Harness: Next Gen Software Delivery in Financial Services

During a panel discussion at Harness’ {unscripted} 2025 event, key industry leaders came together to address how AI and automation have impacted the financial services landscape. 

Additionally, the benefits are weighed against the risk as the panelists explored support measures such as guardrails to continue delivering reliable results. 

How is AI currently impacting financial service providers?

Dill Bath, AI Technical Lead at Allianz says: “So that's investment research, really deep-diving into that and getting decisions quicker. There's so much financial data out there, and we've been using traditional machine learning for a long time, but Gen AI is just another tool that can help us get to those buy-sell signals quicker.”

Generative AI is quickly catching up with fintechs globally. The technology is constantly evolving, and has so far been used for automating mundane tasks to aiding in the fight against financial crime. 

Dill continues: “The other areas are really around operations – automating manual tasks. We have a lot of third-party and client requests in shared mailboxes, and it's a very manual process in operations, so we're really optimising that.”

Companies such as Bloomberg use generative AI for enhanced data analysis, while Starling Bank and Klarna use generative AI to assist with customer enquiries. 

Lloyds Banking Group has also adopted AI, in particular harnessing the technology to assist in frontline operations and customer enquiries. 

Tony Phillips, Engineering Lead for DevOps Services Engineering Platform at Lloyds, adds: “In terms of what we're doing from an AI perspective at Lloyds, now that we've got that foundation of the internal developer portal (IDP), that's how we consider it, people can now onboard and they can now start to consume this. How do we start to expedite it?”

Reports predict that generative AI could expand by US$5.56bn between 2024 and 2029, growing at a CAGR of 36.9%. 

How does efficiency translate into productivity using AI? 

A 2024 report from McKinsey suggested that banks could improve capacity on tech teams by automating processes. 

Bettina Topali, Senior Software Engineering Manager at Hargreaves Lansdown, says: “What our clients need are slick experiences and modern services. They don't want to be on the phone with a help desk all day. So how do we achieve this? We achieve this through automation.” 

Automation comes with risks, on top of benefits. AI still carries risks such as hallucinations, leaks, and compliance disruptions. 

To combat these risks, developers need to introduce guardrails. 

Hargreaves Lansdown has “embedded all of these guardrails that, in reality, help us move faster”, according to Bettina.

Guardrails are a necessity in automation and AI. By introducing guardrails, which are designed to work in line with an organisation’s standards, policies and values, it introduces a vital backboard for any user: trust.

Dill notes that Allianz is heading towards a tech-first approach. 

“OPA, or Open Policy Agent, is very new for us, just codifying all the policies, but not in a way to block our developers, but almost like a copilot to nudge them in the right direction. So we are really investing a lot of time into that. 

“So don't block, but report and say, 'Hey, you might be doing something wrong here.’ So that's working well in our pilots, but we really want to push on that. And the other thing is we want to go more in a tech-first way. 

“When new regulations come in, we really look at the regulations and interpret them from a technology-first perspective rather than, 'Hey, let's add it to the policy, let's create a manual process, and then let's ask once a year whether you're doing this or not.’ It doesn't work.”

Alongside a tech-first approach, companies have to think about people: developers, first. 

Tony Phillips asks: “Then how do we embed a level of security in those guardrails, but also with the notion of flexibility for our developer and engineering community?”

“That's where we're at at the moment. So we very much lean on Open Policy Agent, using a template for a lot of what we're doing. Data is still the thing that we've yet to get a hold of.”

How is AI used in the software developer lifecycle? 

Developers have worked to integrate AI into a number of softwares that benefit fintech, including fraud detection. 

Tony adds: “There's definitely quite a lot in the bank that we're doing from a machine learning point of view, particularly around security, fraud detection, and a lot around those customer applications that have been in place for some time now.”

Machine learning is also employed by developers to reduce time-heavy processes in day-to-day operations. 

Aaron Gallimore, Senior Director for Cloud Engineering at GlobalPay, adds: “Our big focus is making AI scalable and secure and approved so that our developers spend less time moving between the tooling and having to do all of that context switching and heavy lifting, and just let the LLMs do that for them now. It's really exciting.”

The Alan Turing Institute predicts that LLMs are forecasted to be a US$40m market by 2030, based on a joint report with HSBC, Accenture and the UK Financial Conduct Authority.

Daniel Terry says: “We’re looking into a world where developers are not the producers of the code, but more the conductors of agents going forward. And I think when we hit that stage, we also need to look at how we solve this in the pipeline. 

“How do we secure the output of those thousands of lines of code that are generated within minutes instead of months or years?”

While the fintech herd gathers around the AI well, some companies still refuse to drink.

The question remains: how to build a progressive culture around AI?

A key part of the technology revolution reigns supreme; climb aboard or be left behind. 

Embracing change can be difficult, however, varied approaches can achieve desired results. 

Aaron Gillmore says: “I think what's really important is still training, is giving people a foundation on how to use these tools in the right way and not just use them like a usual search engine. 

“What we decided to do is put in place almost like university sessions, where people will just come along and we'll do short demos and riffs on what things they've done in the last week. And you just see that spark in people's eyes of, ‘Oh, I can actually use this and do it like this.’ And I think it's building that flywheel of knowledge and culture to change.”

Utilising academia to introduce technology concepts is one method. Bettina mentions that similarly, businesses “need to invest in our people, in our culture, and guide them. I think one of the biggest challenges in organisations, especially in large organisations, is that people have seen strategies failing and failing. I think we just need to bring people on the journey.”

By investing in people first, businesses provide vital stepping stones for the next generation of software developers. 

Bettina continues: "We need to show visible progress, and we need to take them with us and guide them through these steps. And then hopefully, gradually, all of this misbelief is going to be replaced by people actually believing in this strategy. 

“All of these tools that are coming, they're going to start becoming even more available for us. They're very powerful tools, and of course, it's going to start raising a lot of questions about how engineers are going to work in the future.”

So what is left for the future?

Emerging fintechs will be among the first to use leading technologies, as the market continues to shift and adapt. Catching up is the next task that Hargreaves Lansdown believes is most important for current businesses. 

Bettina says: “New startups and fintechs are coming out, they're going to start getting a share in the market, so we need to move ahead. So with all of these tools, we'll now have that opportunity.”

Further improvements to technology could similarly provide an advantage. 

Dill concludes that he “could see big improvements in the coding agents, but that means all the other steps in the SDLC being embedded.”