Australian FinTech recently chatted with Marc Wilson (pictured), Founder and Chief Executive Ambassador at Appian, to hear his insights on the evolving state of Artificial Intelligence (AI) in financial services across Australia and internationally.
A recent MIT research report noted that 95% of AI pilots fail. From what you’re seeing globally and in Australia, what’s the most common mistake financial institutions are making with their AI investments?
I meet a lot of different financial institutions around the world, and those that make mistakes in their AI approach tend to have common characteristics. The first is at senior leadership level; the board or their leadership has decided they need to do AI, but that’s the full extent of what the direction is. Failure often occurs when people lose sight of why you do things in organisations. Many financial services organisations are focused on doing AI rather than trying to find value where AI can help achieve business goals. Organisations should be thinking more about how AI is going to help processes be faster, more efficient, cheaper, better. They need to think about what they’re trying to achieve first and then find the right tool to apply.
Where in financial services do you see the biggest opportunities for AI to drive transformation? And from your experience, where do banks often start?
There’s undeniably a time and a place for AI from an individual employee productivity perspective. Plenty of organisations are aware that it’s very useful to put these tools in the hands of employees to save a little time here and there. However, the challenge from an overall organisational perspective is that when you’re looking for real value statements, when you’re looking for things that can really move the needle, you have to look at strategically important processes that AI can impact. That’s why Appian is encouraging organisations to look at their processes first. An organisation can look at the process and identify any areas that are causing problems, bottlenecks, delays, and assess whether adding an algorithm to the process might help to make improvements. A common example of a process an organisation will look at first is the onboarding experience process, which in financial services often involves different checks, compliance or regulatory regimes.
The MIT research suggests the greatest returns come from back-office automation. Do you think these less visible areas of banking or even wider enterprise are yielding the strongest results? Or are there other areas or processes you’re seeing with bigger results?
One of Appian’s perspectives, and one I wholeheartedly support, is that we see financial institutions sometimes bring too much of a spotlight to a particular part of the business – the front office, mid office, back office or our KYC process or an AML process. Where sometimes the most strategically valuable places to focus are the white spaces between some of those bigger elements. If we again look at the onboarding process, it inevitably impacts front office, mid office, back office, and organisations are going to be able to identify problems throughout. Sometimes they’re within the transition from the front office to the mid office to the back office, and being able to smooth these transitions can be a strategic win. One area that has seen big success in financial services has been document processing, where AI is removing the manual human double and triple checking of information, and instead bringing the relevant information to employees when needed.
Australian banks face a unique regulatory environment with APRA, AUSTRAC obligations, and strict data sovereignty requirements. Does embedding AI into core processes make compliance harder or easier in the banking sector? How can banks maintain audit trails and explainability?
We know regulation is necessary. A good approach to regulation helps customers. It drives banks to make good choices and develop more efficient ways of doing business. Regulation involves audit trails, data, predictability, transparency of the process (what decisions were made when). However, currently regulatory issues often constrain AI, as AI can in many ways run counter to these regulatory obligations. That’s because most regulatory schemes were historically based on the idea that we’re looking for deterministic outcomes based on intents. This again points to the importance of considering AI in the context of a process and in financial services this applies on an ongoing basis due to continuously updated regulatory schemes. If you consider process first, AI second, this allows for a tailored AI approach. Instead of thinking that you’re going to implement one AI tool to manage the entire customer onboarding process – which I think we’re a long way off – it’s more realistic to consider applying AI to some aspects of onboarding. That might look like applying different AI algorithms in six steps of a total 20 step customer onboarding process. This might include different algorithms that are specialised for specific aspects such as document processing or decision making. This approach allows an organisation to incrementally build up confidence in appropriate use of tailored AI, and to naturally develop an AI audit trail, rather than implement with wild abandon.
What is the role of humans in relation to AI? How does combining AI with automation and human expertise create a more resilient, compliant, and efficient financial institution?
I think it’s important to start with the attitude that this is more of a revolution in how people are going to work, rather than just purely about technology. What we are seeing with a lot of our customers that are making improvements to their processes with AI is that in almost all of these cases, they have started out with AI helping humans, not replacing them. Additionally, much of the value has been in what we sometimes refer to as boring AI: having AI summarise unstructured documents in a few minutes instead of a human reading these lengthy documents over 45 minutes. In financial services we’re seeing this in in a lot of intake AI document work, where there’s digestion of broker forms or credit information – areas where it’s not the best or most rewarding use of a human’s time or skills, which are better spent on more complex judgement-oriented activities.
In financial services, it’s critical that AI systems are not treated as foolproof. Robust validation mechanisms—such as randomised sampling—are essential to assess the reliability and performance of various LLMs and algorithms. Additionally, information within financial institutions varies in sensitivity, requiring tailored governance and risk controls based on data classification. Leveraging specialised AI tools for individual business processes improves both accuracy and oversight, ensuring AI is deployed with compliance and efficiency in mind.
Can you walk us through an example of how an Appian customer has embedding AI into processes to deliver measurable business results?
One great example is Tower Bank, who have been a long-term financial services customer that recently gotten into a hybrid crypto. Under their previous retail banking approach Tower Bank applied various AI techniques into their new user onboarding process, however it was still taking seven hours. This was far too long for the kind of retail business that they want to be. Working with Appian they’ve been able to drive that down to 90 minutes by bringing a lot of AI work into document processing, coupled with human-centered work. They also have their sights set on continuous improvement, which AI help by supporting ongoing evaluation of processes.
