As AI automation (aka, “intelligent automation,” or IA) in financial services quickly becomes mainstream, it attracts increased stakeholder interest as firms explore the possibility of unlocking value via increased efficiency, cost reduction, and enhanced predictive capabilities.
IA in financial services brings the potential to help generate positive ROI across various operations – customer service, lending and mortgages, HR and hiring, internal auditing, and more. Yet, despite the opportunities afforded to it, the financial services industry – a more “traditional” one – has been somewhat slow to take them on.
Knowing that opportunities for intelligent automation exist in financial services is one thing, but how can leaders nail down and identify opportunities in their own business? How can a leader then scale, measure, and demonstrate value for these opportunities?
To discuss this and other related topics in more detail, we spoke to Christophe Makni, Managing Consultant for Lean AI and Automation at Basler Kantonalbank in Basler, Switzerland.
Basler Kantonalbank is one of 24 banks that provides financial services to citizens of the 26 canons (administrative divisions) in Switzerland. The bank’s personal and commercial products are numerous, with offerings from all financial services areas except for insurance.
In our approximate 20-minute long interview, we focus on three distinct topics:
- How to “start small” in AI adoption and where
- Building a framework for early-stage implementation
- Winning C-suite support by showing an early ROI
Listen to the full episode below, skim our interview takeaways, or read the full transcript below:
Expertise: AI project management and consulting using Lean AI models.
Brief Recognition: Christophe has worked in the financial services industry for the past 12 years; working as an AI-focused project manager or another management role. Since 2017, he has lectured on the topic of Lean AI at the University of Applied Sciences and Arts in Switzerland.
- How to “start small” and where: Identify easily automatable, non-complex projects first consulting an individual who should be able to explain the process or how they make a decision in a few seconds to non-technical professionals.
- How to build an early framework for implementing automation: Being able to show a framework to management that challenges the existing process, measures any potential automated process, and displays metrics for continuous improvement is advantageous to optimal automation is essential for building trust.
- How to win C suite support by showing an early ROI: Demonstrate the value of automation processes – real or potential – to stakeholders using tools that ”show” next steps, quantify value, and justify spend. These tools can include flow efficiency, heat maps, and KPIs.
First Interview Transcript
Daniel Faggella: So Christophe, I’m glad to be able to have you on the AI in Financial Services show. We’re talking about automation today. You have a proper background in automation in the financial services space. We’re going to talk about where AI fits in.
But I want to talk first about how automation is thought about in the financial services space today. Efficiencies are so important. These companies are huge. There are so many operations going on. Why is automation efficiency so important? How do we think about it already, kind of pre-AI?
Christophe Makni: So before AI, automation is all about being able to scale, you have lots of customer requests coming in. And at some point, you can classify that into simple requests and complex requests, when it’s about simple requests, where you don’t need to think too much about the case, just go and automate, because that’s the only way you can scale.
Right now, due to COVID, lots of customers are coming with requests, and we need to be able to deliver. So standard automation comes really from the efficiency part where we want to be able to scale. And AI comes into the game where you want to be able to do even more thanks to automation.
Daniel Faggella: And, it seems like automation up until recently, this would mostly have been through process improvement and then through IT. Things that we could hard code if this then that, if this then that, send it through the system – kind of go about it that way. And so rethinking processes, hardcoding, and potential solutions.
And clearly, there’s still a very important place for that. A lot of automation efforts might not require any machine learning at all. If we can figure out a rule, and we can route something, then that’s great.
But AI is making us think a little bit differently, you’re very close to this dynamic in this paradigm shift. What are those differences? What are those changes? What’s the new way to think about automation, now that AI is part of the mix?
Christophe Makni: So when you enter into the AI arena, you have basically some new capabilities, and you have to mix these new capabilities with existing technologies. So if you want to automate some really complex process, you mix different tools, you still have your traditional workflow engines and Rules Engines, which is basically the rules and the processes you were describing.
You can do a little bit of RPA if you have a system, which has no interface, and then you have to orchestrate all the things and have some algorithm learning to make decisions for you. Right, and this is where you bring into the game workflow engines to orchestrate the whole architecture, some RPA bots.
If you have some old legacy system without any interface, and using some machine learning to be able to better classify the data that you get, maybe you want to automate some process where you just want to automate the customer requests which are coming in. that’s not something that you can code in you You need to be able to train your OCR system to classify that. And that’s the way to start some automation with AI.
Daniel Faggella: Got it. So that’s a way to kind of mix and bridge the two, as we look at sort of opportunities for bridging [a] kind of automation with AI and mixing these dynamics, and we’ve done some great interviews in the past about RPA. And then the transition for [a] kind of ML-enabled RPA.
There’s some corollaries here. Where we’re going from more hard-coded, just thinking about the process, to actual learning in the process where there’s a system in place that’s going to be doing classification is going to be learning on the fly from human feedback.
So there’s this opportunity to level things up. You talked before we started recording about kind of digitization as maybe a good place to talk about this. What are you seeing happening in digitization? Also, what do you mean by the term? Because I think it’s a very broad term, so we can define it and then dive into what the lessons are there.
Christophe Makni: Yeah. So if we break it down, at the end of the day, you have lots of customer requests, to process every day, so customers are coming to you because they want your product, or they want to complain about something.
They’re using all the different channels, so they’re coming with a letter with some paper with some phone calls, emails, whatever.
So you need to absorb that and be able to provide the right answer at the right time. So this is from the physical world to the data. And this is where machine learning is coming into place in order to digitize all the inputs that you’re getting, classifying that and triggering the right process for the right request.
Give you one example, if you want to move to [a new] location, you move from New York to Boston, you don’t want to write a letter to your bank. You just want to have you banking on your mobile, put [in] your new address, click and that’s it. And you expected that to be quite automated.
So that’s what you want to achieve, like a self-service opportunity for customers to be able to do what they want immediately. And to get an answer to all the stuff, which is easy, that’s something you can easily automate. And when it’s complex, that’s where you have some human task force entering into the game, to be able to provide the right service to the customer.
Daniel Faggella: We had a previous episode about kind of the innovation process and being able to think through the right ideas that would drive results for the company. It almost feels like that strong innovation background that you have, might also be pretty important for finding opportunities for automation, because again, it feels like there’s so many.
I’m thinking about a bank, you were talking a lot about customer service. But, good lord, I mean, from lending and mortgages to HR and hiring to internal auditing before the big auditors come — there’s an almost unlimited number of things that we could potentially focus on.
For automation, what are good ways for leaders who [say] “We’ve got an exciting opportunity with automation because we can think about processing it hard-coding ways…” but we can also think about how these new AI capabilities can unlock even further automation, take that farther.
So it’s never been a more tantalizing time to kind of look at our processes. How do you recommend folks kind of survey the landscape and find those pockets where real opportunity could lie.
Christophe Makni: So here are two approaches. The first one is to have a good understanding of your business processes because this really helps if you have a process landscape where you see your core processes, and where you can even measure your automation level. And then you can just have a heat map for your management where you can display those are like the 50 value streams.
This is an automation level. And this is why we have some automation potential. So that’s the first heat map that we are using, to work on the real stuff, where we have some potential. And the second thing for me is before you automate a process, just make some due diligence to be sure that the process is really a good process.
So the first step of automating a process is not automating the process, but really, try to challenge the process, make it simple, try to eliminate stuff from the past, make it easy, and once it’s you know.
My test would be to ask a guy how he takes a decision. If the guy says, “I’m able to solve this case in one second,” then it’s something you can automate without any problem. If the person is not able to describe to you in less than a second how they solve the problem, maybe you want to go after. Maybe you don’t want to do that as the first test, maybe you want to go to do that later on.
Daniel Faggella: This is a really interesting lens to look through because I think it’s extremely important, Christophe, for our listeners and enterprise leaders and even consultants who might be listening in to, of course, not just go in with the technology hammer and nail and say “Okay, it’s time to go,” but to assess properly, to cut away the fat and then to find the real opportunities.
You brought up two important points I’m going to try to highlight. One is the idea of looking at our value streams, looking at what’s already automated at what level, and it feels a little bit subjective to do that, but it’s clearly very useful if we can see all the streams. And one of them has a tremendous amount of, let’s say, nine out of 10 kinds of automation and another one has, nothing.
It’s all zeros, it’s like we can see where the relative opportunities are, I think the heat map is a really fun idea. The second one is to go in first, before technology, and say, “Well, what is this process? Are there ways we can cut out this step, we can simplify this step, we have things come in through one stream here?”
Have, simplify the decisions that these people make. So you’re saying that when it comes time to drill down, think through that level first, and then see what’s left to handle with technology that might be the right fit for tech?
Christophe Makni: I would extend that with the first idea.
So I have a three-step process: the first step is challenging the process, simplify, standardize the process. That’s the preparation of the automation, then you go into the automation itself. And you know what, at the end of the automation, you’re not done, because the third step is [to] try to measure and improve continuously the result of your automation.
That is what software vendors would usually sell your robots, and you can automate only one step. But I want to be sure that I challenge my process, I automate it, and then I track it, and I continuously improve it. And that is kind of tricky.
Daniel Faggella: Well, you’re right. It’s tricky. And I can speak from some experience here helping pretty big financial services firms decide on vendors, even the best vendors that have raised $50 million: Sometimes it’s really hard for them to get hard benchmarks for measuring ROI. In part because number one, they’re affecting a lot of different elements of the business.
Number two, sometimes they’re only affecting one juncture of a workflow. So there’s a whole workflow and they impact one part. And it’s really hard to pull out and extrapolate how much that one part affected the time reduction versus so many other factors.
So it’s hard to measure this stuff. When you start to drill down and find those opportunities. And you decide, okay, we’re going to double down here, we think this is a strong opportunity.
It almost feels like there would have to be another whole deliberate effort to say, “Wait a second, guys, we’re challenging this, we’ve decided to go in, let’s think ahead of time, what is measurement going to look like?” I mean, it feels like it’s probably its own step hop.
Christophe Makni: Yes. And what, the things that you’re describing is something that we are measuring. And it’s really not difficult to measure for one business process, what’s the automation level? You can measure that and report that to management. And what’s the flow efficiency? And you need to have some KPIs in place, some kind of concrete for management, and then you can just put that in front of them and show this is where we have the highest potential, this is where we save time.
And at the end of each automation project, you have to report to the stakeholders, the efficiency that you have, and how much time have you saved thanks to the project, thanks to the automation. The transparency is the key – the transparency and the communication to the stakeholders.
Daniel Faggella: That’s excellent. Okay, cool. Yeah, I like that. So, we can get our heat map out, we can think process first. And that’s with your challenge step. And then when we decide to go in for automation, we’re going to also be prepared to have a strong and accountable way to do measurement, and hopefully, to see improvements over time.
And of course, machine learning is potentially what’s going to help us see those improvements over time, if we can have a fraud model that we can consistently provide feedback to and get less false positives, less false negatives. Hopefully, we can see that, or some automated responses to customers, [and] we can have a higher success rate for those low-hanging fruit questions. People keep asking in different languages, that might be another place so I can see sort of how that circle might continue there. Yeah, go ahead.
Christophe Makni: If I can add to that, it’s pretty tricky. Once you have an automated process, you still need to have some people in production, having a look at the process execution in order to keep improving the model, your AI algorithm, and to make sure that you improve the process in production. Lots of people do like the other way that puts robots in production.
The robots are performing but they’re not improving the process over time and you have zero improvements. So you want to be sure that you keep improving your stuff you can automate using workflow engines or robots or whatever you have and once you’re in production keep working on that, actually. You’re never done, it’s a never-ending story.
Daniel Faggella: So yeah, having that kind of I don’t know what we want to call it. Like it’s almost like an automation project owner of some kind who can really instill –
Christophe Makni: … Like a process owner
Daniel Faggella: … who’s really accountable for making sure that measurement is right, making sure we’re improving things in the right direction.
And I really like your idea about challenging the process first. I think it’s so critical because even when we look at AI projects – and you’ve probably seen something similar Christoph – where it’s almost like we’re using AI in a place that’s almost building technical debt and kind of handling things with more complexity than we would be required.
You used OCR and a previous reference. So OCR is great. And there are some instances where we can’t get around it, where we have vendors from a dozen countries sending us invoices and paperwork – we’re never going to get them all to send it in the same template. We don’t control these other companies, it doesn’t work like that. So we’re going to need to have a system that drinks it in.
But sometimes, we’ve seen OCR used for internal communication stuff where people are sending reports and other things. It’s like, “Hey, guys, instead of using OCR, why don’t we stop having paper transfers between these departments and have one way to handle this report? And let’s not use OCR there.”
Christophe Makni: Yeah, I totally agree. So the first step of an OCR project is trying to get rid of the incoming documents. Yeah, that’s the first step.
Daniel Faggella: That’s a great point to sort of hammer home as we get to the end of this interview is that probably the OCR one is just one example. But almost every AI project could probably serve your challenge idea upfront because there might just be a way to get rid of this challenge without thinking tech-first at all – and I think that’s a ubiquitous lesson.
Any other closing notes? We’ve got folks listening in who work within companies of all sizes, and they’re excited and interested in automation, obviously, listening all the way through on this, this episode, any, any other quick takeaways?
You’ve shared some great tools and ways of thinking. Anything else you’d share for leaders like yourself who want to get better at automation?
Christophe Makni: So I think at some point, you need to be able to scale in a very efficient way. I tend to focus on all the stuff which is easy. That’s what I want to automate. And then through AI, I want to be able to provide more tools to my experts, so that they can make better decisions thanks to AI.
So I don’t want to automate super complex processes, I just want to have the AI engine being a good help to my expert to make the best decision possible for the customers. That’s our approach. And that’s something that all the small companies can do, and leverage existing technologies.
Always try to understand what you’re doing. When you embark on an AI journey, try to really understand the tools. Otherwise, it’s kind of difficult to convince your stakeholders to invest in these kinds of projects.
So you have to kind of have a basic AI literacy, engage your stakeholders into showing them the results and make sure that they understand what you’re talking about. Because that will be a good way where you will be able to scale that process internally.
Daniel Faggella: So instead of, “eating the whole whale,” so to speak, what are those opportunities where we think we really could win some budget and support? What are those opportunities? We really couldn’t move the needle, and maybe, beginning there, even if we’re not the biggest enterprise, that might be a way that we can kind of shuffle forward.