KuppingerCole's Advisory stands out due to our regular communication with vendors and key clients, providing us with in-depth insight into the issues and knowledge required to address real-world challenges.
Compare solution offerings and follow predefined best practices or adapt them to the individual requirements of your company.
Meet our team of analysts and advisors who are highly skilled and experienced professionals dedicated to helping you make informed decisions and achieve your goals.
Meet our business team committed to helping you achieve success. We understand that running a business can be challenging, but with the right team in your corner, anything is possible.
Artificial Intelligence is getting increasing attention also in Financial Services. We want to take you on a brief use case safari what we have seen and want to give a deeper look on some selected examples. Finally we want to make you aware of typical pitfalls to avoid if you start your own safari.
Artificial Intelligence is getting increasing attention also in Financial Services. We want to take you on a brief use case safari what we have seen and want to give a deeper look on some selected examples. Finally we want to make you aware of typical pitfalls to avoid if you start your own safari.
Good morning. I'm France from kg. Happy to be here with you. And we wanted to take you on a safari, basically from the use case safari in 20 minutes. So that's a challenge conference own, we trying to do it. And so if you go on a safari, maybe one of you have done this before or recently come from some occasion. You might experience the big five, the big five in Africa, South Africa. You look for animals for specific types of animals. And that's what we also want to do here with AI. We are looking for the big five and just sorted them by size.
So AI, it's almost fascinating for me. So 25 years ago, when I started mass, I was fascinated how linear optimization can her to save energy with trains and long tube. But now we have much better technology have much more opportunities.
And, and if we categorize them in five categories, we, we start with descriptive and analytics. So the ultimate question basically is what happens. So we are looking on historic data. We are looking into the path, right? To understand the, is the historic data and do the analytics on that. So the next evolution or the next more advanced technology is diagnostic analytics. So we are looking now, why did it happen? And the technology that we use here is all these correlations visualize things better in a better way, analyze the feedback.
And so the more advanced we get, the more opportunities that offer to us, and it is fascinating to me, but opportunities are there. What actually can be done. And we wanted to go onto use safari to get with you in order to see, look, this is not only theory because it's also that in use cases with our clients. So next level, predictive analytics. So now we are looking on a question what will happen. So we try to predict the future.
And we used to do regression to extrapolate things, regression analysis, but it has been now we also have technology that helps us for better recognition for better forecasting and, and give much better prediction of the future. The next evolution is prescriptive and analytics, and here, the question is what should be done. So where next evolution is that we are looking to the question, what is the best recommendation and technology can help us with self learning with machine learning and other technologies to, to predict, or to, to help us to find a decision on our own.
And if you go to the last evolutions that cognitive analytics, it's a question about what will be done. And it's most advanced questions of how and what is the best answer and decisions are not only taken by the human person, but by machines. So if we know these are the big five, so that the animals that we are looking for now, we can go on our safari and see what the experience. So this is just a random pick of some use cases.
And as we are looking for AI in finance, we find some animals that we don't are interested for example, RPA or use cases, not in the finance industry, but we also will see some use cases that are AI and finance. We have picked three of them, and I'm happy to hand it over to my colleague who will introduce these use cases to us now. So good morning, working from my side so mentioned already.
I, if you really pre overview about three special use cases, well, we want to show you have a really quick picture. We go into will go into details and for sure, just have an advertiser for the date, not possible using the AI.
And we, we heard about fraud detection already. This is big topic, which is discussed quite often at the moment. And lots of solutions are kind of solution some credit card where you decide this is right wrong. You define the rules and then you just track if there's some thought available. And we were asked two years ago already from, from a health insurance company, if it's possible to find some specific fraud, which is quite high value behind it is called.
I, I, I think it's also called an English Al prescription. It's a direction Germany police also arrive in, in English. And how does it work?
It's, it's a case where usually you go to your doctor, you you've got a diagnosed and you've got a prescription. You go to the pharmacy, get a medicine and insurance company. You pay for that. But basically you don't even need the customer side because there's no information if your pharmacy really solve the products. So basically a doctor and a pharmacist is able to just handle out a prescription.
Of course, there's, there's a customer of a patient on it, but he's not really at doctor and still the, the insurance company pay for that medicine. And there's some medicine in markets really about thousands of heroes. And so there are really cases out there. Usually they're identified by, by manual hit cause somebody saw something strange. And then there's pretty often about millions, millions of heroes.
And so the insurance companies nowadays, so us, even by law to proactively identify those, those fraud or misbehaving in the market, that's the case where, where company came to us and ask us to help with AI using those technologies to identify on product debates, those cases. And there were really a low level of, of using those technologies. And there was that expectation of people using a fancy whatever predictive model to directly predict those, those use those, those cases directly, which was not possible cause the data was not available.
And also, sorry, if data was not available, there was no, no label where you can really use some machine learning techniques where you can yeah. Use, use supervised learning things.
So that's, that's kind, if you go back to these maturity level, it's quite on the left hand side, which is it's unsupervised learning approach, which doesn't mean that it was easy because we had the challenge to really drill in the data and filter into the data using still AI, AI algorithms with an unplaced approach to find out if they Analyst within the data. So what we try to do is to handle a look from different perspective, using different algorithms.
If, if the doctors doing strange things and SIS that most of the people doing the write stuff. And then if you can use other than to cluster the data to based cluster, for example, if, if there's some unusual things happening in there and that's exactly what, what was mentioned beginning, that this was a case where this absolutely relevant, important that subject matter experts, which were the doctors and, and data scientists where directly on one table, because we need to knowledge to really bring the information into the data.
That's able to get out the noise and get an information out of those, those techniques, because the algorithms on won't give you an answer. So it's, it's necessary to really get the knowledge directly from such my expert to bring those information into data. And then we, we tried to identify from this perspective, but the most interesting part was to measure the interaction between the doctors and the pharmacies. Cause I'm a hypothesis. This is a kind of the, the distance matters. How often do you usually go to from one doctor to pharmacy?
Because usually you get you prescription and you probably will take the pharmacy right on, on the same house if it's available. And those assumption that it's normally distributed regarding the distance between the doctors and the pharmacy. And if you do different perspectives, it is interesting. Cause there are some interactions too often, which could be basically not, not right from, from the data perspective. But then again, you go back to a deception made expert, you have to look into the data, you've got an animal detected and then they have to look manually on it if it's really fraud.
Or if you can explain manually, if you find out that it's still some specialty, you have to get back that, that information into the data. And so it was quite long running, working and feature engineering together with social media experts to, to really ate that. Yeah. So the next case goes a bit further regarding, so those, those GE stuff, and it's now completely different industry. It's it's about bank and how hast change. And we've address that if, if it's possible to get information, how, how is the value of specific location, for example, bank branch.
And we build a solution, first of all, basically for the retail market using external, external data that you can purchase as yeah. Map data, of course, but also information from institutes. What's the bargaining power and specific household motion data from cell phones. How do you move from your household to somewhere else? Then it's basically possible to calculate really where the money passed by and, and the model build based on attractiveness and specification and distance.
And then we gonna ask if it's passing up those, those model from the market also to, to S in the bank sector, for example, to calculate the value of ATM machine, yes, it's going down to use cash, but still in Germany, at least we've really used lots of cash and there's still bakery. You cannot pay by credit card. And so it's still one important contact point from, from the bank to the customer. And it's on the other side, it's a quite expensive thing. So cash machines ATM are, are pretty expensive. So there's the question where to locate it a machine and there's a value in it.
And basically there's possible also with that solution, because if you model attractiveness by how easy is it to access those, those machine is a parking lot. Are there stores nearby? Are there competitor ATMs nearby? You put all those data engine model and then train the model. Then we are in a predictive case, we have the information of the existing transaction from, from ATM machines. And then you can calculate the value of a machine. You can get an information why machine's not used for example.
And also you can put any dot of a map and then the algorithm calculates you the value of an ATM if you put in those locations. So using external data, internal data, using those predicted technologies is able also traffic perspective to, to calculate while on any spot on the map with those, with those solution. So that's kind of the bigger animal already where we see predictive algorithms in it. And the next case goes into, let's say the biggest part type cognitive contract management. And we started the use case combined with the Eagle reform.
Cause there are lots of contracts out in market, and there's now a risk. What happens if this reference interest rate is not available anymore? And how do you handle those contracts and how do you have to change something in there? And it's hard manual work to really go through editing.
And we, we found out in a pre-analysis that if you do it manually by, by who's really educated, you're simply not able to be concentrated long. Cause if you read 15, not to 1000 pages, you're not able to do it or you get lazy or something. And she simply doesn't, and it's easy. It's quite easy even to train, training a model to such a task. And that's, that's based on quite a lot of different technologies. We've got OCR component to just scan the data, get, get structured data of text and the engine and LP engine that also using word bedding.
So you can integrate it to also understand the combination of word in it. And then you have the machine learning technology that basically predicts the information from the contract. If there's a before clause in it, if it's qualified or not, and if it's not qualified, what kind of information is is in. So you've got five non-qualified for scenarios, it's all in there. And there's also one other point where nothing in the contract and you have to do something manually. And that system basically classify anything, those categories.
Then if you got the final dashboard where you can see if it works well or not, if the contracts are qualified or not, and the tax passages is extracted already, they can directly have a look into it. How is it managed in the contract and how do you have to change it? And that is kind of a fully technology where you can read it all kind of contracts and, and yeah. Understand it from the machine. And we found out that it's a lot faster for you to hire in beginning because if manual labeling, but afterwards I even at least 10% human being.
And I dunno how much faster because to the manual basis it's and that machine just do it anything. So I pass over again regarding some pitfalls we often see in the market because those cases are not easy and expectations are all this going a different way. And so we have a little pitfalls where Should look up. Yes. Thank you. So basically we wanted to take you on this very brief safari to one customer case with the ATM G to one operations case, the fraud protection, which is quite typical, of course, in operations and quite good case.
And to the last case where we are more in the risk, very finance area. So interpreting contracts and, and getting our information of contracts, but very much more reliable, much faster than things like that.
But, and it couldn't be linked better to our previous speaker. There are some pitfalls and one message that we got in the previous keynote was also our first learning basically to say, look, AI on everything. That's not the right way to do. So it's not about having a blockchain use case because I want to have a blockchain use case. The sequence always should be, have identified a type of business problem type of business question.
And then the second step we look, what is the best approach to handle it and is probably AI the right thing to do, or is AI capable of helping you with this business special, this business challenge? So maybe for the second and third learning, I hand back to my colleague about just summarizing our, our, our shop safari. So if you go on safari, be aware of the pitfall, don't step into the pitfalls yourself. We have seen this several in market and we wanted to give you some takeaway from, from our, Yeah. So just pick some insights directly from, from these cases, we had the second point.
So about the expectation or expect best case is pretty wrong because in the beginning you have no idea, basically what happens. It's it's data driven. You learn by doing it. We also heard it's important to get HR and because you simply cannot plan it. But quite often, what we see is that people expect the best case from the beginning. And that's never the case because you don't even know what's possible using those data.
And you learn it when you start, which is a big challenge and also kind of different thinking in the organization, running those data through the things and the third part about technology. Yeah.
I mean, this is everybody tells you, you can use our technology, that's everything in it's plug and play, and you have a lot of different technologies in, in your, in your systems. And then it's the question, how you source it. Do you do it on a cloud structure? Do you want to build up whole on premise, not just technology, also in an organization and then to train those models, you need really, really in some use cases, lots of machine power. And so you really, and that's, that's the message we have in the next talk as well.
And we heard this morning already, you really should think about technology and how, how to change and the innovative from that perspective that you don't simply start to, let's start with an AI case. And we had that also that they just put us a local machine with talking about ma on, on it and us let's build any AI use case. And that simply doesn't work. So if you have any questions, happy to be at your disposal afterwards and heading back to our next weekend. Moderat so Stefan, thank you very much for.