Gagan Kathuria

Early Fraud Detection in Auto Insurance In Insurance space, especially auto insurance, the most damaging risk is of a claims fraud. By the time it’s discovered (mostly after the payout), the damage is already done. It’s rarely a big indicator, but the small inconsistencies that slip through. One process where I think an early detection would be really helpful is the first notice of loss (FNOL). When someone reports a claim, most systems treat it as routine unless something obvious is off. AI can help us with this issue. AI can scan for patterns like how often someone has changed their coverages, subtle patterns in claimant's reports, if the accident timing and location matches with public data. I have seen claims that look fine on the surface, but if you go in details, turn out to be fraudulent. AI could catch those early. The challenge is not to overwhelm the claim adjudicators with false alarms. If every odd detail is flagged, people will start ignoring alerts completely. Building in a layered alert system - along with the yes/no flag, a risk score like "low, medium, high" with simple explanations attached. For instance: “This claim shares a contact number with three other previous suspicious claims”. Also, feedback loops should be added too. If an adjuster disagrees with the AI’s flag, that input should go back into the system to help it learn and get better. In short, it is not just about catching the fraud. It’s about making intelligent systems which learn and grow with time to spot risks earlier, and let examiners focus on real customers who need help. Here, AI can help in early risk detection by helping the claim adjuster, not replace him/her.

Apart from the obvious parameters like volume forecasts and shift timings, there are several other ways in which AI can be useful in work allocation. Some of these are listed below: Well being of the employees: integration with well being devices like smart phones, watches and smart rings can help work allocation AI can check these metrics before assigning the work. Cross-Training: AI work allocation system can easily monitor the type of cases that are being assigned to different employees in a team. It can start assigning some simple yet varied task to create a corss trained team. Forecasting: AI model can create patterns using the historical data and predict surge periods. It can be also integrated with holiday calendars and other tools to predict the number of people required to complete the task over next few days. AI can ensure the fairness via: Shift time Balance: AI can easily plan the shifts for all the employees and distribute day time and night shifts equitably. Balance Work Assignment: If the AI is connected someway to the ERP where it can gather the AHT for different type of transactions. The work assignment could be very fair. Bias audits: Audits to monitor bias could be done at regular intervals to monitor and keep the AI bias free. AI will eventually be able to improve efficiency and at the same create a trust rich environment in the team.

Every interaction feels personal with AI due to many factors. Some of these are listed below. Randomness: One can choose the randomness in GPT to create varying answers, this is done exactly to make the users feel that they ain't talking to a bot. Hence, it gives the interaction a human touch. This feature in AI is defined by the term ‘temperature’. Higher the temperature, more the randomness. Historical Data: The AI keeps on learning from the interactions. It adapts to what the user is looking for and also from the external world. This learning makes the AI feel more and more personal as more interactions take place. Transformers and Sentiment Analysis: The GPTs use transformers to drive the tone, meaning and context of a conversation. They can also understand when someone is angry/happy/sad via sentiment analysis. Both these tools help them be more accurate and adaptive while responding back. Omnichannel presence: AI reads through all the interactions and touchpoints like emails, chats, notes, etc. This makes them more personalised. They really know you well with all this data. This in turn helps AI become as personal as possible. Pattern Finding: We all know AI is super fast at crunching numbers and processing data in this new era of smarter and faster computational power. With the omnichannel presence, AI captures data at each and every touchpoint, but it does not stop there. AI uses that data and its innate computational power to create patterns out of it. Ex: I play songs on youtube while driving to office. But I never play songs when I am home, I only listen to news or watch some reels. Youtube is smart enough to know this pattern and recommend me songs when I am not home and news when I am in my home. When we talk about an organisation and particularly a customer interaction using AI, know that AI has all the above ability to become more and more personal. But, the data points we collect here can also invade his/her privacy. There are some points to keep in mind while utilising AI as a customer answering agent, these are: Data capture consent: All the data points and their source should be made clear to the customer upfront. Also, only data that is relevant to the conversation (not personal) in nature must be captured. Data Usage: Along with what data is being captured, it is very important to inform the customer about the way it will be used. Prompt Guardrails: In the backend of an AI agent one can add some guardrails. Like not inferring finances, health, other very personal things when replying back to a query. Remember, only the relevant data should be collected, which is data required to sensibly answer the question. And true picture about data collection should be presented to customer to add a value without infringement of privacy.

AI can definitely become a trusted advisor for Leaders. AI can do data crunching like no human and even extract trends more efficiently from that data (much better and faster than a human); but their are other aspects like ethical judgement, human touch and leadership which AI cannot replace. In short, AI can become a trusted partner and not replace a human. AI as a trusted advisor Decision making: Any amount of data can be analyzed by AI in few minutes. AI can find out the monthly/yearly trends, risks, market outcomes in a much faster and efficient manner than any human. Mitigating bias: AI can identify bias via the organizational data and help companies become more inclusive. Automating mundane tasks: Al can automate repetitive tasks, freeing up the time for leaders to focus on more strategic things. Sentiment Analysis: sentiment analysis can be easily done by AI and help leaders gauge the customer sentiments, understand communication patterns, and flag potential issues. Same goes for employees to understand the morale of the team. The human elements which are not replaceable are critical thinking, reading the room and ethical judgement.

Bias in, Bias out in AI simply means the responses/text/images being generated by our GPTs inherit the bias (Bias-in). Since these models also learn form the responses they give, Bias gets further into their veins (Bias-out). This Bias is not a technical issues, it is a dataset issue. The dataset that the models are fed are biased in the first place, the bias could be of any type, human races, ethnicity related, societal prejudices. Example: Visa for US sees a much higher rejection rate for Indian applicants compared to Australian. If the historical data for these applications is fed into an AI model, which is asked to start approving rejecting the applications now, it will inherently be rejecting more applications for Indians than Australians. Some of the major reasons for these biases are: Algorithmic design flaws: Bias can also be introduced through algorithm design, where certain variables—like zip codes—can act as proxies for race or socioeconomic status, leading to discriminatory outcomes. Training data bias: The visa rejection examples we covered above is a training data bias. It is amongst the most common type of bias in AI today. Human bias: The labeling provided to the AI models can be biased to the tune of biasness in developers' assumptions. Incomplete samples: If the data used to train the AI model does not represent the population in the correct manner that bias will inevitably infuse into the model outcomes. The underrepresented group will not have the accurate outcomes. Training loops: The decisions made by AI can be fed back to the AI dataset, which in turn can further worsen the bias in an already biased model. How can we break the bias cycle inclusive data collection: Use of data sampling exercises can be used to correct the data that is fed into the AI models, remember if you feed in right, the right outcome will come. Bias measurement tools: Their are tools available in the market to test for biasness, test for bias regularly. Source and Quality scores: Show the source info and quality scores in output results so that the end user has the exposure to input taken. Bias training and awareness: developers and users should be educated about the types of biases and the ways to avoid them, this will help build better solutions.