason being that with agentic AI generally, you need to put the right controls and guardrails in place before you can introduce it. So the predictive element has already reached customers.” He further explains that while there’ s a lot more to be done, progress in implementing agentic AI is well under way. For example, Sapient started putting in machine learning platforms to enable data scientists at high street banks around 2016 to 2017. He says:“ That’ s when people started adopting machine learning in UK high street banks, enabling data scientists to work on improving it ever since.”
AI is being adopted gradually as banks have a wider conversation around implementing guardrails and platforms needed to facilitate and introduce it.
“ What matters most in harnessing the power of agentic AI, beyond the large language models themselves, is the context you provide”
Dave Murphy Head of Financial Services – EMEA & APAC Publicis Sapient
Dave says:“ There’ s a lot of thinking still needed in the agentic space, because it isn’ t always 100 % accurate – there’ s a non-deterministic element to it. Banks need to think carefully about how they introduce that to customers. In many ways, what we’ re seeing is organisations starting from the back office and moving toward the front, so they can get comfortable that it’ s working before they put it in front of customers.”
Banks face two approaches Fintechs globally now face the option to adopt and scale using AI, however, there are a number of approaches to consider. Despite having a more hands-on approach compared to banks, there is still opportunity for guidance to ensure that business gets it right.“ Think about your workflows and capabilities” Dave says.“ Gen AI can be an assistant, giving you some improvements and efficiencies, like an autocomplete. When you’ re typing into a search box and it completes your words for you, that’ s the idea. That will typically only give you around a 15 % improvement.
“ The other side is that you can rethink the process entirely – and this is where it gets interesting for workflows. If you want to be more than just an assistant to a human, you want agents actually doing work on your behalf. To do that, you need to think hard about the workflow, because what matters most in harnessing the power of agentic AI, beyond the large language models themselves, is the context you provide.”
30 September 2026