Customised AI solutions for fintech sector prove a Keeper

Chatbots have already proved their value, with the latest generation able to generate surprisingly sophisticated responses to users’ queries. But when will they take on more complex tasks? Can artificial intelligence (AI), for example, automate or assist in credit decisions?

A general purpose AI can’t, but it could be done if someone developed a tailored or special purpose AI chatbot.

Founded in 2011, Limerick-based Keeper Solutions designs, develops and deploys AI-powered products for fast-scaling fintech companies across both the United States and Europe, and recently achieved ISO/IEC 27001: 2022 information security certification.

Notably, the company uses its AI expertise to deploy credit decision-making systems and predictive analytics – two very clear areas where AI can have a significant impact on the finance sector’s top and bottom lines.

Interestingly, Keeper Solutions started out offering AI to augment a wider range of sectors. But the already dominant fintech business grew explosively and thus became its singular focus.

“We’ve been working with fintech since we started in 2011, and it was probably 80 per cent of our business to begin with,” said Keeper’s founder and chief executive Stephen Walsh.

Since then, however, fintech has become Keeper’s exclusive focus, Walsh said, driven by the growing market for AI-based solutions in the sector.

“Since ChatGPT came on the scene we have felt we needed to lean into AI more deeply than ever, because we could see a need for it in the fintech sector for all kinds of customers, frankly.”

Keeper has now set up an AI lab focusing on developing generative AI specifically for finance.

“Our role, traditionally, was to develop scalable fintech products, oftentimes as SaaS [software as a service]. Now we see our mission as designing and optimising AI fintech products for our clients.

“Our clients and their customers are all talking about doing things related to generative AI and we want to be able to help them with that,” he said.

Today, Keeper’s partnerships include Fexco, Taxamo, Brite:Bill, Momnt, Umba and AccountsIQ, placing the company at the cutting edge of how fintech is changing finance.

“Nearly all of our clients are VC-funded start-ups or scale-ups,” Walsh said.

Delivering more with lower risk

Working not only with cutting-edge technology like AI, but also in such a fast-moving sector, has seen Keeper develop methodologies designed to both speed up results and reduce risk. High among them is the so-called ‘design sprint’: a time-constrained, multi-phase process that uses design thinking to reduce the risk of bringing a new product or service to market. This, Walsh said, has now been tailored for the delivery of AI projects.

“One of the methods we use in product design is the design sprint – we call it the Keeper Design Sprint: you run it over two to four weeks, and we’ve upgraded it to an AI design sprint,” Walsh said.

In practice, this means designing and developing AI products that seek to create value specifically in fintech.

“We try to create use cases based on the data available to them, and identify the high-value use cases and, based on that, we build a prototype chatbot,” he said.

The company’s ambition goes further, however. It intends to deliver an AI library specifically for fintech to give its clients a head start.

“Keeper is building prompt libraries with custom-made prompts which enable our team to work more consistently with AI. Taking the time to map your products and services at a fairly detailed level and catalogue that in a prompt library gives a common foundation for AI use by teams of designers and developers,” said Walsh.

This can integrate with all of today’s large language models (LLMs), such as those from OpenAI and others. However, it is only the beginning, Walsh said, pointing to the potential value of custom-developed LLMs that can deliver not only sectoral expertise but meet the exact needs of a given business.

Keeper Solutions’ AI Lab is working with tools like Hugging Face, ChatGPT API, Python and Chatterbox. Its focus is on developing the capability to design and train customised LLMs that are specific to fintech.

“It can help you do more with commercially available LLMs, but we see a lot of value in creating custom LLMs that are specific to their needs,” he said.

This may seem a rapid development given that LLMs only appeared in the public consciousness with the arrival of ChatGPT 3.5 in November 2022. Nevertheless, such is the pace of progress and the desire for augmenting business with AI that it is a realistic prospect, and one that Keeper’s clients are already discussing.

“Talking to our clients, we’re seeing that they are seeing that as well: being able to create custom LLMs that specifically work on high-value use cases, that’s a valuable thing,” Walsh said.

“That’s why we’re in the process of acquiring GPUs and CPUs so that we have the firepower needed to train models.”

Interestingly, Walsh said that this need not be in the cloud: on-premise AI could be an option, given the power of Nvidia’s GPU chipset. This may put some minds at rest because, while cloud has become the norm for much enterprise IT, it has not proven itself cheaper than an on-premises set up.

“The cost of the cloud services that you can run up just provisioning, the pipelines can be massive,” said Walsh.

There is also the underlying question with all IT today, one that has only come into sharper focus with the rise of AI: what is the environmental cost of technology?

“We’re very conscious of the cost of electricity and the need for carbon neutrality. We’re looking to get as much AI as possible using as little electricity as possible,” Walsh said.

Keeper, which already has deep ties to the US city of Atlanta, Georgia, is now seeking to develop its client base in financial centres including London, Paris and Dublin, as well as in the US. In the course of this, Walsh said, the company has found that the most innovative work is taking place in the fintech start-ups and scale-ups, not in the traditional banks and insurance companies.

“We’re seeing generative AI-native start-ups in the US. I think all of them are looking to see what they can do, and they’re in a better position to do something than the corporates, who are focused on their data governance frameworks, which is very important but it can slow them down,” he said.