Search

EP-3475911-B1 - LIFE INSURANCE SYSTEM WITH FULLY AUTOMATED UNDERWRITING PROCESS FOR REAL-TIME UNDERWRITING AND RISK ADJUSTMENT, AND CORRESPONDING METHOD THEREOF

EP3475911B1EP 3475911 B1EP3475911 B1EP 3475911B1EP-3475911-B1

Inventors

  • NAYAK, NITINCHANDRA
  • OLSEN, Jayne
  • SREE NANDY, Nanditha
  • ABROKWAH, Stephen Ofori
  • MARTIN, Joy
  • YUNG, JIMMY

Dates

Publication Date
20260506
Application Date
20160622

Claims (20)

  1. An automated, real-time mortality classification and signaling system (1) with autonomous operation for real-time risk assessment, and adjustment based on an automated selective multi-level triage process (81, 82, 83) relying on electronic and technical means, process flow and process control/operation, wherein risks (913, 923, 933) associated with a plurality of risk-exposed individuals (91, 92, 93) are at least partially transferable from a risk-exposed individual (91, 92, 93) to a first electronically automated insurance system (2) and/or from the electronically automated first insurance system (2) to an associated second electronically automated insurance system (3), wherein a risk (913, 923, 933) is related to the probability for the occurrence of a life risk event in relation to a risk-exposed individual (91, 92, 93) based on defined physical measuring parameters to detect the occurrence of a risk event at the risk-exposed individual (91, 92, 93), wherein the system (1) comprises a table (10) with retrievable stored risk classes (101, 102, 103) each comprising assigned risk class criteria (110, 111, 112), wherein individual-specific parameters (911, 921, 931) of the risk-exposed individuals (91, 92, 93) are captured relating to criteria (110, 111, 112) of the stored risk classes (101, 102, 103) by means of the system (1) and stored to a repository unit (6) and wherein a specific risk class (101, 102, 103) associated with the risk (913, 923, 933) of the exposed individual (91, 92, 93) is identified out and selected from said stored risk classes (101, 102, 103) by means of the system (1) based on the captured parameters (911, 921, 931), characterized in that the operation of the system (1) is dynamically controlled, monitored and steered by a control module (7) with an associated machine learning-based pattern recognition module 8), wherein individual-specific parameters (911, 921, 931) of the risk-exposed individuals (91, 92, 93) comprise at least individual-specific parameters (916, 926, 936) indicating captured self-declaration of smoking or non-smoking of the risk-exposed individuals (91, 92, 93) and medical condition data based on medical screening tests and family history data of impairments of the risk-exposed individuals (91, 92, 93) captured by capturing or measuring devices (914, 924, 934), in that upon triggering (71/711) the individual-specific parameters (916, 926, 936) indicating a captured self-declaration of smoking (8111, 8112, 8113) of a risk-exposed individual (91, 92, 93) by means of first trigger parameters (7111, 7112, 7113), the risk-exposed individual (91, 92, 93) is automatically assigned to a first triage channel (81), and upon triggering (71/712) individual-specific parameters (916, 926, 936) indicating captured self-declaration of non-smoking of risk-exposed individuals (91, 92, 93) by means of second trigger parameters (7121, 7122, 7123), the triggered individual-specific parameters (911, 921, 931) are processed by the machine learning-based pattern recognition module (8) automatically assigning risk-exposed individuals (91, 92, 93) with detected non-smoking patterns (8211, 8212, 8213) to a second triage channel (82) as predicted non-smokers, and automatically assigning risk-exposed individuals (91, 92, 93) with detected smoking patterns (8311, 8312, 8313) to a third triage channel (83) as predicted smokers, in that the machine learning-based pattern recognition module (8) is based on random forest processing as an ensemble learning structure for classification, regression and prediction or based on a Gradient Boosting Machine (GBM) as a machine learning structure for regression, classification and prediction or based on support vector machines (SVM) as a supervised machine learning structure for regression, classification and prediction or logistic regression as a machine learning structure for regression, classification and prediction, wherein the probability of a binary response is estimated based on one or more of the individual-specific parameters (911,921,931) as predictors, wherein the captured input parameter are non-actual-smoking related measuring parameters at least comprising demographic data and employment-related data and previous tobacco usage parameters and medical condition parameters, family history data and residence location linking the applicant to characteristics of the community the applicant resides in, in that for detected risk-exposed individuals (91, 92, 93) in the third triage channel (83), the system (1) requests and captures laboratory-scaled individual-specific parameters (915, 925, 935), wherein the laboratory-scaled individual-specific parameters (915, 925, 935) are measured by means of laboratory measuring devices (914, 924, 934), and the laboratory-scaled individual-specific parameters (915, 925, 935) are triggered for measured smoking and not-measured smoking, in that for the real-time risk assessment, a relative mortality factor (918, 928, 938) is measured based on the captured risk-related individual data (911, 921, 931) and the measured smoking or non-smoking parameter corresponding to the assigned channel (81,...,83), wherein the relative mortality factor (918, 928, 938) is measured based on the captured individual's specific parameter (911, 921, 931) assignable to corresponding risk class criteria (110, 111, 112) of the risk classes (101, 102, 103), wherein the risk class criteria (110, 111, 112) comprise at least a risk class criterion (101) indicating smoking or non-smoking, and wherein for risk-exposed individuals (91, 92, 93) in the first triage channel (81), the risk class criterion (101) indicating smoking or non-smoking is automatically set to smoking, for risk-exposed individuals (91, 92, 93) in the second triage channel (82) to non-smoking, and for risk-exposed individuals (91, 92, 93) in the third triage channel (83) according to the laboratory-scaled, measured smoking or non-smoking parameters (915, 925, 935), in that based on the real-time risk assessment by means of the measured relative mortality factor (918, 928, 938), the risk associated with the risk-exposed individual (91, 92, 93) is transferable from the risk-exposed individual (91, 92, 93) to a first insurance system (2) and/or from the first insurance system (2) to the associated second insurance system (3), in that for transferring a risk associated with the risk-exposed individual (91, 92, 93) from the risk-exposed individual (91, 92, 93) to a first insurance system (2) and/or from the first insurance system (2) to the associated second insurance system (3), an appropriate activation signaling is generated by the automated system (1) by means of the control circuit (7) and transmitted to the first insurance system (2) and/or to the associated second insurance system (3) and wherein the risk transfer is mutually synchronized between the first and second insurance system (2/3), and in that the system (1) is operable in an ongoing validation process diverting a definable percentage of the risk exposed individuals (91, 92, 93) with detected non-smoking patterns (8211, 8212, 8213) to the third triage channel (83) requesting and capturing laboratory-scaled individual-specific parameters (915, 925, 935), and comparing the captured laboratory-scaled individual-specific parameters (915, 925, 935) against predicted smoking or non-smoking patterns and re-learn the machine-learning based pattern-recognition module (8) and predictive model if comparison indicates high error rates.
  2. The system (1) according to claim 1, characterized in that the machine learning-based pattern recognition module (8) is based on random forest processing as an ensemble learning structure for classification, regression, and prediction, wherein the pattern recognition module (8) operates by constructing a multitude of decision trees during training and outputting the class that is the mode of the classes or mean prediction as regression of the individual trees.
  3. The system (1) according to one of claims 1 or 2, characterized in that the machine learning-based pattern recognition module (8) is based on Gradient Boosting (GBM) as a machine learning structure for regression, classification and prediction, wherein the pattern recognition module (8) operates on a set or ensemble of weak prediction structures using classifiers which are only slightly correlated with the true classification and optimization by means of an arbitrary differentiable loss function.
  4. The system (1) according to one of claims 1 to 3, characterized in that the machine learning-based pattern recognition module (8) is based on support vector machines (SVM) as a machine learning structure for regression, classification and prediction, wherein for given sets of individual-specific parameters (911, 921, 931) of the risk-exposed individuals (91, 92, 93), each is marked for one of two definable categories, and wherein the pattern recognition module (8) builds a structure by assigning new sets or ensembles into one category or the other, making it a non-probabilistic binary linear classifier.
  5. The system (1) according to one of claims 1 to 3, characterized in that the machine learning-based pattern recognition module (8) is based on logistic regression as a machine learning structure for regression, classification, and prediction, wherein the probability of a binary response is estimated based on one or more of the individual-specific parameters (911, 921, 931) as predictors.
  6. The system (1) according to one of claims 1 to 5, characterized in that the system (1) comprises one or more of the first risk transfer systems (2) to provide said first risk transfer based on first risk transfer parameters (211, 212, 213) from at least some of the risk-exposed individuals (91, 92, 93) to the first risk transfer system (2), wherein the first risk transfer system (2) comprises a plurality of payment transfer modules (22) configured to receive and store (23) first payment parameters (221, 222, 223) associated with risk transfer of risk exposures (9) of said risk-exposed individuals (91, 92, 93) for pooling of their risks (913, 923, 933).
  7. The system (1) according to claim 6, characterized in that via the machine learning-based control circuit (7) of the system (1), risk-related data (911, 921, 931) captured from the risk-exposed individuals (91, 92, 93) are processed, wherein said first risk transfer parameters (211, 212, 213) and correlated first payment transfer parameters (221, 222, 223) are generated by means of the machine learning-based control circuit (7) and transmitted to the first risk-transfer system (2), and wherein, in the case of triggering the occurrence of one of defined risk events (721, 722, 723) associated with transferred risk exposure (913, 923, 933) of the risk-exposed individuals (91, 92, 93), the occurred loss (917, 927, 937) is automatically covered by the first risk transfer system (2) based on the first risk transfer parameters (211, 212, 213) and correlated first payment transfer parameters (221, 222, 223).
  8. The system (1) according to one of claims 1 to 7, characterized in that the system (1) comprises a second risk transfer system (3) to provide a second risk transfer based on second risk transfer parameters (311, 312, 313) from one or more of the first risk transfer systems (2) to the second risk transfer system (3), wherein the second risk transfer system (3) comprises second payment transfer modules (31) configured to receive and store (32) second payment parameters (321, 322, 323) for pooling of the risks (9) of the first risk transfer systems (2) associated with risk exposures transferred to the first risk transfer systems (2).
  9. The system (1) according to claim 8, characterized in that the second risk transfer parameters (211, 212, 213) and correlated second payment transfer parameters (311, 312, 313) are generated by means of the machine learning-based control circuit (7) and transmitted to the second risk transfer system (3), wherein the occurred loss (917, 927, 937) is at least partly covered by the second insurance system (3) based on the second risk transfer parameters (311, 312, 313) and correlated second payment transfer parameters (321, 322, 323).
  10. The system (1) according to one of claims 8 or 9, characterized in that the first and second risk transfer parameters (211, 212, 213/311, 312, 313) and the correlated first and second payment transfer parameters (221, 222, 223/321, 322, 323) are dynamically adapted and/or optimized by means of the machine learning-based control and signaling circuit (7) based on the captured risk-related individual data (911, 921, 931) and laboratory-confirmed individual-specific parameters (915, 925, 935) and the related assignment of the risk-exposed individuals (91, 92, 93) to the respective first, second or third channel (81/82/83), and based on the pooled risks (9) of the first risk transfer systems (2).
  11. The system (1) according to one of claims 8 to 10, characterized in that the first and second risk transfer parameters (211, 212, 213/311, 312, 313) and the correlated first and second payment transfer parameters (221, 222, 223/321, 322, 323) are dynamically adapted and/or optimized by means of the machine learning-based control and signaling circuit (7), further based upon measuring the cost impact of prediction errors by misclassification of the risk-exposed individuals (91, 92, 93) (smokers as non-smokers) in comparison to savings from no lab testing for majority of applicants.
  12. The system (1) according to one of claims 1 to 11, characterized in that by triggering predefined smoking or non-smoking detection pattern parameters (8211, 8212, 8213/8311, 8312, 8313) in the captured individual-specific data (911, 921, 931), additional individual-specific parameters are requested by the system (1) and transmitted to an independent control unit (4), wherein only upon capturing the transmission of a check back confirmation of the control unit (4), the automated mortality classification and underwriting system (1) accepts a possible risk transfer for the individual (91, 92, 93) for the classes (101, 102, 103) by transmitting appropriate accept or decline data.
  13. The system (1) according to one of claims 1 to 12, characterized in that the risks associated with a plurality of risk-exposed individuals (91, 92, 93) are at least partially transferable on an optional basis by means of the automated mortality classification and underwriting system (1) from a risk-exposed individual (91, 92, 93) to a first insurance system (2) and/or from the first insurance system (2) to an associated second insurance system (3), if the exceedance of a predefined uncertainty threshold is detected based upon the detected non-smoking patterns (8211, 8212, 8213) and/or detected smoking patterns (8311, 8312, 8313).
  14. The system (1) according to one of claims 1 to 13, characterized in that each of the risk classes (101, 102, 103) of the table (10) with retrievable stored risk classes (101, 102, 103) is associated with at least one financial product accessible in a dedicated data store, wherein the system (1) determines an expected occurrence rate for each of the risk classes (101, 102, 103), wherein the system (1) divides the expected occurrence rates by an average rate and determines a relative risk ratio as relative mortality factor (918, 928, 938) for each of the risk classes (101, 102, 103) based on the data relating to the criteria (110, 111, 112) associated with said risk classes (101, 102, 103), wherein the system (1) calculates correlated risk ratios between at least two of the risk classes (101, 102, 103) that are identified in said step of identifying and determining a dependence between the at least two different risk classes (101, 102, 103), wherein the system (1) compares the relative risk ratios and the correlated risk ratios with empirical data and generates comparative risk data to characterize the relative risks associated with the plurality of products, wherein the system (1) corrects the relative risk ratios if the comparative risk data is outside a defined range compared with the empirical data, and wherein the generated activation signaling is adapted based on the corrected risk ratios.
  15. The system (1) according to claim 14, characterized in that for captured individual-specific parameters (911, 921, 931) of the risk-exposed individual (91, 92, 93) comprising at least age and gender and face amount as risk-related individual data (911, 921, 931), parameter requirements and/or ranges are generated via the system (1) for a client-specific life or financial product with a positive net present value (NPV) given by the measured sum of the present values (PV) of incoming payment transfers (221, 222, 223) to the first insurance system (2) and outgoing payment transfers covering the occurred loss (917, 927, 937) at a risk-exposed individual (91, 92, 93).
  16. The system (1) according to one of claims 1 to 15, characterized in that the system (1) comprises means to automatically negotiate the risk class criteria (110, 111, 112) between the first insurance system (2) and second insurance system (3), wherein the generated activation signaling is adapted based on the negotiated risk class criteria (110, 111, 112).
  17. The system (1) according to one of claims 1 to 16, characterized in that said one or more risk classes (101, 102, 103) are associated with one or more risk class criteria (110, 111, 112), and the system (1) further modifies one or more of said criteria (110, 111, 112) and re-determines the relative risk ratio, and determines an impact of said modification on the relative risks (918, 928, 938) associated with the products.
  18. The system (1) according to one of claims 1 to 17, characterized in that one or more of said risk classes (101, 102, 103) are associated with different criteria (110, 111, 112), and the system (1) further compares the risk classes (101, 102, 103) based on said relative risk ratios.
  19. The system (1) according to one of claims 1 to 18, characterized in that the system (1) further redefines one or more of said risk classes (101, 102, 103) based on the relative risk ratio.
  20. The system (1) according to one of claims 1 to 19, characterized in that system (1) further determines a separate relative risk ratio for sub-groups of risks.

Description

Field of the Invention The present invention relates to automated life and/or mortality classification, signaling and automated underwriting systems for real-time risk assessment and adjustment. Based on the real-time risk assessment and adjustment, specific risks associated with a risk-exposed individual are transferable from the risk-exposed individual to an automated insurance system by means of an expert system providing a fully automated underwriting risk transfer process based on the expert risk assessment and forecast classification. Background of the Invention The problems associated with risk transfer and risk pooling are integral elements in the operation of life insurance systems. By grouping individuals' risk, the insurance systems are able to cover losses based on possibly future arising risks, out of a common pool of resources captured by the insurance systems from associated individuals for the transfer of their risks. However, in order to maintain some degree of equity among individuals exhibiting different mortality risks, i.e., in order to derive a balance between a specific individual's transferred risk and the amount of its resources pooled in return, the insurance systems must capture, assess and classify the individual's risk according to appropriately selected or filtered criteria and accepted characteristics. Automated classification is the process of assigning an input pattern to one of a predefined set of classes. Classification problems exist in many real-world applications, such as medical diagnosis, machine fault diagnosis, handwriting character recognition, fingerprint recognition, and credit scoring, to name a few. Broadly speaking, classification problems can be categorized into two types: dichotomous classification and polychotomous classification. Dichotomous classification deals with two-class classification problems, while polychotomous classification deals with classification problems that have more than two classes. Classification consists of developing a functional relationship between the input features and the target classes. Accurately estimating or forecasting such a relationship for future events is key to for precise classifier systems. Instrument or object underwriting, such as requests for human risk transfers, financial credits or loan risk transfers, catastrophe risk transfer or liability risk transfer is another area where these classification problems exist. An automated underwriting process for a risk instrument or object may consist of assigning a given real world object, described by its risk factors and other key input parameters, to one of several risk categories (also referred to as risk or rate classes). A trained human expert traditionally performs risk instrument or object underwriting, since this is normally a highly complex and non-linear process comparable to automated weather forecast systems, cannot be automated as such. A given application of factors for the specific risk instrument or object may be compared to a variety of underwriting rules/standards sets, which are typically predefined. Using underwriting rules/standards enables the instrument or object application to be classified into one of several risk categories available for a type of coverage requested by an applicant. The risk categories can affect the payment transfer structure (e.g., in terms of amount and timing) paid for the applied object or instrument, e.g., the higher the risk category, the higher the overall payment transfer balancing the corresponding risk transfer. A decision to accept or reject the risk transfer for the instrument or object may also be part of this risk classification, as risks above a certain tolerance threshold value may simply be rejected. One problem associated with this approach in underwriting an instrument request is that there are a large number of features (individual measuring parameter, external or environmental measuring parameters, financial parameters, credit rating parameters, corporate structure parameters, market parameters) and rules/standards that underwriters must take into account in assigning the instrument request to one of several risk categories (or rate classes). With the large number of features, rules/standards and risk categories, it is very difficult and time consuming or even impossible, especially for human experts, to consider all of the information necessary to make a decision; furthermore, the results are often inadequate in terms of consistency and reliability. The inadequacy of this process becomes more apparent as the complexity of object or instrument applications increases. Another technical problem with automated underwriting processes is that the underwriting standards typically do not cover all possible cases and variations of a request for risk transfer of a real world object or instrument. The underwriting standards may even be self-contradictory or ambiguous, leading to uncertain application of the standards.