Kathy Gibson reports from the Microsoft AI Tour – The financial sector is one of the world’s biggest technology users and has been among the leaders in the adoption of artificial intelligence (AI).
But, before they can reap the benefits of AI, financial organisations have to overcome the significant challenges of compliance and security, says Rupert Nicolay, director: industry advisor worldwide financial services at Microsoft.
At the same time, they operate in a very competitive environment where it can be difficult – and expensive – to recruit and retain customers, he says.
“AI in financial services is a big workload for Microsoft today,” Nicolay says. “We believe it is one of two or three industries that will be disrupted by AI – and particularly generative AI (GenAI) – in the coming years.”
At the Microsoft AI Tour in Sandton today, many of South Africa’s largest banks shared their experiences in deploying and using AI.
Stuart Emslie, head of actuarial and data science at Discovery Bank, says his organisation was focused on using AI to increase its return on investment (ROI).
As an insurance company that moved into banking, Discovery started off by creating a centralised database that enabled it to use its ecosystem of products to create what it calls behavioural fingerprints and would help to enrich clients’ financial health.
“Our use case was how to embed data products across the client lifecycle, then be able to measure against a single framework, the shared value index,” Emslie explains.
The net result has been a high ROI across the business.
Going further, Discovery is using data to provide better services.
“In the services space, we have embedded assistants to enable agents to service clients in a hyper-personalised way.”
This has moved services from a cost centre to a profit centre
By embedding actions into the services environment, agents can proactively assist clients to attain financial health, and push actions.
“So clients give us a better score, and agents feel empowered,” Emslie says.
This has helped Discovery to convert its contact centre from a cost centre to a profit centre, measured against shared value to create ROI as a business.
The key to implementing successful AI projects lies in data, Emslie adds, and particularly in integrating data across business units.
“We quickly saw the power of integrating data, with a clear use case being banking data, giving us financial fingerprints,” he explains. “Applying that data to insurance pricing is a powerful way of delivering value to our clients. We can offer pricing discounts because of the information we know about them and how they manage their money.
“Developing that in a highly secure and controlled way was highly important for us.”
Emslie stresses that AI is not the end goal in any of these implementations. “You need to get the base layer right – the data and the management. Once you get the framework right, you can build the GenAI on top of that.”
Andre Jansen, security consultant at Nedbank, agrees that the framework is key.
“Initially, when we started using AI, we created user standards – both to enable the technology and also to understand what the known threats were at the time.
“We have now created a technology standard to build a framework for AI.”
This framework would include threat modelling, he adds.
People are still a vital cog in any AI implementation, says Jansen, particularly with LLMs (large language models) still prone to hallucination.
Adit Mehta, head of MLOps at Discovery Bank, points out that any AI implementation must be built on a solid foundation that starts with the team.
“Our philosophy is to keep the teams lean but highly specialised, which means people have to wear many different hats. This lets us build teams with a good mix of data science, software, and engineering talent.”
When it comes to scaling, Mehta believes this can only happen if there is business buy-in right from the beginning. “This is very important.”
“From an engineering perspective, having an application that is scalable and useful comes down to good architectural practices right from the get-go.”
Embedded standardised security practices and employing the right tools are also key to successful implementation, he adds.
Asokan Moodley, executive head: intelligent hyper automation at Nedbank, says his organisation started out by creating a centre of excellence across the whole bank.
This CoE included not just IT people, but security, group data and governance, risk and compliance, enterprise architecture, communications, scrummaster and more.
“This set the scene for how we plan to overcome the challenges,” he says.
Skills is a massive challenge for any IT project, and particularly so for AI. “We had to embark on employing skills, training people, and partnering with the likes of Microsoft on skills development.”
Setting ethical guidelines was vital, ensuring there are no ethical issues, biases, or privacy concerns. “We had to look at mitigating the risks of not just cybersecurity, but also bias and privacy concerns,” says Moodley.
A final building block was governance frameworks, detailing user guideline policies. And the team had to ensure that users read and understood the guidelines.
Developing and deploying new applications is just the first step – if people don’t use them, it’s just so much wasted effort.
“Adoption and change management are key,” says Moodley. “We tend to forget about these, and forget there are both tech and non-tech people in the organisation.
“For us this was a key point when we started with adoption and change – getting buy-in and adoption from the C-level down.”
The approach seems to be working: Moodley says Nedbank’s CoE has delivered more than 50 uses cases – and they are delivering ROI.
“We have been engaging with business value to measure ROI and figure out what we can do with the time we get back after enabling people.”
Importantly, Moodley says Nedbank’s AI initiatives have gone beyond summarising emails and meetings to enabling business units. “We have champions in each business unit to help deliver use cases.”
At the end of the day, measuring value is key, and CEOs want to see real ROI.
Christoph Nieuwoudt, chief data and analytics officer at First Rand Group, agrees that usability is key – and too often new tools and applications languish unused.
First Rand aims to counter this by ensuring that AI is embedded in the tools and apps that employees use every day.
“Building models is hard, but once they are deployed, people don’t know – it just happens automatically.
“For any technology, user adoption is key,” Nieuwoudt adds. “You have to take people on the journey with you and show them how a solution practically integrates into their workflows.
“For this integration, APIs and function calling are key.
“To give people the ability to work smarter, you need to get the underlying data into a graph network so you can use the integration – and then you don’t have a user adoption problem.”
Nieuwoudt points out that group banks hold rich data sets on customers that give them incredible insights into retail, commercial and corporate activity.
Models can only be improved using this rich data and it is plugged into all kinds of decision-making at the banks.
None of this is new for First Rand. Nieuwoudt points out that the organisation has been doing AI for over a decade and that about 30 projects are currently underway.
Many of the existing implementations aim for internal efficiencies, with several customer-facing projects about to go live.