At Cloudera’s flagship EVOLVE event held in New York, Adrien Chenailler (pictured), Global Director for AI Industry Solutions, Financial Services at Cloudera, shared his valuable insights on AI and its relationship with financial services, fintech and the jobs market.
Cloudera is the only data and AI platform company that large organizations trust to bring AI to their data, anywhere it lives. Cloudera are well entrenched in the financial services arena, counting 8 out of the top 10 global banks as their clients, and 3 out of the top 4 credit card networks too.
Q: What are the most exciting AI opportunities in financial services and fintech?
AC: Let’s talk about the banks, especially our largest customer that I cover. So banks onboarded AI and Gen AI two years ago. Now they are starting to look at what are the key use cases that they want to drive. The key message that I see is that in the next one or two years there is going to be a more realistic approach, which is ‘what can we automate without disturbing our customers too much?’
I know there are some Gen AI chatbots coming to financial services, but they are highly constrained, they are not like ChatGPT. You are however looking at more cost savings and cost optimization by deploying Gen AI into your operations, payment processing, but how can you make the flow a bit smoother, especially with cross-border payments as they are quite impacted? As an example of what we were doing, when you send a cross-border payment with SWIFT, you can receive a message back. And these messages are very ugly. As an example, the bank – that is asking for information about this payment – may send you a message of hundreds of words, and then you have someone that has to read this, decipher it and determine what to do.
Banks have built a lot of rule-based systems over time based on this, but you still get a lot of other requests. For instance, the Australian banks are really strange about this because every payment that you send to an Australian bank, they send you a message back: please send us the address of the recipient – which is not required by any other country – and in the middle somewhere they may include this, but they may also add the entire legislation, Australian banking regulation, telling you why you should be doing this. And this piece’s address could be right in the middle of it. So we started deploying Gen AI to understand this messaging and then categorize them, which retrieves the address automatically, and sends it back. It’s something that is very boring and very mundane, but it works. It has improved the speed of the payment and customer expenses are cut back.
AI is impacting on how you deal with your clients too and how they reply to their emails for example. Not that an AI is going to reply to your emails, but AI may prepare the email. In every bank, if you receive an email from your bank, especially if you have sent something like, ‘Can you please send me the last transaction that I did on this credit card?’ the person who is answering from the operations center does not hit send – this person prepares a message, then it gets sent to another person and this person then checks the message – because there’s personal information in it – and then it’s allowed to be sent. So it’s a two-step process with two humans being involved. With these kinds of processes, it’s very easy to replace one of the two employees with AI, especially the first one that prepares it, then the second one can just check and press send if it’s okay to be sent. These types of changes in the operation centers and contact centers, will be impacted much more than the customer facing processes.
During the next two years, you will see the customer experience aspect coming in. How can you generalize a chatbot so that it can actually do something meaningful for you, and have a more conversational side to it? One of the issues we saw recently with a large Australian bank was that they tried to replace a service with a chatbot and then they quickly rolled it all back. They did not explain exactly what happened.
There’s a lot of edge cases in handling conversations with clients. And the last part is that every time you talk to your customer, it’s an opportunity to sell and not just to answer the query. And Gen AI is still not there when it comes to integrating all the data feeds and so it requires having a human doing this kind of fix. So, we’ll first see changes in operations in the next one or two years, then customer experience being impacted a lot more. Those are my predictions – I may be wrong, but that’s how I see it.
Q: When you’re dealing with all types of financial institutions, are they excited about the prospects of what AI could do? Are they also a bit apprehensive about what they’re going to do with the data and compliance concerns?
AC: I think it really depends on who you’re talking to, and you know it fuelled a lot of the POCs (Proof of Concepts) for the last two years. Many of these POCs did not make it to success, many do not see a way to production, and what you’re describing is exactly this – some people are really excited. The tech folks can build some demos and can get something out of the box quickly, and then the reality of the implementation kicks in – processing customer data, at scale, processing personal data inside an agentic or Gen AI system – it’s complicated. So you need to have a layer of foundation that is already very strong and you need to understand your data. And I always pitch it like this when I put on my Cloudera ‘sales hat’ – you need to understand your data from ingestion all the way to your monitoring, and Cloudera is the only one to do this on-prem and on the cloud. This on top of a layer observability that we provide via the Octopai observability tools, then you get a full view. That’s where we come in.
Q: Is it a challenge for you to put the bank’s minds at ease to say Cloudera can do everything you need to do? Or are they still saying they’re not sure?
AC: No, I think they do understand the message about doing it all on a single platform. It’s cheaper to operate than having five platforms – one to do your data processing, your model training, your prompt engineering, your agent development, and another to do your monitoring. They do see that there is value in having everything in a single place. And when you have to manage access rights across five different systems to just develop a modern system, that’s obviously a further cost.
Q: Most banks are slowly moving away from all their old legacy systems, but some banks still have some legacy systems. How does all this new technology connect with the old? Or do you say you need to start fresh using all this new technology and get rid of your legacy systems?
AC: There are some banks that have done it and completely rebuilt from scratch, but I would say this is oly 2% of the banks. Most banks are doing more progressive migration, or a progressive rebuild of things towards a more modern platform.
One example on how we can enable this is that if you have a legacy data warehouse or legacy system, we can offer Apache Trino as a way to connect with it. When I was at OCBC, we utilized Apache Trino because there was a legacy data warehouse that according to IT would have apparently cost a lot of money to migrate to Cloudera. Because other departments were involved and it was not in my department, I couldn’t tell them to shut it off because I didn’t want to use it anymore. But when you implement products like Apache Trino, you are still able to query them and slowly migrate the data at your own pace, while also adding some good economies of scale. So this kind of tool is quite useful.
Also, products like Octopai – we had a customer that used it to do an offload from a data warehouse because the first thing to do when you want to migrate a system is also to understand the data lineage and how it’s fitting in.
So we do see a small and slow migration working well because that’s how banks really operate.
Q: There are assumptions that there’s going to be lots of jobs disappearing in the financial services sector because of AI. What do you think?
AC: Actually, I have a very different perspective. If you look at the evolution of the number of jobs in financial services in the last 20 years, it has gone down a lot mainly because of digitalization and the fact that we don’t physically go to bank branches that often anymore. In Europe, many of the branches have simply disappeared. And that’s really a reality that has been happening for the last 20 years or so. Now, whether we label them as ‘AI-related’ layoffs or not, it creates new opportunities. IT departments and AI departments of financial institutions are growing, at around 20% to 50% year-on-year over the last couple of years
Are there going to be more layoffs? Yes, there will be more layoffs because some institutions do not want to retrain or re-skill their people. But a lot of them are also just going to be moved into different types of roles.
When it comes to the physical interaction in financial services, there has not been a large shift caused by AI in that space.
Operations within banks is already extremely tight and controlled. Yes, it’s going to keep reducing, but that has been ongoing for the last 20 years. What some are predicting is that with big banks their operations department will disappear because it will all be done by AI. This is absolutely not happening. The attrition rates in most operation departments of banks is high at around10% percent per year, which does give financial institutions a lot of leeway to reduce headcount without firing. Firing can be seen as quite negative and can always be managed better in my opinion, because the attrition rate is actually quite high within the operations department of French institutions.
Let’s look at entry-level level jobs. How do we change the mindset of these people, especially in basic research jobs. Imagine you have someone starting at an investment bank and they see more senior people than juniors.
And what’s happening when I talk with some more senior people, they tell me that ‘in the past I would have given the work to my junior, but now I just go to AI and AI will do it for me’. So this part is a bit more troublesome, because how do you get senior people in the future if you are not training the junior ones now?
There’s a part where some banks might shoot themselves in the foot by not thinking about their own future. So in five to ten years, they might realize that within a couple of years they won’t have enough senior talent for this particular role because they decided not to hire enough juniors in that position years ago.
Q: How do you handle implementing compliance for banks?
AC: With the compliance process, banks are all subject to regulation, but they can implement it differently for various reasons – different internal constraints, different risk appetites – some banks are more willing to take on risks than others – and different business profiles also. You can’t put the same effort in a process when you are a business bank compared to being a retail bank. You will have to focus differently. And this also dictates how much effort you’re putting in to implementing AI.
As an example, in Southeast Asia a lot of the processes that I was seeing were like 5 FTEs (Full Time Equivalents) and in terms of the ROI, many business leaders question: do I want to put on two data scientists and an AI engineer to reduce the work of five people? Not sure. But then you talk to the large American banks and you go through the same process, and the same process has 600 people doing it. Then it becomes a lot more valuable to invest into these AI solutions. So it’s different conversations we can have with different banks and it obviously dictates what we are going to recommend to them. Having Cloudera is a great because companies can build whatever they need on it. But what they need might not be an AI system, or it might be a more simplified AI system, where what a larger institution needs is going to be a fully-fledged system that is bulletproof with guardrails and a lot of compliance on it. Same rules apply, but you don’t have the same economics on it, and you need to think twice.
Q: How do you ensure fairness and transparency in AI systems? Especially in dealing with sensitive data – like customers data – which the banks want to protect at all costs.
AC: AI governance is part of the AI building process, and many banks are now realizing that it cannot just be an AI governance department within the bank. They cannot do this anymore like what we had in the past, so it has to be integrated into the AI building process by empowering AI engineers and data centers to make the right decisions. That’s the first point.
The second point is back to data lineage and the data understanding that we need to have when building something. We are far from the day when an AI agent is going to crawl through the bank database to give an answer. So we need good lineage and traceability of the data that we have. Using Octopai is one possibility. That’s the first solution. Then good AI practice means also integrating solutions that complements others. Cloudera provides out of the box observability, and observability is the first step about governance. Then you need to integrate guardrails. Cloudera’s new partner Galileo provides real-time guardrails, which means that on the fly, when there’s an LLM response or an agent response, it can block the response if it feels like you are leaking some personal data on a chatbot that you should not be leaking. So I think there’s set of tools that need to be put in place, real-time guardrails, offline evaluation, observability, and data lineage is really how you are getting the full sector of what is being done. But still, we are also back to select the right use case. If you select the use case that is not feasible, that uses too much personal data, there’s a high likelihood that you’re not going to pass compliance, you’re not going to pass anything, and you need to first establish a lot of trust on what your practices are and how you are building around this.
Q: In recent times, many fintech companies that have been using their technology for years, are now re-branding it as “AI”. But is it really AI? Or are they just saying that because ‘AI’ is the newest buzzword?
AC: There is a lot of hype and marketing around AI right now. We have been doing machine learning, data science and AI for many, many years. So I have seen the various levels of hype going on. Obviously some of the companies are just trying to catch up with it all.
The reality for startups is that if you want funding, you need to have AI in your in your pitch. But they are also exploring a lot of AI solutions that they are potentially developing. Look at how many startups are now coming up in wealth management – and particularly for accredited investors – where in the past this was really an area for large banks and private banks to offer their products. Now they have chatbots, some AI-driven solutions and some are using AI to do some investments. It’s nice that you can start talking to your portfolio and an agent can help you understand what’s happening and you can ask ‘Why was my portfolio down?’, ‘What news is related to my portfolio?’, which was all very cumbersome in the past. They are not AI-driven products, but AI is definitely empowering the consumer experience much more.
