1. This mind map shows some of the most relevant use cases of AI in Insurance industry
    1. Made by Danny Bravo
      1. Student - Master in Artificial Intelligence
      2. Three Points - OBS Business School
      3. Universitat Politècnica de Catalunya
    2. The information here was taken from diverse sources, was analyzed, organized and writen down sometimes in a textual way
      1. To find out more, you can access to the hyperlinks toward the original sources in each part of the mind map
      2. The hyperlinks can only be used downloading the original .xmind file which is avalible in the right upper corner
    3. August 2007
  2. Intro
    1. "While there have been a number of insurance entities attempting to harness AI to their processes, the level of adoption has not been as straight forward as has been suggested."
      1. OECD (2020), The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector
      2. This report benefited from the input of OECD member contributions, which includes ministries of finance, insurance supervisors, and private sector associations, and is a result of discussions in the Committee over a period of one year.
    2. The insurance industry is a competitive sector representing an estimated $507 billion or 2.7 percent of the US Gross Domestic Product
    3. Challenges that insurers face today are
      1. Promising hassle-free claim support to customers
        1. AI can help
      2. Opportunity Cost
        1. Tapping into the potential customers at the right time
          1. AI can help
      3. Right advice
        1. Providing the right set of products/services that meets customer requirements
          1. AI can help
      4. Time Consuming
        1. Fastest claim support to loyal customers
      5. Cost
        1. High-cost claims taking the edge of firms to make them get marginal profit
      6. Frauds
        1. Increasing number of False Claims and fraudulent
      7. Bulky operations
        1. Large data is being processed manually making operations bulky
    4. $45 billion in annual compensation from insurers around the world
    5. AI is growing due to the ever-increasing ―datafication‖ of business interactions, private life, and public life
    6. This work can be complemented with this paper, which include several examples of real use cases, including their results.
      1. Artificial Intelligence in Insurance Sector
        1. Naman Kumar
        2. November 2019
  3. Claims
    1. It's the formal request by a policyholder to an insurance company for coverage or compensation for a covered loss or policy event
      1. The cost of inpatient treatment makes up 30% to 40% of a typical health insurer's total budget; on average, however, between 8 and 10 percent of all claims received are incorrect."
        1. McKinsey says that Smart audit algorithms enable reliable identification of those, and only those, claims that are in fact incorrect
    2. Use case for Claims management
      1. From inflexible rule book to Intelligent algorithms that learn from historical cases and continuously evolve
        1. Data science has enabled predictions based on
          1. real events
          2. in real time
          3. using large datasets
        2. rather than samples to make the best guess.
      2. AI processes can speed up claims payment significantly
        1. Planet AI's Intelligent Document Analysis (IDA)
          1. convert heterogenous input data (e.g. images, PDF, handwritten, machine printed)
          2. into an electronic standard format (e.g. PDF, JSON) in a fully automated process
          3. 50,000 pages per hour on a single 4xGPU server
        2. companies like Claim Technology have been providing machine learning systems for claims handling
        3. Chinese insurer Ping An thinks it has found in AI the answer to the painful claims process
      3. Role of AI and related technologies
        1. Vehicle crash claim
          1. A customer involved in a motor accident, for example, can take a photo of the damage with a phone and send it to Ping An, a chinese insurer
          2. The insurer’s algorithms assess what type of car is involved, how significant the damage is, and how much it will cost to fix.
          3. It can then send over an offer to settle the claim straight away.
          4. AI tech
          5. Artificial vision
          6. Deep Learning
          7. Regression and Classification Algorithms
        2. Chatbots for customer assistance
          1. NLP
        3. RPA
          1. automate the process by which a claims
          2. assessor receives evidence, or shares the resulting outcomes,
        4. Ej. AI for improving claims at Hospitals
          1. compile and preprocess data
          2. BigData & DataScience
          3. Digital records should exist for at least the last two years, and ideally more
          4. analyze data
          5. BigData & DataScience
          6. Statistical models
          7. Descriptive Analytics
          8. At this stage, it is already possible to determine correlations between certain diagnoses and successful reductions
          9. develop the model
          10. Predictive Analytics
          11. Machine Learning
          12. Deep Learning
          13. evaluate the results
          14. model benchmarked in terms of specific metrics
          15. pilot the approach
      4. Benefits
        1. First estimates indicate that German health insurers could save in about EUR 500 million each year this way (2017)
        2. Thanks to automated prioritization, administration staff no longer have to check every claim deemed unusual, but can instead focus on those cases that have the greatest reduction potential and the best prospects for successful intervention
          1. additional costs for redundant audit and rejection processes are eliminated, while available resources can be focused on the "right" cases
        3. the system frees up capacity among administration staff and auditors so that they can correctly pinpoint reduction potential and properly prepare intervention cases
        4. In 2017, New York-based insurance start-up Lemonade said that it had paid a claim for a stolen $979 Canada Goose jacket in just three seconds
      5. Risks
        1. there are also concerns that the rapidity can compromise the optional payment, as well as potentially being open to fraudulent claims
        2. it is not clear whether current machine learning would be able to handle complex analytical and sorting work, such as assessment of insurance claims for car crashes or burglaries. Humans had to override the computer’s decision too often in the past.
        3. In its early years, Lemonade paid a lot of claims and its loss ratio — a measure of insurance payouts as a proportion of premiums — was very high.
          1. For most of last year, Lemonade’s loss ratio was above 100 per cent
          2. suggesting that it was paying out more in claims than it was receiving in premiums.
          3. AI “may have played a role” in the high loss ratios". Lemonade CEO
          4. the average loss ratio for US home insurers in 2018 was 73.9 per cent.
        4. possible bad use of claims optimisation
          1. the practice of paying out the minimum amount that a customer will accept without complaining.
          2. it could create a “vicious circle” for the industry
          3. if customers believe that insurers will not pay out the full claim, then there is an incentive to inflate the size of the claim that is submitted
        5. The median salary of an insurance adjuster who assesses auto damage was $63,510 in 2016. It is not required anymore due to AI
  4. New Business / Underwriting
    1. Evaluate and analyze the risks involved in insuring people and assets. Establish pricing for accepted insurable risks
    2. Use cases for Personalized offers
      1. As the world's digital footprint is getting bigger, the outburst of data is facilitating a new age of personalisation
        1. Customers can receive personalized offers based on their personal risk profile and behavior, not only their age and other static information
      2. Healt insurance
        1. development of customized offers for patients suffering from chronic diseases or for identifying clinical pathways that fail to adhere to guidelines
      3. Life insurance
        1. on the birth of a child, the insurance company can remind the individual about the importance of buying a family cover or a life insurance policy to protect the future of the child
      4. AI Role
        1. predictive analytics
        2. Image recognition applied on historic assests
        3. ML & DL
      5. Related technologies
        1. BigData
        2. Cloud
      6. Benefits
        1. Reduce compensation costs
        2. Better fit for the insurance premium
        3. Inactive customers now able to take a policy
        4. Identify sales opportunities by analyzing data assets
        5. Cross selling
          1. A customer bought an IoT sensor, so he can receive a new offer to adjust his policy
      7. Risks
        1. Large loss can impact severely the insurance business
        2. Customer privacy concerns
    3. Use cases for Behavioral Policy Pricing
      1. Ubiquitous Internet of Things (IoT) sensors will provide personalized data to pricing platforms
        1. safer drivers pay less for auto insurance
          1. known as usage-based insurance
        2. people with healthier lifestyles to pay less for health insurance
        3. Smart fridge
          1. Can order food
          2. Tell me what do you eat and I tell you how healthy you could be
          3. The data source not only come from a smart bridge, but through retail food stores alliances
      2. Policy discounts to users of sensorized loss prevention technology
        1. enabling cross-selling of devices and insurance
          1. Take Neos Ventures, a company that provides smart home monitoring and emergency assistance IoT along with a home insurance policy
      3. Role of AI
        1. Use of deep learning algorithms to identify patterns on data coming from IoT sensors
        2. Predictive analytics to evaluate the most likely events/sinisters to happend and triger prevention actions
      4. Other technologies involved
        1. IoT sensors
          1. home
          2. cars
          3. body
        2. Almost any smartphone can detect what type of action the user is doing and send it to the cloud
          1. walking
          2. running
          3. in car
          4. in bicycle
        3. BigData
        4. Cloud
      5. Benefits
        1. Risk assessed in real time
      6. Risks
        1. Privacy invasion
          1. Sensors in the body
          2. Sensors and cameras in the house
  5. Policy Servicing
    1. The process to change or update a policy's coverage and details that might affect the coverage and premium of the policy
    2. Use case for Inspection
      1. analyze in-car camera feeds and detect and provide feedback on unsafe behavior such as distracted driving and texting
        1. An example is State Farm’s Drive Safe & Save platform
      2. Within the Commercial sector insurance companies carry out inspections to validate their underwriting decisions based on the exposures presented by that risk
        1. Due to the complexity and/or size of the job, inspections can often be time consuming
      3. Role of AI
        1. deep learning
          1. Classification
        2. image recognition models
          1. Object detection
      4. Other technologies related
        1. High resolution cameras
        2. Drones
        3. Cloud
      5. Benefits
        1. reduction of inspection time
        2. increased surveyor productivity with the use of AI
        3. assist the Underwriter in making informed underwriting decisions
  6. Customer experience
    1. Use case for Customer support
      1. Seamless automated buying experience with chatbots
        1. Carriers will also allow users to customize coverage for specific items and events
      2. Ej. Previously, LITA’s experts had to manually go through these documents to answer questions about regulations and compliance, which usually took several days per query.
        1. To optimize the process, the LITA team used the ton of data they had gathered through their interactions with their customers to train a question-answering NLP model.
        2. They were able to develop a system that automated a large part of the consultation process and improved the service-level agreement from five days to less than an hour.
      3. agentes de inteligencia artificial cognitiva (AI)
        1. colaboran con agentes reales en más de tres millones de llamadas, reduciendo la duración de llamadas y logrando aumentos de la tasa de resolución en la primera llamada, aumentando de 67% a 75%
        2. Debido a su compleja naturaleza, los seguros se basan en un proceso de "resolución de preguntas", razón por la cual ha estado dominado por agentes. Ahora, las compañías podrán proporcionar conversaciones con los clientes en tiempo real.
      4. Role of AI
        1. The most impacting AI technology is NLP - Natural Language Processing
        2. Deep learning
        3. Predictive and prescriptive analytics
    2. Use case for Robo-advice
      1. Automated advice could assist pockets of population that do not have access to financial advice to gain input in a more cost efficient way than a human advisor
        1. developing a financial plan addressing multiple goals
          1. retirement
          2. protection needs
          3. estate planning
          4. health/long-term care coverage
          5. investment management
      2. Role of AI
        1. ML and deep learning algorithms applied to
          1. data assets coming from the customer
          2. investment offers in the marker
        2. Predictive and Prescriptive analytics
        3. NLP for smart enough chatbots
        4. Machine learning techniques
          1. probabilistic
          2. decision tree based
          3. deep learning
          4. for "decoding" hidden patterns
          5. autoencoders
          6. deliver better accuracy than machine learning techniques with larger, more complex data sets that require more compute power
      3. Benefits
        1. provide quotes with automated advice and offerings calculated through algorithms
        2. Robo-advice has the privacy which some may feel more comfortable with given the sensitivity in discussing money matters
      4. Risks
        1. Insurance start ups such as Lemonade and PolicyGenius use AI to support their policy offerings. AI can
        2. simplify and tailor policy offerings to match the needs and financial situation of the policyholder. A number
        3. of start ups are integrating AI to their processes, and their success will affect how the wider insurance
        4. sector introduces AI into its businesses as well.
        5. 60% of consumers have expressed about purchasing coverage via chatbot
          1. "It can be a Trojan horse for denying their claims"
  7. Final thoughts
    1. It was interesting to see how the insurtechs are facing the insurance market
      1. Lemonade for example, made huge investments to be as much "smart" as possible, using AI in its internal processes
        1. Specially the claiming process
        2. After paying the price for their excesive confidence on AI for approve claims, now are in a more stable scenario
        3. And even better, with AI models fine tuned supporting the fast processes they already have
    2. The more confidence the insurers have in AI, bigger the impact if some risk is materialized
      1. The impact could go from wrong pricing for premium policies, to bias the claim results
    3. AI can be a source of bad practices in insurance
      1. Some insurancers could find out that AI can identify the minimum compensation that a customer would accept
        1. and abuse of it
        2. Customers could realize there does not make sense to buy high premium policies
        3. so, the industry could be impacted
    4. Finally, what about AI/ML insurance?
      1. AI systems can fail and generate big impacts
        1. Intentionally
          1. For example, inducing bias in the training model
          2. Thorugh and adversarial attack
        2. Unintentionally
          1. When the decisions made by the AI are technically right, buy not expected in fact.
          2. due to faulty assumptions by ML developers that produce a formally correct — but practically unsafe — outcome
      2. Uber’s self-driving car killed a pedestrian in Arizona because its machine learning system failed to account for jaywalking
        1. This is not covered by a cyber insurance
        2. Is there an AI/ML insurance to cover this sinister?