FORM-2
THE PATENT ACT,1970
(39 OF 1970)
AND
THE PATENT RULES, 2003
(As Amended)
COMPLETE SPECIFICATION
(See section 10;rule 13)
"PREDICTING EARLY SURRENDER OF DIFFERENT TYPES OF CONTRACT BY A USER BASED ON MACHINE LEARNING
TECHNIQUES"
ICICI Prudential Life Insurance Company Limited, a corporation organized and existing under the laws of India of
ICICI Prulife Towers, 1089, Appasaheb Marathe Marg, Prabhadevi, Mumbai 400025, India
The following specification particularly describes the invention and the manner in which it is to be performed:
2
PREDICTING EARLY SURRENDER OF DIFFERENT TYPES OF
CONTRACT BY A USER BASED ON MACHINE LEARNING
TECHNIQUES
TECHNICAL FIELD
The present invention relates to 5 the field of machine learning, in particular to
predicting probability of early surrender of different types of contracts by a user
based on machine learning techniques.
BACKGROUND
10 [0001] Machine learning techniques find their applications in various fields these
days. One or more machine learning models can be trained using sample historical
data so that the models can be used to learn based on the sample data. Based on the
learning by the machine learning models, the models can be used to predict an
outcome.
15
[0002] One such application of the machine learning technology can be found for
identifying the propensity to early surrender of contracts by a user. The contract
may be an insurance policy (for example, a unit linked insurance policy).
Surrendering contracts before maturity period is over, may effectively cancel the
20 long-term benefits immediately and only provides an immediate reduced benefit.
There may be numerous reasons for surrendering of an insurance policy by a user,
for example a policyholder (or user) feels that policy does not give as much returns
as expected or when the beneficiaries for whom the insured bought the policy may
have expired or incapable of paying the premium or have intent to gain cash value
25 under the policy.
[0003] The conventional art fails to identify early surrender of contract by the user.
What is required in the art is the intelligent models which can analyze diverse
dataset to accurately predict a possibility of surrender by the user. The conventional
3
art also lacks techniques for identifying engagement strategies with the user
ensuring it can handle large volumes of data and provide actionable insights in a
timely manner. Therefore, the need arises to provide improved techniques to predict
the probability of surrender and categorizing the users based on their propensity to
5 surrender.
SUMMARY
[0004] The following presents a simplified summary of the subject matter in order
to provide a basic understanding of some aspects of subject matter embodiments.
10 This summary is not an extensive overview of the subject matter. It is not intended
to identify key/critical elements of the embodiments or to delineate the scope of the
subject matter.
[0005] Its sole purpose is to present some concepts of the subject matter in a
15 simplified form as a prelude to the more detailed description that is presented later.
[0006] The primary objective of the present invention is to provide machine
learning based techniques to predict the probability of a user to surrender a policy.
20 [0007] Another objective of the present invention is to provide machine learning
techniques to categorize the users based on their probability to surrendering the
policies.
[0008] In one embodiment, the present invention provides for a method for
25 predicting early surrender of different types of contract by a user. The method
comprises obtaining a first set of parameters corresponding to the user and a first
type of contract, determining whether the tenure of the first type of contract is more
than a first threshold and less than a second threshold, inputting the first set of
parameters to a first machine learning model when the tenure of the contract is more
30 than a first threshold and less than a second threshold to generate a first output,
4
predicting a probability of surrender of the first type of contract by the first machine
learning model based on the first output, obtaining a second set of parameters
related to the user and the first type of contract, determining whether the tenure of
the contract is more than a third threshold, inputting the second set of parameters to
a second machine learning 5 model when the tenure of the contract is more than the
third threshold to generate a second output, predicting a probability of surrender of
the first type of contract by the second machine learning model based on the second
output, obtaining a third set of parameters related to the user and a second type of
contract, determining whether the tenure of the second type of contract is more than
10 a fourth threshold, inputting the second set of parameters to a third machine learning
model when the tenure of the second type of contract is more than the fourth
threshold to generate a third output and predicting a probability of surrender of the
second type of contract by the third machine learning model based on the third
output.
15
[0009] In another embodiment, the present invention provides an apparatus for
predicting early surrender of different types of contract by a user. The apparatus
comprises a memory and a processor coupled to the memory to perform the
operations of obtaining a first set of parameters corresponding to the user and a first
20 type of contract, determining whether the tenure of the first type of contract is more
than a first threshold and less than a second threshold, inputting the first set of
parameters to a first machine learning model when the tenure of the contract is more
than a first threshold and less than a second threshold to generate a first output,
predicting a probability of surrender of the first type of contract by the first machine
25 learning model based on the first output, obtaining a second set of parameters
related to the user and the first type of contract, determining whether the tenure of
the contract is more than a third threshold, inputting the second set of parameters to
a second machine learning model when the tenure of the contract is more than the
third threshold generate a second output, predicting a probability of surrender of the
30 first type of contract by the second machine learning model based on the second
5
output, obtaining a third set of parameters related to the user and a second type of
contract, determining whether the tenure of the second type of contract is more than
a fourth threshold, inputting the second set of parameters to a third machine learning
model when the tenure of the second type of contract is more than the fourth
threshold generate a third output and 5 predicting a probability of surrender of the
second type of contract by the third machine learning model based on the third
output.
[0010] These and other objects, embodiments and advantages of the present
10 invention will become readily apparent to those skilled in the art from the following
detailed description of the embodiments having reference to the attached figures,
the invention not being limited to any particular embodiments disclosed.
BRIEF DESCRIPTION OF FIGURES
15 [0011] The foregoing and further objects, features and advantages of the present
subject matter will become apparent from the following description of exemplary
embodiments with reference to the accompanying drawings, wherein like numerals
are used to represent like elements.
20 [0012] It is to be noted, however, that the appended drawings along with the
reference numerals illustrate only typical embodiments of the present subject
matter, and are therefore, not to be considered for limiting of its scope, for the
subject matter may admit to other equally effective embodiments.
25 [0013] FIGURE 1 illustrates a block diagram, a system for predicting probability
to surrender of a contract, according to an embodiment of the present invention.
[0014] FIGURE 2 illustrates a simplified flow diagram of predicting the
probability of surrender of a contract using different machine learning models,
30 according to an embodiment of the present invention.
6
[0015] FIGURE 3 illustrates an exemplary embodiment for predicting probability
of surrender of a contract when the contract is a unit linked insurance policy,
according to an embodiment of the present invention.
5
[0016] FIGURE 4 illustrates an exemplary embodiment for predicting probability
of surrender of a contract when the contract is not a unit linked insurance policy,
according to an embodiment of the present invention.
10 [0017] FIGURE 5 illustrates a simplified flow chart of a method for predicting
probability to surrender, according to an embodiment of the present invention.
[0018] FIGURES 6A and 6B illustrates a detailed flow chart of a method for
predicting probability to surrender, according to an embodiment of the present
15 invention.
[0019] FIGURE 7 is a block diagram illustrating an exemplary computing device
in which one or more embodiments of the present invention may operate, according
to an embodiment.
20
DETAILED DESCRIPTION
[0020] Exemplary embodiments now will be described with reference to the
accompanying drawings. The disclosure may, however, be embodied in many
different forms and should not be construed as limited to the embodiments set forth
25 herein; rather, these embodiments are provided so that this disclosure will be
thorough and complete, and will fully convey its scope to those skilled in the art.
The terminology used in the detailed description of the particular exemplary
embodiments illustrated in the accompanying drawings is not intended to be
limiting. In the drawings, like numbers refer to like elements.
30
7
[0021] It is to be noted, however, that the reference numerals used herein illustrate
only typical embodiments of the present subject matter, and are therefore, not to be
considered for limiting of its scope, for the subject matter may admit to other
equally effective embodiments.
5
[0022] The specification may refer to “an”, “one” or “some” embodiment(s) in
several locations. This does not necessarily imply that each such reference is to the
same embodiment(s), or that the feature only applies to a single embodiment. Single
features of different embodiments may also be combined to provide other
10 embodiments.
[0023] As used herein, the singular forms “a”, “an” and “the” are intended to
include the plural forms as well, unless expressly stated otherwise. It will be further
understood that the terms “includes”, “comprises”, “including” and/or
15 “comprising” when used in this specification, specify the presence of stated
features, integers, steps, operations, elements, and/or components, but do not
preclude the presence or addition of one or more other features, integers, steps,
operations, elements, components, and/or groups thereof.
20 [0024] Unless otherwise defined, all terms (including technical and scientific
terms) used herein have the same meaning as commonly understood by one of
ordinary skill in the art to which this disclosure pertains. It will be further
understood that terms, such as those defined in commonly used dictionaries, should
be interpreted as having a meaning that is consistent with their meaning in the
25 context of the relevant art and will not be interpreted in an idealized or overly formal
sense unless expressly so defined herein.
[0025] The figures depict a simplified structure only showing some elements and
functional entities, all being logical units whose implementation may differ from
30 what is shown. The connections shown are logical connections; the actual physical
8
connections may be different. It is apparent to a person skilled in the art that the
structure may also comprise other functions and structures.
[0026] The present invention proposes to develop various machine learning
techniques that predicts the propensity of 5 a user to surrender different contracts (for
example, a linked insurance policy, or a traditional insurance policy) and also
identifies categories of users based on their probability to surrender. In the case of
Unit linked insurance policies (ULIP) policies, the ULIP become eligible for
surrender once they complete 5 years. This is the most vulnerable time in policies
10 lifecycle wherein user decide to surrender the respective policies. Once policies
vintage age is more than 6 years, the relative surrender rate of policies goes down.
Hence, it becomes imperative to monitor the policies that are going to be eligible
for surrender in their 6th year.
15 [0027] The present invention identifies the propensity to surrender once a user
reaches the end of lock-in period (i.e., 5 years) using machine learning techniques
and enables insurer to take intervening actions. The proposed machine learning
techniques according to the present invention can also be developed to predict
surrender in pre and post lock in period.
20
[0028] The invention consists of various machine learning models for different
stages of contract life cycle starting from pre-lock in stage wherein customer is
locked in and cannot surrender and also during post lock-in of the contract when
needed. These models are machine learning models like logistic regression and
25 xGBoost with early stopping models which is used to classify contracts into
different categories based on propensity to surrender using historical profiling of
similar contracts. Contracts with higher chances of surrender are classified into one
bucket as most vulnerable cohort.
30
9
[0029] The proposed solution consisting of various machine learning models like
Logistic and XGBoost Trees Classifier with early stopping model which takes input
variables from various data sources like:
• Customer Demographics
5 • Product Characteristics
• Payment details
• Channel details
• Credit score details
10 [0030] The various machine learning models also takes in as input unstructured
data pertaining to customer calls and based on the context of the call, transformed
it into a categorical variable for our supervised machine learning model. Data
sourcing was followed by an extensive EDA to identify the variables affecting
surrender significantly. Bivariate analysis gave us insights for feature engineering.
15
[0031] Various models like logistic regression, random forest, support vector
machines, KNN along with few boosting algorithms like xGboost, adaboost were
used to fit the data to predict if a policy is going to surrender. Using F1 score as a
performance metric and with aim to reduce Type-2 error, logistic regression and
20 xGBoost model was chosen as the best fit model. The selected model then classifies
customers into different categories based on their probability to surrender.
[0032] Referring to Figure 1 now, a block diagram of the surrender prediction
system 100 for predicting probability of surrender of different types of contract by
25 a user is illustrated. The system 100 identifies the probability of a user to surrender
the policy and categorizes the users based on their probability of surrendering the
policy. The system 100 comprises of an input module 110, a surrender prediction
module 120, a memory 130, and a user categorization module 140. Although, only
a few components or units have been shown in Figure 1, the components or units
30 in system 100 are not restricted to the ones mentioned here.
10
[0033] Since the system 100 can be used to predict propensity to surrender for
different types of contracts, the system can take different input depending on the
type of contract. The input module 110 receives the user demographics,
characteristics of the contract, 5 etc. For example, the input module 110 can receive
the following as input:
• Subscription related information of the contract;
• Channel Details (Channel Categories);
10 • Age & Earning, credit score of the user;
• Fund Type (Equity, Balanced or Debt);
• Whether user has checked fund value;
• Campaign Name (D2FSC or PFA);
• Nationality of the user ;
15 • User Interaction which subscribing to the contracts.
[0034] In one embodiment, different set of parameters are used for different type
of contract. For example, the ULIP type of policy use different input variables (a
first and a second set of variables) while the non-ULIP type of policies use different
20 input variables (a third set of variables). Further, variables like fund type, fund value
check are very specific and significant for predicting surrender of linked polices.
For example, users who have frequently checked their fund value in recent months,
users whose fund type is equity are found to be more prone to surrender. The
variables are not limited to the one mentioned here and may include any other
25 variables as well.
[0035] The input module 110 provides the input variables to the surrender
prediction module 120 for further processing. For processing, the surrender
prediction module 120 may use various machine learning models/algorithms such
30 as logistic regression, lasso regression, decision tree, random forest etc. These
11
models are used to fit the data to predict if a user is going to surrender a policy. In
one embodiment, the system 100 may prefer to use Logistic regression method
since the system 100 is used to predict the binary dependent variable using a given
set of independent variables. The machine learning techniques/algorithms are
5 stored in the memory 130.
[0036] The prediction module 120 contains different machine learning models for
processing the input received by the input module 110. The different machine
learning models may be selected based on the different types of contracts. The
10 prediction module 120 also takes into account the tenure of the contract to select
the different machine learning model. For example, as discussed above, the lock-in
period for the ULIP ends after the period of 5 years. Hence, the period before the
end of the 5 years represents pre-lock in stage and the period after the 5th year
represents post lock in stage. A user is monitored such that the machine learning
15 models can predict the surrender of the policy before the user is about to surrender
a policy both in pre and post lock in stage. The first machine learning model runs
from the period beginning after 37th month from the issuance of the ULIP and
before 57th month from the issuance of the ULIP. Similarly, the second machine
learning model runs after 5th year from the issuance of the ULIP. Further, a third
20 machine learning model can be selected for the non-ULIP type of policy which have
crossed 59th month of their life span till the 72nd month (6th year policies). Thus,
the threshold and the second threshold may be a period between 37th month to 57th
month from the issuance of the policy respectively. The third threshold may be after
the 5th year from the issuance of the policy. Similarly, the fourth threshold may be
25 the period after 59th month from the issuance of contract.
[0037] The first set of parameters, i.e., the variables for the first type of contract
(for example, ULIP) are input to the first machine learning model. The first set of
parameters may be used during the pre-lock in stage of the contract (for example,
30 the pre lock in stage of the ULIP). Further, the surrender prediction module 120
12
determines if the tenure of the first type of contract is more than the first threshold
and less than the second threshold. The surrender prediction module 120 uses
various algorithms such as logistic regression, random forest, support vector
machines, KNN along with few boosting algorithms like xGboost, adaboost to
predict the probability to surrender 5 the first contract based on determination that
tenure of the first type of contract is more than the first threshold and less than the
second threshold.
[0038] Similarly, the second set of parameters, i.e., variables for the first type of
10 contract are input to the second machine learning model. The second set of
parameters may be used during the post lock-in stage of the contract (for example,
post lock-in stage of the ULIP). Further, the surrender prediction module 120
determines if the tenure of the first type of contract is more than the third threshold.
The surrender prediction module 120 uses various algorithms such as logistic
15 regression, random forest, support vector machines, KNN along with few boosting
algorithms like xGboost, adaboost to predict the probability to surrender the first
contract based on determination that tenure of the first type of contract is more than
the third threshold.
20 [0039] Similarly, the third set of parameters, i.e., variables for the second type of
contract are input to the third machine learning model. The third set of parameters
may be used for the non-ULIP type of policies. Further, the surrender prediction
module 120 determines if the tenure of the second type of contract is more than the
fourth threshold. The surrender prediction module 120 uses one or more algorithms
25 such as logistic regression, random forest, support vector machines, KNN along
with few boosting algorithms like xGboost, adaboost to predict the probability to
surrender the second contract based on determination that tenure of the second type
of contract is more than the fourth threshold.
30 [0040] In one embodiment, the first, second and the third machine learning models
use logistic regression model to predict the probability of a user to surrender a
13
policy. The logistic regression model using binary classification and producing a
probability value between 0 and 1. The output can be classified either under the
class 0 or the class 1. These classified probability are fed to user categorization
module 140 for further classification..
5
[0041] The user categorization module 140 classifies users into different flagging
(Red, Amber, Yellow, Green) based on their probability to surrender. That is, the
predicted outcome 150 of the user categorization module 140 will be the
categorization of users into different flagging. Users with Red flagging have the
10 highest propensity to surrender and the users with green flagging being the least.
[0042] The output from the user categorization module 140 enables senior
managers to intervene, have a direct discussion with the vulnerable customer,
explain about the product in detail and all the benefits pertaining to staying invested
15 and not surrendering the policy. This ensures that the at-risk customers are taken
special care of, and steps are being taken to ensure they also have a better experience
and benefits from the policy. This in turn gives a sense of attachment for such
customers with the company, as they can see that they are being taken special care
of and are made sure that the right product is shared with them. Post lock-in
20 monitoring enables AUM team to focus on vulnerable customers and connect with
them and retain them with additional options like cover continuance without further
premium payment. This also leads to optimization of follow-up efforts with the
policyholders.
25 [0043] The first, the second and the third machine learning models identifies high
surrender risk policies and take proactive measures to reduce surrender rates such
as targeted retention efforts or personalized offers. The present invention helps in
planning and forecasting financial reserves and capital requirements by better
understanding potential surrender liabilities. Further, by predicting which
30 customers are likely to surrender their contracts, the company implements tailored
retention strategies, including personalized communication and incentives.
14
Enhancing the customer experience through proactive engagement based on
predictive insights can reduce churn. Also, analyzes patterns and behaviors leading
to higher surrender risks, helping in understanding customer needs and preferences
better. The present invention also helps in optimizing resource allocation for
customer service 5 and retention efforts by focusing on high-risk segments,
identifying areas for improving processes related to customer interactions and
policy management to reduce surrender rates and integrating with broader business
analytics to support strategic decision-making processes. Overall, the invention can
significantly impact insurance and financial services by improving risk
10 management, enhancing customer retention, and optimizing operational efficiency.
[0044] Referring to FIGURE 2, a simplified flow diagram of predicting the
probability of surrender a contract using different machine learning models is
illustrated, according to an embodiment of the present invention. At block 202, the
15 first, second and third set of parameters are obtained. The first, second and third set
of parameters include user and contract related preferences. At block 204, the first
machine learning model identifies a chance to surrender of the first contract after
before end of lock in period. At block 206, the second machine learning model
determines the probability of user to surrender the first contract after out of lock-in.
20 At block 208, the third machine learning model determines the probability of user
to surrender traditional policies (non-ULIP).
[0045] FIGURE 3 illustrates an exemplary embodiment for predicting probability
of surrender of a contract when the contract is a unit linked insurance policy,
25 according to an embodiment of the present invention. For the unit linked insurance
policy, 2 stages are defined- pre-lock in stage and the post-lock in stage. The prelock
in stage occurs when the ULIP policies have paid their 37-month premium.
The first machine learning model runs on the various input parameters of the
customer (such as age, gender, marital status, income, occupation, etc.), policy
30 details (such as annual premium, payment channel, fund value, fund type etc.),
customer interaction details and channel details and predicts probability of
15
surrender for these contracts when they will cross the lock-in period after 2 more
years. The model is present on DataRobot where we input the data and compute the
propensity to surrender.
[0046] Based on the model 5 output, contracts are grouped into various cohorts
depending on their risk of surrender and the details are shared with the AUM team
with recommendations. Model is run on batch processing on monthly basis for the
set bucket of contracts which are from 37th month to 57th month and the output is
shared with the AUM team respectively. At this stage, based on surrender risks
10 attached with different cohorts of the classifications from model, various
communication strategies are planned for the customers keeping them in sync with
the overall goal of financial benefit and long-term better gains on their premiums.
AUM team takes initiative to connect with customers and explain them regarding
the benefit of keeping their policy running and try to access factors which might
15 lead to surrendering of the policy for high-risk customers and accordingly help them
with all possible options from the company.
[0047] At the post- lock in stage, all the relevant data is collected for the ULIP
policies at 58th month which will be surrender eligible in next 2 months (i.e., at 60th
20 month). The input data (i.e., the second set of parameters) is obtained at the real
time basis for the all the required details such as customer demographics, policy
details which includes the fund value at that time, fund type money is invested in,
etc, customer interaction details such as number of times customer has checked his
fund value in last 6 months, has the customer connected for any type of complaints
25 regarding his policy or experience and group all the data into respective buckets as
required for consumption by the second machine learning model. The data is made
on complete one year book from the policies being surrender eligible in two months
to the policies currently in 6th year of their lifespan.
16
[0048] The post-lock in stage of invention runs from the 6th year using the input
data created for the ULIP policies in previous steps. The contracts are classified
into different colour cohorts namely Red, Amber, Yellow and Green based on the
predictive risk of surrender, with red cohort being at the highest risk of surrender
and Green at the lowest surrender 5 risk moving in ordinal fashion from Red to Green
and the data is stored in a table in the database
[0049] At the post-lock in stage of invention, the data is shared with the AUM team
with the classification of the policies in different surrender-risk buckets on which
10 they further implement their outreach program to curb surrenders and retain them.
Also, at this stage, customers with high surrender risk are prioritized for various
efforts and communications in the outreach program. The efforts of the team which
has to retain the customers and stop surrenders are concentrated on high-risk
customers and for the lower risk customers simple communications are made via
15 email and SMS. AUM team using the model results are better able to handle the
surrender and get better outputs from similar efforts concentrating them on
appropriate customers.
[0050] FIGURE 4 illustrates an exemplary embodiment for predicting probability
20 of surrender of a contract when the contract is not a unit linked insurance policy
(also called traditional policies), according to an embodiment of the present
invention. Here, input data is gathered and created for all the policies which are
traditional and have crossed 59th month of their life span till the 72nd month (6th
year policies). The data collected in then cleaned and pre-processed to convert into
25 required formats of categorical data as required for input into the model. The data
here consist of customer details, policy details, customer interaction details and
credit bureau information of the customer.
[0051] This stage of the invention occurs for running the eXtreme Gradient Boosted
30 Trees Classifier with early stopping machine learning model which is implemented
in DataRobot. The input data from the previous stage is fit into the DataRobot and
17
the Surrender probabilities are computed on the above model. Based of probability
values from the model, the policies are classified into the colour cohorts Red,
Maroon, Amber, Yellow, Green 1 and Green 2 with surrender risk moving in
descending order in the above consecutive cohorts, red being at highest risk of
surrender and Green 2 5 being at lowest risk of surrender for the Traditional policies.
[0052] At this stage of invention, the data is shared with the AUM team with the
classification of the policies in different surrender-risk buckets on which they
further implement their outreach program to curb surrenders and retain them. At
10 this stage, customers with high surrender risk are prioritized for various efforts and
communications in the outreach program. The efforts of the team which has to
retain the customers and stop surrenders are concentrated on high-risk customers
and for the lower risk customers simple communications are made via email and
SMS. AUM team using the model results are better able to handle the surrender and
15 get better outputs from similar efforts concentrating them on appropriate customers.
[0053] Figure 5 illustrates a simplified flowchart of a method 500 involved in the
process of identifying the probability of a user to surrender their policy and
categorization of users based on their propensity to surrender the policy according
20 to an embodiment of the present invention. Although the flowchart has been
explained from the point of view of a user surrendering insurance policy, the present
invention can be used to predict propensity to surrender by a user in any application
where surrender is involved.
25 [0054] The various steps involved in this process can be implemented by the system
100. At Step 510, the method comprises obtaining the first, second and the third set
of parameters from various data sources. The first, second and the third parameters
may include user related data and the contract related data. Further, the active
policies data may also be extracted that may include policy and premium related
30 fields are extracted from internal tables. At Step 520, the method comprises creating
input features that have been identified as important features for the first, second
18
and the third machine learning models and consequently are the input features to
the Surrender prediction module 120 of system 100. In detail, features like fund
type as balanced, debt and equity, fund value check, how was the user contacted in
past etc. are created. Next, different data columns has been arranged in required
format and perform binning 5 wherever needed. Here, binning is performed to
transform the numerical variables to their categorical variables to simplify the data
and to make it more manageable for analysis.
[0055] At Step 530, the method comprises providing the created input features to
10 the surrender prediction module 120 of system 100. At Step 540, the method
comprises obtaining the probability of a user to surrender different contracts based
on output from the various machine learning models. At Step 550, the method
comprises categorizing and tagging the users based on their predicted probability
to surrender the policy. The users are then flagged into Red, Amber, Yellow and
15 Green categories (RAYG) according to their probabilities. The user tagging are then
shared with the Assets Under Management (AUM) team and further they
implement their outreach program augmented by PTS model tagging.
[0056] As explained above, predicting the probability of a user to surrender and
20 classifying the users based on their probability to surrender, the AUM team can take
prevention measures to reduce the surrender rate. In detail, after identifying the
vulnerable user that can surrender at end of lock-in period with different tagging
(RAYG) two month ahead, the data will be sent to AUM team monthly and
according to their propensity to surrender, AUM team deploys various retention
25 campaigns, including connecting with users using different telecommunication
methods, conversation with investment specialists etc.
[0057] FIGURE 6A and 6B illustrates a detailed flow chart of a method 600 for
predicting probability to surrender, according to an embodiment of the present
30 invention. At step 602, the method comprises obtaining a first set of parameters
corresponding to the user and a first type of contract. At step 604, the method
19
comprises determining whether the tenure of the first type of contract is more than
a first threshold and less than a second threshold. At step 606, the method comprises
inputting the first set of parameters to a first machine learning model when the
tenure of the contract is more than a first threshold and less than a second threshold
to generate a first 5 output. At step 608, the method comprises predicting a probability
of surrender of the first type of contract by the first machine learning model based
on the first output.
[0058] At step 610, the method comprises obtaining a second set of parameters
10 related to the user and the first type of contract. At step 612, the method comprises
determining whether the tenure of the contract is more than a third threshold. At
step 614, the method comprises inputting the second set of parameters to a second
machine learning model when the tenure of the contract is more than the third
threshold generate a second output. At step 616, the method comprises predicting a
15 probability of surrender of the first type of contract by the second machine learning
model based on the second output.
[0059] At step 618, the method comprises obtaining a third set of parameters related
to the user and a second type of contract. At step 620, the method comprises
20 determining whether the tenure of the second type of contract is more than a fourth
threshold. At step 622, the method comprises inputting the second set of parameters
to a third machine learning model when the tenure of the second type of contract is
more than the fourth threshold generate a third output. At step 624, the method
comprises predicting a probability of surrender of the second type of contract by
25 the third machine learning model based on the third output.
[0060] In one embodiment, the above-mentioned functions may be implemented
using one or more hardware components present in a computing device. For
example, the computing device may include the components as mentioned in Figure
30 7.
20
[0061] Figure 7 is an exemplary computing device 700 in which one or more
embodiments of the present invention may operate, according to an embodiment.
In the system schematic of figure 7, bus 710 is in physical communication with
Input/ Output device 702, interface 704, memory 706, and processor 708. Bus 710
includes a path 5 that permits components within computing device 800 to
communicate with each other. Examples of Input/ Output device 702 include
peripherals and/or other mechanism that may enable a user to input information to
computing device 700, including a keyboard, computer mice, buttons, touch
screens, voice recognition, and biometric mechanisms and the like Input/ Output
10 device 702 also includes a mechanism that outputs information to the user of
computing device 700, such as a display, a light emitting diode (LED), a printer, a
speaker, and the like.
[0062] Examples of interface 704 include mechanisms that enable computing
15 device 700 to communicate with other computing devices and/or systems through
network connections. Examples of memory 706 include random access memory
(RAM), read-only memory (ROM), flash memory, and the like. The memory 706
store information and instructions for execution by processor 708. The processor
708 includes, but not limited to, a microprocessor, an application specific integrated
20 circuit (ASIC), or a field programmable object array (FPOA) and the like. The
processor 708 interprets and executes instructions retrieved from memory 706.
[0063] In one embodiment, the computing device 700 may be responsible for
implementing the above-mentioned steps. For example, input parameters of the user
25 may be received using the Input/ Output device 702. The machine learning models
may be stored in the memory 706 and may be implemented by the processor 708.
[0064] In present invention, the functions described may be implemented in
hardware, software, firmware, or any combination thereof. If implemented in
30 software, the functions may be stored on or transmitted over as one or more
instructions or code on a computer-readable medium and executed by a hardware21
based processor. The implementations described herein are not limited to any
specific combinations of hardware circuitry and software.
[0065] In the drawings and specification, there have been disclosed exemplary
embodiments 5 of the invention. Although specific terms are employed, they are used
in a generic and descriptive sense only and not for purposes of limitation of the
scope of the invention.

CLAIMS
WE CLAIM:

1. A method for predicting early surrender of different types of contract by a user, the
5 method comprising:
obtaining a first set of parameters corresponding to the user and a first type
of contract;
determining whether the tenure of the first type of contract is more than a
first threshold and less than a second threshold;
10 inputting the first set of parameters to a first machine learning model when
the tenure of the contract is more than a first threshold and less than a second
threshold to generate a first output;
predicting a probability of surrender of the first type of contract by the first
machine learning model based on the first output;
15 obtaining a second set of parameters related to the user and the first type
of contract;
determining whether the tenure of the contract is more than a third
threshold;
inputting the second set of parameters to a second machine learning model
20 when the tenure of the contract is more than the third threshold generate a second
output;
predicting a probability of surrender of the first type of contract by the
second machine learning model based on the second output;
obtaining a third set of parameters related to the user and a second type of
25 contract;
determining whether the tenure of the second type of contract is more than
a fourth threshold;
23
inputting the second set of parameters to a third machine learning model
when the tenure of the second type of contract is more than the fourth threshold
generate a third output;
predicting a probability of surrender of the second type of contract by the
third 5 machine learning model based on the third output.
2. The method as claimed in claim 1, wherein after predicting the probability of
surrender of the first type of contract by the first machine learning model, the
method further comprises:
10 classifying the user as a high risk user if it is determined that the probability
of surrender of the first type of contract by the first machine learning model is
above a pre-determined threshold; and
sending a prioritized communication to the user when the user is classified
as a high risk user.
15
3. The method as claimed in claim 1, wherein after predicting the probability of
surrender of the first type of contract by the second machine learning model, the
method further comprises:
classifying the user as a high risk user if it is determined that the probability
20 of surrender of the first type of contract by the second machine learning model is
above a pre-determined threshold; and
sending a prioritized communication to the user when the user is classified
as a high risk user.
25 4. The method as claimed in claim 1, wherein after predicting the probability of
surrender of the second type of contract by the third machine learning model, the
method further comprises:
24
classifying the user as a high risk user if it is determined that the probability
of surrender of the first type of contract by the third machine learning model is
above a pre-determined threshold; and
sending a prioritized communication to the user when the user is classified
5 as a high risk user.
5. The method as claimed in claim 1, wherein the first type of contract is unit linked
insurance policy.
6. The method as claimed in claim 1, wherein the first threshold and the second
10 threshold corresponds to tenure of the first type of contract in pre-lock in stage.
7. The method as claimed in claim 1, wherein the third threshold corresponds to
tenure of the first type of contract in post-lock in stage.
15 8. An apparatus for predicting early surrender of different types of contract by a user,
the apparatus comprising:
a memory;
a processor coupled to the memory and configured to perform the
following operations:
20 obtaining a first set of parameters corresponding to the user and a first type
of contract;
determining whether the tenure of the first type of contract is more than a
first threshold and less than a second threshold;
inputting the first set of parameters to a first machine learning model when
25 the tenure of the contract is more than a first threshold and less than a second
threshold to generate a first output;
predicting a probability of surrender of the first type of contract by the first
machine learning model based on the first output;
obtaining a second set of parameters related to the user and the first type
30 of contract;
25
determining whether the tenure of the contract is more than a third
threshold;
inputting the second set of parameters to a second machine learning model
when the tenure of the contract is more than the third threshold generate a second
5 output;
predicting a probability of surrender of the first type of contract by the
second machine learning model based on the second output;
obtaining a third set of parameters related to the user and a second type of
contract;
10 determining whether the tenure of the second type of contract is more than
a fourth threshold;
inputting the second set of parameters to a third machine learning model
when the tenure of the second type of contract is more than the fourth threshold
generate a third output;
15 predicting a probability of surrender of the second type of contract by the
third machine learning model based on the third output.
9. The apparatus as claimed in claim 8, wherein after predicting the probability of
surrender of the first type of contract by the first machine learning model, the
20 processor is configured to perform the following operation:
classifying the user as a high risk user if it is determined that the probability
of surrender of the first type of contract by the first machine learning model is
above a pre-determined threshold; and
sending a prioritized communication to the user when the user is classified
25 as a high risk user.
26
10. The apparatus as claimed in claim 8, wherein after predicting the probability of
surrender of the first type of contract by the second machine learning model, the
processor is configured to perform the following operation:
classifying the user as a high risk user if it is determined that the probability
of surrender of the first 5 type of contract by the second machine learning
model is above a pre-determined threshold; and
sending a prioritized communication to the user when the user is classified
as a high risk user.
10 11. The apparatus as claimed in claim 8, wherein after predicting the probability of
surrender of the second type of contract by the third machine learning model, the
processor is configured to perform the following operation:
classifying the user as a high risk user if it is determined that the probability
of surrender of the first type of contract by the third machine learning
15 model is above a pre-determined threshold; and
sending a prioritized communication to the user when the user is classified as a
high risk user.
12. The apparatus as claimed in claim 8, wherein the first type of contract is unit linked
20 insurance policy.
13. The apparatus as claimed in claim 8, wherein the first threshold and the second
threshold corresponds to tenure of the first type of contract in pre-lock in stage.
25 14. The apparatus as claimed in claim 8, wherein the third threshold corresponds to
tenure of the first type of contract in post-lock in stage.

Dated this 31st day of August 2023