This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.
Bud Financial: using data and AI to drive bankable insights
Turning raw transactional data into actionable insights is what fintech platform Bud Financial offers its large finance clients so that they can improve customer engagement and carry out affordability assessments.
According to Bud’s VP of engineering and data Michael Cullum, the firm has been developing Large Language Models long before they became the buzz word, as the firm uses them to process and analyse hundreds of thousands of banking transactions to build up a 360 view of customers.
“For the past 15 years the focus in our sector has been about getting data, acquiring data, hoovering it all up and then paying companies like us,” say Cullum, speaking at an AI showcase earlier this month, organised by cloud-based database vendor Datastax.
“But frankly, the truth is most of us have this data just sat there, not doing very much. That’s’ a shame. Why are we collecting all this data and not making use of it?”
Transactional AI
Spotting a gap in the market created by Open Banking and the dawning realisation that banks could use customer data for more than just fraud detection and asset and liability, Bud first started creating mathematical models in 2015 to understand more about the unstructured ‘messy’ transactional data banks were gathering, and turn it into structured data.
“We didn’t call it this then, but what we call it now is transactional AI,” explains Cullum.
Transactional data with context allows banks to build up a profile of customers and potential customers to assess what financial packages would be appropriate for their circumstances.
There is also another reason firms needed to make sense of these strings of underutilised transactional data sitting in their systems: explainability.
As Cullum notes: “When you work in a regulated industry you need to be able to explain why you’ve made an affordability decision – you can’t just say ‘yes’ or ‘no’ to someone. So, the way you get that ‘explainability’ is by turning unstructured data into structured data and then adding context.”
Fast verification in assessing whether someone is right for a loan or not has been one of Bud’s USPs, but recent advances in generative AI has enabled the platform to focus more on the context side of things: personalisation and really building up a profile of customers through data.
This extra context has allowed Cullum and his team add two new products to its wheelhouse.
Chatbot
The first, ‘Jas’, Cullum explains, is a chatbot UI for personal finance management, which enables bank customers to have more personalised and relevant conversations with their lenders.
“This is taking that GenAI chat experience down to individual consumer. If you go back 40 years that’s almost like the relationship you used to have with your bank manager — they would get to know you quite well and your circumstances and they would give you tailored advice, which was good for them because they know what to offer you.
“Now we can use this technology to bring back that same level of personalisation and understanding in a much cheaper way than has been developed in call centres over the last 30 years,” he adds.
Taking this to the next level, Cullum says they then started to experiment with Open AI’s generative AI ChatGPT.
While ChatGPT doesn’t understand the world like a human, Cullum claims that it kind of understands questions like: ‘What sort of a person might be good for a mortgage?’ and it will give you some answers, even if they are not very good.
“But the important thing is that it has this understanding of the world. So, we incorporated that and built up this view of a customer with added context,” he explains.
Building Jas involved adding a series of prompts and scripts, and for parts of this, such as product information and help information, Cullum simply asked ChatGPT to write the script in Go (a programming language) which scrapes the website for product information in HTML and spits it out as commands that feed into the script.
“For that entire step, you can just use ChatGPT to write the script for you. We did this as a PoC, and it worked. ChatGPT is great!” he says.
Drive
Cullum explains that the team then started to think more broadly: now Bud has created 360 customer profiles, has more context as to what’s going on, it is now able to look at this info collectively as an entire customer base.
This is what its second product, ‘Drive’ offers, allowing banks to use info based on broad customer datasets to drive personalisation for better target marketing.
Cullum notes that in the fast-paced world of finance, banks sometimes have to make swift decisions about a customer’s credit risks: he uses HSBC’s rescue of SVB UK earlier this year as a case-in-point.
“When SVB had issues, earlier this year what was the first question on the mind of HSBC: It would have been how many of their customers also have their money in SVB?”
“All this data is really important information. But it normally takes weeks to find out. So, making this data more accessible and easier to use empowers decision making faster.”
“That’s essentially what Drive is — it’s also about being able to show analytical dashboards and responses to your queries,” he adds.
Tech stack
To handle the immense dataflow that AI generates, Bud requires a highly reliable and scalable backbone.
The firm uses Datastax Astra DB for its “dependable fast scalable data architecture” to handle the huge volumes of real time transactional and open banking data that the firm generates.
Bud’s Astra runs on Google Cloud — mainly using the tech giant’s Kubernetes engine which, Cullum says, works to its advantage as it operates across multiple regions.
“Working with Astra on Google Cloud makes life a little easier and they have a strong focus on being able to deliver really good data products,” Cullum adds.
On the analytics side, Bud uses BigQuery, a serverless, scalable data warehouse, capable of handling large data sets.
“Everyone at Bud uses [business intelligence platform] Looker and Big Query for data on a daily basis — it’s helped us become more effective as an organisation,” says Cullum.
Results
According to Cullum the investment in this tech stack and AI is paying off for Bud, their clients and their client’s customers.
Having the right data to hand, he claims, has enabled one high street bank to increase its upsell rate by 25%. Another broker client was able to increase the acceptance rates of loans they offered by 20% with 75% of its customers getting better rates.
Meanwhile, a credit provider client reported a 50% increase in applications while reducing costs by 20%.
Cullum adds that the AI chat function has resulted in one digital only bank see a 350% increase in app engagement and a 10% drop in customer call centre contacts.
Even in traditional areas where transactional data is used, one credit provider reported a 90% drop in fraud.
“These are a great bunch of figures, and it proves that AI is not a fad,” says Cullum.
“Crypto was good at a few small use cases and it takes time to test them out, but with AI we are in this situation where we’ve already validated this as a need for the industry. And it’s not just in one sector. AI is going to impact lots of industries in lots of different ways,” he adds.
To read more stories on big data click here
#BeInformed
Subscribe to our Editor's weekly newsletter