Flagstone innovates with Microsoft Azure OpenAI to enhance client experience

Flagstone is the UK’s leading savings platform with a vision to revolutionise the cash deposit and savings market. The UK-based business began life as a fintech disruptor born in the cloud. Today, it administers more than £11bn of client assets, working with more than 60 banking partners. 

In March 2024, the fast-growing UK fintech announced one of the largest equity investments in UK technology since the start of the year. US specialist financial services private equity firm Estancia Capital Partners will invest £108m in Flagstone to capitalise on expansion opportunities in the UK and internationally. 

Flagstone migrated its infrastructure onto Microsoft Azure in 2019. Lee Provoost, Chief Technology Officer at Flagstone, explains the reasons for the switch: “When Flagstone was founded some ten years ago, it was common for tech startups to use a different cloud service – but that service never really felt like home to us.” 

Microsoft feels like home

By contrast, Microsoft Azure is a great home for Flagstone. Lee Provoost explains, “When we took the opportunity to revisit our original decision, we concluded that Microsoft Azure is a much better option for us.” 

“We build all our software with Microsoft .NET technology and all our engineers are Microsoft developers. Azure gives us fully integrated user authentication and access control all the way from our developers’ laptops to how they log into the Azure cloud. The end-to-end platform as a service offering takes away a lot of the mundane, manual tasks with better support for our .NET applications and databases.”

Microsoft is an innovation partner

Innovation success is at the heart of Flagstone’s DNA. “From day one, we have had to build our platform from the ground up,” explains Lee Provoost. “That innovation culture is in our DNA. Exploring how technology can help us do things better and faster has always been a big part of Flagstone culture.” 

The Flagstone savings platform is used by charities, corporates, wealth managers and high-net-worth individuals. Flagstone technologies power services for its financial services partners, helping 600,000 clients to save. 

Lee Provoost continues, “Flagstone innovation ranges from basic things like using Microsoft Power Platform to automate processes through to exploring how Microsoft Azure OpenAI can help us transform the employee and client experiences. We have been lucky enough to work with Microsoft’s Digital Cloud Solutions Architecture team to advance this agenda.” 

Speed to competency is hampered by complexity

The scope of services and products offered by Flagstone means that complexity is also inherent in its model. Flagstone must bring simplicity to complex client requests and the complex services and requirements of 60+ different banking partners. This inevitably means an extended time to competency for new staff members.  

“Our activities are quite complex from a policy and procedures point of view,” Lee Provoost acknowledges. “It takes a long time to get new staff members up to speed.” 

This lengthy speed to competency can be a brake on growth, particularly unplanned or unpredictable growth.  

“It takes about six months for a new joiner to get fully up to speed at Flagstone,” confirms Mark Nichols, Interim Director of Operations at Flagstone. “Our customers are used to a white-glove service in an advisory context where they can go to one person to get all the answers. But there’s a lot of information needed to move our customers through each of their customer journeys. Stitching all that disparate data together and making sure that information is up to date and based on the latest version is quite difficult.” 

Even for experienced staff, this complexity can be a problem. Mark Nichols adds, “If we look through our compliance documentation libraries and there can be numerous different documents all purporting to be the right one, it is very easy for staff to become very unproductive, very quickly.” 

The benefits of using familiar tools

Flagstone has been exploring how to hasten speed to competency without compromising quality, accuracy or the client experience.  

Because Flagstone’s collaborative culture ensures much sharing of knowledge and expertise, Flagstone’s hub for collaboration and knowledge sharing across the business – Microsoft Teams – became a key focus. 

“We saw that an AI chatbot could be a good way to help accelerate speed to competence,” enthuses Mark Nichols. “With so much of our work happening in Microsoft Teams, having a chatbot make sense of all that structured and unstructured data is a natural progression.”  

“We want the chatbot to put all that information in front of staff so we free them up to be much more customer centric,” Mark Nichols continues. “They can focus on what is the best action for the customer at that moment, rather than trying to recall information or find which bit of process documentation will fit. That’s important, for example, in a bereavement case. The last thing you want to do is fumble around, delaying the conversation while you dredge up the right information and lose the sensitivity for the client.”

Developing an AI-powered chatbot to simplify the complexity

Following discussions with Microsoft, Flagstone decided to assess the possibilities of artificial intelligence (AI) with Microsoft to answer this challenge. Azure OpenAI felt like a natural choice for Flagstone, from the point of view of user experience, technology and data governance. 

“While one of our first AI use cases is AI-powered chatbots,” Lee Provoost says, “the last thing that Mark wants is for me to come along saying ‘I’ve got another tool to make your life easier’. He’s already got ten different tools. An eleventh tool isn’t going to make his life easier. The idea of being able to access AI capabilities in a single interface, in a familiar tool like Microsoft Teams where so much of our work is already done, it was an obvious choice.” 

“Plus, of course, we’re already using the Microsoft platform in our business,” he continues. “The data security and governance around that is already assessed and proven. The fact that Microsoft offers an OpenAI solution means that we can experiment and upload sensitive data and policy documents. We can train and experiment with our chatbot without worrying where that data will end up because with Microsoft we know we have control of our data.” 

Fast tracking a proof of concept

“We began a compelling proof of concept (POC) to see how we arm our customer-facing teams with all of our information and make it easy for them to discover rather than having to retain it,” recalls Mark Nichols. 

Flagstone’s existing innovation strengths and collaborative culture were the ideal conditions for launching the proof of concept quickly.  

“We often invest in new technologies and our engineers get a lot of innovation time for tinkering with new technologies,” explains Lee Provoost. “We’ve had six engineers working on this alongside six people from client services. Those super users have the experience to action quality control and internal advisories. They are closely involved with testing and iterations, working with the developers to make it better.” 

By leveraging the strengths of Azure OpenAI, Flagstone has made rapid and impressive progress. While Azure OpenAI On Your Data enables you to run advanced AI models on your enterprise data without needing to train or finetune them, the service also makes it possible to upload training data to optimise or “Flagstone-ise” a model. The Flagstone team is excited by the early results. 

“First of all, I was impressed by how quickly we could get up and running,” reports Lee Provoost. “The current results have exceeded my expectations. It was my belief that we needed a lot of data to train these models before they became even remotely useful. We have a lot of policy documents but, in the grand scheme of things, we’re talking about a small number. And they’re big. And complex. The fact that we fairly rapidly got some decent answers to the complex questions we put in far exceeded anything I was expecting.” 

Early results are impressive  

The success of the project has strengthened the commitment to innovation and continuous improvement in the business, says Mark Nichols. “This has been a great experiment,” he states. “It’s enabled people to look up and forwards and explore how to do something we’ve always done better.” 

The Flagstone team is now looking forward to how it can successfully deliver a wider rollout through the business. 

“We’re at that first hurdle,” Lee Provoost explains. “We’re all pretty impressed with how quickly we are up and running and the answers we get are decent. The next stage is to turn this from a POC to a properly funded project with a proper team behind it.” 

Quality is paramount

Given the compliance and regulatory environment in which Flagstone operates and the high expectations of its client base, the business has a strong focus on perfecting the quality of the chatbot output. 

“We have very high level of trust and appreciation for the client service we provide. And our clients manage significant amounts of money on our platform,” emphasises Lee Provoost. “We cannot afford for a client service executive to give out the wrong information.” 

For this reason, the team have plans to further train and finetune the Flagstone Azure OpenAI models.  

“We are going to continue to invest to finetune the solution,” Lee Provoost confirms. “We need to develop the search index so it gets better at pinpointing where to source information from. And, because a lot of our information sits in different systems, we are building some integrations to make sure we always get the latest version of policy documents and procedures and policy and regulatory changes.”

Guided by responsible AI principles

Another way Flagstone is underwriting quality is through its clearly stated responsible AI policies, which are in close alignment with Microsoft’s Responsible AI Principles. Clear rules around technology use, keeping a human in the loop and good data governance are aligned with Microsoft tools for data residency, data governance and cyber security. 

The Flagstone team worked closely with Microsoft Digital Cloud Solution Architects to engineer a solution that adhered with these principles. It also makes use of existing Microsoft security tools. 

Lee Provoost explains, “We have a fully integrated SIEM tool with Microsoft Sentinel. It’s a familiar environment that is fully integrated with our billing and controls and it covers all security events, no matter where they’re from. It’s pretty much a no-brainer to adopt Microsoft Sentinel as your primary security platform.” 

Planning the wider rollout

“The next step will be to roll this out across client services,” Lee Provoost states. “Then, once the client services team is really happy, the next step will be to open it up to external clients. There is a question over how we will do this. We are modernising our contact channel landscape at the moment. We’ve been rolling out human chat to provide a quicker service for some of our clients. We need to understand whether our clients actually want to speak with an AI bot or whether we use the bot to surface answers to direct queries live in the moment and support our human agents that way.” 

Mark Nichols agrees, “We spend a lot of time hiring our people. We should be making their job less about having a core competency in navigating document libraries and making it more interesting and engaging.” 

“There are a bunch of use cases that flow from this,” he continues, “Another area where we see the potential is in quality management and quality control of our customer contacts. It is a manual exercise right now to listen to the call, observe the sentiment and distil the key moments. If technology could be applied to review, summarise and quality assess all those calls it would be phenomenal.” 

Boosting existing AI achievements with the use of Microsoft Copilot

“AI is very high up on our agenda,” confirms Lee Provoost. “We’re trialling a number of different Microsoft Copilots at the moment to see how we can use them to increase productivity.” 

Microsoft Copilot for Microsoft 365 has been the key focus initially, driving some diverse use cases which not only deliver time savings and boosts to productivity but are making people more effective in surprising ways.  

Lee Provoost explains: “We have a small-scale trial using Microsoft Copilot for Microsoft 365 in Teams and Office. Copilot does a good job summarising meetings and topics in a safe and controlled way which we know we have control over. That has a big focus on supporting people with neurodiversity. We know it’s helpful for people to get really good quality summaries and transcripts after a Teams call. People really benefit from that.” 

Mark Nichols adds, “People are using Copilot alongside the editor in Microsoft Word to adapt texts to be more objective while staying in their own style. People have been using it a lot during our end of year review cycle.” 

While Microsoft Copilot for Microsoft 365 has been the primary focus, the team are also exploring the potential of other Copilots, including GitHub Copilot for code development and review. Lee Provoost confirms, “Our engineers are starting to experiment with GitHub Copilot. That’s also high on our agenda because it is all about productivity.” 

The future of AI at Flagstone

The Flagstone team is confident that complexity and speed to competency will no longer be a potential brake on growth.  

“We’re now able to project the ambition that product mastery is available to all of us, so that our executives and client-facing teams really live amongst the clients to add value,” says Mark Nichols. “My end vision with this is for us to serve 100 times more clients with two times more staff because they are now focusing on the value add, not the information search and retention.” 

With one successful project already riding high, the Flagstone team is very excited about pursuing its AI objectives through the coming year. 

“Our whole 2024 roadmap is littered with initiatives to bring us closer to our vision of using AI to provide more rapid responses to complex queries,” says Lee Provoost. “That’s from stitching things together with Power Platform to making a big investment in Microsoft Azure Data Lake to ensure we have the right data in the right places to support our vision. And we’re putting all the guardrails and processes in place to support risk management and compliance.” 

Mark Nichols agrees, “We have a really good strategic partnership with Microsoft now, with monthly review meetings where we learn about the latest new technologies. We’re having those strategic conversations with Microsoft cloud architects so we can explore how technology can help us deliver on our ambitions to empower our customers to be masters of their own cash management.”