 Semua orang mendengar saya atau tidak? Kita okey? Terima kasih kerana menikmati saya hari ini. Terima kasih Michael, Max dan SGNVT dan SGNJSG untuk meminta saya. Ini adalah kali pertama saya di SGNJSG. Terima kasih kerana mempunyai saya. Untuk mengenai diri saya, saya mengambil alias penyelamatan dan alias saya. Saya juga mengambil alias penyelamatan di Singapura. Saya gembira melihat beberapa alias penyelamatan, tetapi kita perlu mencari lebih banyak untuk datang. Sebelumnya, kami akan mempunyai alias penyelamatan dan alias penyelamatan. Kita akan berkongsi dengan alias penyelamatan lebih banyak. Saya akan mengembangkan channelnya dan menjadikan alias penyelamatan. Jadi alias penyelamatan di sini, saya tidak akan menjadikan alias penyelamatan dan alias penyelamatan juga. Saya lebih baik membuat alias penyelamatan dan alias alias penyelamatan. B dan juga keadaan yang berlaku dengan jangka-jangka yang berlaku. Jadi, ia juga sebabnya saya mempunyai keadaan. Apa yang melewati saya adalah idea bahawa saya mungkin boleh membantu untuk meminatkan keadaan untuk mengalami keadaan, untuk membuat keadaan lebih keadaan dan untuk menghidupkan keadaan pada keadaan. Jadi, sebabnya, banyak pelajaran yang berlaku untuk mempunyai keadaan adalah sebabnya saya mengalami keadaan, banyak pelajaran yang berlaku untuk mengalami keadaan adalah sebabnya keadaan yang berlaku mempunyai keadaan yang berlaku sebabnya saya mempunyai keadaan. Ini adalah alian. Kami berlaku 126 tahun ini. Pada tahun terakhir kami, ia berlaku sekitar €135 bilion. Kami adalah salah satu institusi yang paling besar. Kami mempunyai keadaan yang berlaku sebabnya keadaan yang berlaku. Kami juga mempunyai banyak pelajaran yang berlaku untuk menghidupkan keadaan dan menghidupkan keadaan. Setiap hari saya akan melihat banyak pelajaran yang berlaku dan berkata, saya ada solusi yang boleh membantu mengalami keadaan. Dan segera, keadaan yang besar yang berlaku untuk mengambil keadaan sepanjang pelajaran. Jadi, pelajaran yang berlaku sedikit untuk digitalisasi dan juga untuk keadaan data untuk mencuba dan mengalami keadaan kita di sini. Jadi, Team Data Science telah berlaku selama 3 tahun sekarang. Apa yang kita telah lakukan adalah membuat apa yang kita nampak keadaan yang berlaku. Kita telah membuat produk data science yang berlaku sehingga kita dapat menghidupkan keadaan. Jadi, Alian Singapura adalah keadaan yang berlaku untuk Pasir Asia. Kita mempunyai pekerjaan di Malaysia, Thailand, Taiwan dan semua negara ini di seluruh keadaan. Jadi, ia berlaku dari keadaan yang berlaku dan ia sebenarnya menghidupkan keadaan yang berlaku terutama keadaan yang terlalu besar. Keadaan yang besar adalah keadaan sepanjang pelajaran, keadaan sepanjang keadaan dan sebagainya. Jadi, kita sebenarnya mempunyai keadaan yang berlaku untuk membuat keadaan yang se-ezy. Jadi, kita membuat keadaan data yang berlaku, kita memiliki keadaan model dan keadaan output. Saya yakin untuk orang yang telah mendapat keadaan data, It's not that straightforward for a lot of these models over here. And so what we really want to do is to simplify the process of this and move towards being able to serve some of this rescor through APIs. And that's what we are building. Multiple different data science products to be able to be plugged as part of the insurance value chain. So if I can go a bit more in depth into the different products that we have, we actually built our data science products based on the insurance customer journey. So if you look at this chart over here, it's actually a very simplified view of a person coming into an insurance company. So when you first buy an insurance, that's where you start. And then if you continue having good relationship with an insurance company, then you'll get happy and happy and that's where you grow your value. But if something bad happens, that's where the satisfaction actually drops and then you actually fall off. So maybe I'll take some time here to share a story. So I've asked a lot of my friends and before I joined the insurance industry also that what do you think of an insurance company? And they would say that usually insurance company will not want me to claim. So they do all kinds of things to stop me from claiming. I see no-thing like really, really, seriously here. So yeah, the first time I actually sat on the board meeting for Malaysia General Insurance Alliance. It's for the general insurance, so like the mortal insurance side. And they were discussing how can we make our customers claim more? Because they are very, very certain that with their claims processed, once the person claim, they would become a promoter for them. So you would be so sold and know that Alliance is a good brand because if I'm in trouble, Alliance will be there for me. So I think it's a different kind of mindset that we are trying to bring into now. So we first have a smart agent, product suite to help our agents become smarter in dealing with customers. So I think now even though we are moving to digitalisation, how many of you would actually buy your life insurance online? You would? Great. So I think there is still a long way to go. Hopefully we will get there. So a lot of our insurance is still being sold by agents. And that's why we have different products from recruitment. How do we identify potential suitable agents to identify agents with high potential? And also how do we identify that agents are going to leave us? So what kind of traits are they going to show? To show that they are actually going to leave us so that we can bring in another agent to take over him before he actually leaves the customer. And then there is also one part on smart pricing. Another story to share, Alliance Malaysia has actually recently pushed out a 3H product. It's for high diabetes, high blood pressure, and high... I can't remember what's the last high. So anyway, hypertension. Most of the time if you have any of these 3 highs, you will be turned away by insurance company. And now there is this product out there that allows you to be protected even when you have one of these 3 highs. And that is being done because of smart pricing. You are able to break down individuals by their risk and you don't pay for the unnecessary risk that you don't carry. And that's how you balance off. And then we have the usual which is when is the best time to contact the customers, when is the best product to sell. I would like to highlight smart health. Health is what we are doing to predict chronic diseases early. So there are signs of you claiming on particular diseases that may lead to a particular chronic disease in let's say 1 or 2 years. And that's something that we don't want to see our customers have because it's a lose-lose situation if that happens. If this customer goes on to get chronic diseases without getting worn that he's going to go there, then we're going to pay out more and this customer is going to be in bad health. So early detection of chronic diseases is something that we are working very hard on. And then the last part is more on operations. So smart deploy is putting our claim successors at the best location so that they are able to reach car accidents at the fastest time. And then the last one is on smart catch which is detecting abnormal health claims, detecting abnormal policy submissions and also mortal claims. Okey, so I'm going to go deeper into 3K studies. So smart catch risk is on predictive underwriting. So what happens is that while we are trying to improve our customer experience so we are trying to let more customers get their policies faster, we also need to manage the risk that the company is taking. So how we do that is through a straight through processing engine by putting in variables of customers at policy submission we are able to detect which customer is likely to not tell us that he ever did this. So what happens is the underwriting team found that there is this group of customers who are likely to not disclose the conditions that they have currently. Like for example if you have a chronic disease maybe one year before you would have a bead of symptoms on something is happening to you but when you actually apply for an insurance policy you may not really declare that these are the conditions that you have and these are the unnecessary risks that we want to stop and that's how we actually built this model. And I think one of the stories to share for anyone who maybe wants to become a data scientist is that we actually built like a neural network model and we define to all the parameters and stuff like that but we went back to the business people the underwriters who have got 20 years of experience and they say no why do you think that this son has high risk? None of what he is saying is high risk to me. So then we said it's neural network it works like a brain, right? So I can't really tell you how it happens. So then that's the part where we have to manage business side of things to what technical capabilities what the technologies allow us to do. So we went back and we redid the model and we did a decision tree. We did extra tools actually so it's still not like we can grab a particular branch and say because it's weighted across but it actually it's easier to visualise that this is what's happening so eventually we went with this and we got into deployment actually increase the straight through rate from 55% send 5% and we did random sampling as well and we measured all the different performances. Sorry? Ah, UVB it's the underwriters' rules so it's called underwriting black box it's a set of rules that underwriters have used for many, many, many years. So the rules can be things like if you have BMI of more than 35 then they are like high risk so it's like fixed set of things that they think are correct. So the UVB rules are things that we are trying to get rid of so in the ideal case we can just ignore the UVB error and just go straight into the data science model and so this layer over here will be gone but Ya Ya, that's the reality of things where the business side is still very insistent that I cannot just let the floodgate lose, right? I need to have some of my rules in place still to be able to safeguard some of the risk the underwriters. So ya, they have like many, many years of experience and probably like logistic logistic regression or something like that to project forward but it's not like machine learning that allows you to find pockets of patterns. So the next one that's on health claims anomaly detection so this one we did unsupervised learning there was to actually say that a claim, health claim is abnormal it's actually not easy especially for insurance company it's very hard to prove that someone is trying to fraud a claim, right? and it's bad publicity and it's not good on an insurance company so there are no true labels that we have so we did unsupervised learning for this I guess the main story around this is really that it's about coming out with a different way to do the things that it's running currently so what the usual process that they are doing is that for every claim that come in they will go through every claim and see that okay if this person is claiming 100% more than what the other people of this same condition is claiming then it must be a fraud but we all know that people who wants to fraud will not go to 100% maybe I will go I will charge 5% more but I will charge 100x5% more so how do you catch this kind of patterns and try and prove them in an unsupervised way it's something that we have worked with the team on so ya maybe I can show you how the dashboard looks like 2 minit we can we can so we just build a table dashboard so this is actually the agent quote and then this is a tracker of the performance of the agent so every month we will actually run a report run the model once and then score each agent by their abnormality compared to the rest of the agent so one is the most abnormal agent and then this dashboard can actually help claim successors take a look at which agent is actually performing worse or which agent after they actually got scolded they are starting to show better behaviour right so it's a lot of education also ya and then these are the agent details this is the different kinds of factors that we put into the model to be able to detect the abnormality and then you can actually drill into the actual cases for ya so you all pull the data from the database that the agent the agent so when they only manage something through the system then you guys get the data so when they claim for it it has to come through the system for us to process the claims okay i think my time is up that's all i have