The American Journal of Engineering and Technology
25
https://www.theamericanjournals.com/index.php/tajet
TYPE
Original Research
PAGE NO.
25-32
10.37547/tajet/Volume06Issue01-06
OPEN ACCESS
SUBMITED
19 December 2023
ACCEPTED
04 January 2024
PUBLISHED
30 January 2024
VOLUME
Vol.06 Issue 01 2024
CITATION
Rachit Jain. (2024). AI-Driven Personalization in Usage-Based Insurance: A
Game-Theoretic Roadmap to Smarter Risk Assessment. The American
Journal of Engineering and Technology, 6(01), 25
–
32.
https://doi.org/10.37547/tajet/Volume06Issue01-06
COPYRIGHT
© 2024 Original content from this work may be used under the terms
of the creative commons attributes 4.0 License.
AI-Driven Personalization
in Usage-Based Insurance:
A Game-Theoretic
Roadmap to Smarter Risk
Assessment
Independent Researcher, Downingtown, PA 19335, USA
Abstract:
Usage-Based Insurance (UBI) is revolutionizing
how insurers calculate premiums based on observed
driving habits, with telematics and connected vehicles
providing growing potential for more responsive and
fairer insurance. The traditional way of calculating the
premium is based on the static models that curate the
premium for an individual based on the past driving
history, and neglecting the driving habits. This old
method has both advantages and disadvantages, but it
doesn’t provide a premium based on the
risks of the
drivers’ driving habits. Insureds were asked to pay the
premium based on the algorithm, which focuses on the
static rating tables rather than using the real-time user
driving habits data. However, these system creates
complex interactions between the insurer and insured,
specifically for privacy, data manipulation, and self-
interested driving behavior. This article highlights the
role of artificial intelligence (AI) in enhancing Universal
Basic Income (UBI) by analyzing data, refining risk
modeling, and enabling dynamic pricing in real-time.
Additionally, we model these interactions using dynamic
game theory under incomplete information. For this, we
define an insurer as a leader who sets pricing schemes
and monitors strategies, and an insured as the follower
who reacts to the incentives and possibly changes
behavior. We propose a ready-for-action AI platform
with individualized driver feedback, fraud detection, and
dynamic pricing mechanisms, and derive equilibrium
strategies for both insured and insurer, and propose a
robust pricing method for strategic manipulation. The
simulation-based synthetic driving data highlights how
game-theoretic pricing can perform better than
The American Journal of Engineering and Technology
26
https://www.theamericanjournals.com/index.php/tajet
traditional pricing methods in all aspects. The study also
elaborates on key regulatory and moral implications and
charts the way forward with future evolution and
research gaps in this new area of driving, where
technology drives the future.
Keywords:
Usage-Based Insurance, UBI, Artificial
Intelligence, Machine Learning, Risk Assessment,
Telematics, Personalized Premiums
1.
Introduction
As the number of road traffic fatalities is increasing in
the United States, this research is vital. This number
keeps increasing every year, and this is something that
can be addressed by using technology. The motor
insurance industry is moving from traditional static
pricing models to dynamic, behavior-based pricing
models. Usage-Based Insurance (UBI), or Pay-As-You-
Drive (PAYD) or Pay-How-You-Drive (PHYD), uses
telematics data to monitor driving behavior and adjust
premiums accordingly. The combination of AI and
telematics provides an opportunity to make insurance
more personalized, equitable, and responsive. Insureds
try to attempt the game the system by driving in a
different way when they are monitored or tampering
with the data-capturing devices. For this reason, it is
important to understand how these behaviors can be
reduced. This paper analyzes how AI technologies can be
effectively used in UBI and what future developments
are required to implement such integration more
effectively and effectively [1][4]. Additionally, how to
leverage dynamic game theory to capture the
interactions that can help with the right pricing and
monitoring strategies of the insurers.
2
Literature Review
Over the past few years, Usage-Based Insurance (UBI)
has undergone a significant transformation, largely
driven by the rise of connected technologies and the
adoption of big data across industries. In the insurance
sector, UBI initially took shape through the Pay-As-You-
Drive (PAYD) model, which focused on how much a
person drove but didn’t fully capture how they drove. As
the need for more accurate risk assessment grew, the
industry shifted toward the Pay-How-You-Drive (PHYD)
model. This approach analyzed post-trip driving
behavior, offering a better picture of driver risk, though
it still operated reactively. To improve real-time
engagement, the Manage-How-You-Drive (MHYD)
model emerged, allowing insurers to provide immediate
feedback and alerts during a trip. This evolution from
PAYD to MHYD reflects a broader shift toward more
personalized and preventative insurance models. As
more insurers adopt PHYD and MHYD programs, there's
a noticeable increase in the volume and complexity of
data shared between drivers and insurers, making big
data capabilities essential. Recent research highlights
the need to go beyond driving patterns by factoring in
behavioral and emotional cues, especially in cases of
aggressive driving or road rage. These insights are now
shaping how insurers evaluate risk and tailor premiums
on a more individualized basis[9].
As the insurance industry continues to evolve, many
firms are increasingly turning to InsurTech solutions to
stay competitive, with Usage-Based Insurance (UBI)
standing out as one of the most influential trends,
particularly in the auto sector. UBI enables insurers to
integrate real-world driving behavior into actuarial
models, moving beyond traditional static risk
assessments. Recent empirical studies have shown that
UBI adoption can lead to significant improvements in
underwriting performance, particularly for private
passenger auto liability (PPAL) insurers. Interestingly,
these benefits appear most prominently among
companies that were early adopters of UBI technology.
This early-mover advantage not only translates into
lower loss ratios but also drives a measurable increase
in market share, largely by attracting safer, low-risk
drivers who may be underserved by traditional pricing
models. However, the research also emphasizes that the
advantages of UBI are not immediate
—
it takes time for
these systems to mature and deliver consistent returns.
For early adopters, UBI implementation has been linked
to modest but meaningful increases in both return on
assets (ROA) and return on equity (ROE), reinforcing the
strategic value of timely technological investment.
These findings underscore UBI’s growing importance,
not just as a pricing tool but as a long-term asset for
insurers seeking both financial performance and market
differentiation[10].
While the technological and financial benefits of UBI are
becoming increasingly clear for insurers, customer
perception remains a critical factor in its broader
adoption. Recent research focusing on consumer
attitudes toward UBI reveals a complex landscape of
acceptance and resistance. The study highlights that
while many drivers are open to the concept of usage-
based pricing, their willingness to adopt it depends
heavily on demographic variables such as age, gender,
and geographic location. Additionally, the frequency of
vehicle use, prior premium amounts, and individuals’
self-perception of their driving abilities also influence
their openness to UBI. Interestingly, although customers
express a general readiness to explore new pricing
models, they still display a strong attachment to
traditional practices, particularly the no-claims bonus
The American Journal of Engineering and Technology
27
https://www.theamericanjournals.com/index.php/tajet
system, which they perceive as a familiar and reliable
reward structure. This indicates that while the insurance
market may be technologically ready for UBI, consumer
readiness requires more targeted education, trust-
building, and potentially hybrid models that balance
innovation with familiar incentives[11].
As UBI models continue to evolve, the integration of
smartphone-based sensing has emerged as a practical
and scalable approach for real-time data collection and
user engagement. A recent framework demonstrates
how smartphones can serve as dual-purpose tools,
simultaneously supporting road traffic monitoring and
UBI applications. This modular system spans from low-
level sensor functionality and data transmission to the
high-level business model, emphasizing the need to
align technical design with user incentives. The study
highlights that in addition to providing traffic-related
insights beneficial to public infrastructure and
environmental planning, the same data streams can be
used to generate individualized driving profiles for UBI
programs. Importantly, the sustainability of such a
system hinges on offering tangible benefits to users,
such as reduced insurance premiums based on good
driving behavior. Results from a ten-month pilot
campaign, involving over 250,000 kilometers of data,
underscore the feasibility of this approach and its
alignment with successful real-world deployments like
the Berkeley Mobile Millennium Project. This dual-value
proposition
—
societal benefit coupled with personalized
insurance incentives
—
positions smartphone-based UBI
as a promising frontier in both transportation and
insurance innovation[12].
Furthermore, insurers use historical data for pricing
calculations, but game theory has emerged as a tool to
study moral hazards and adverse selection in insurance
contracts. Using game theory, Insurers can make sound
decisions while maximizing the insuran
ce company’s
payoffs[14].
Based on the literature review, it is clear that a lot of
research has been conducted on UBI, but the
development of oncology and an interoperability
framework, Explanation of AI frameworks, and driver
coaching through behavior AI is the gap that is
highlighted in this literature and makes this literature
novel. The below graph shows the road traffic fatalities
in the United States from 2012 to 2024, which makes
this literature even more vital to explore the ways to
reduce fatalities using innovation and technology[13].
Fig. 1. Number of road traffic fatalities in the United States from 2012 to 2024
3
Game Theory Framework
We model the insurer-insured interaction as a
The American Journal of Engineering and Technology
28
https://www.theamericanjournals.com/index.php/tajet
Stackelberg
game
with
incomplete
information[15].
Players:
Insurer (Leader) and Insured (Follower)
Strategy Space:
Insurer: Chooses a pricing function P(d,m)
where d is the driving score and m is the
monitoring level. also chooses the detection
strategy δ.
Policyholder: Chooses driving behavior b
∈
B
and manipulation level
μ
∈
[0,1].
Payoff Functions:
Insurer:
ΠI = P(d, m)
- E[C(b)] - Cmon(m)
Policyholder: UP = -P(d, m) + Ucomfort(b) - R(b) -
Cmanip(μ)
•
Information Structure:
o
Insurer cannot observe b or
μ
perfectly;
uses signals from telematics with
uncertainty.
•
Game Type:
o
Dynamic Bayesian Stackelberg game
with asymmetric and incomplete
information.
4
Equilibrium Analysis
We solve for subgame-perfect Bayesian equilibria[16].
Under certain assumptions (e.g., linear utility and cost
functions), best-response functions can be derived
analytically:
•
Policyholders exert minimal effort when
monitoring is low or the manipulation cost is
low.
•
Insurer's optimal pricing balances expected
claims with the cost of monitoring and privacy
backlash.
We characterize equilibria where:
•
Honest
driving
dominates
under
high
monitoring and penalties.
•
Strategic manipulation emerges under lax
monitoring and high comfort rewards.
We define the best-response function of the
policyholder as:
b*(m,P) = arg maxb UP (b, μ) subject to the constraints
of telematics feedback.
5
UBI Landscape and Challenges
UBI systems traditionally apply pre-set rules to calculate
each risk score based on mileage, speed, braking habits,
and time of day. Major insurance firms like Progressive,
Allstate, and Metromile have introduced UBI products
based on mobile apps or on-board devices. These kinds
of systems have limitations in flexibility, data processing
scalability, data manipulation, and behavioral bias.
Moreover, standardization of data formats and
interpretability of AI-based decisions continue to be a
problem. Explainable AI models that not only predict but
also interpret pricing and risk logic for users and
regulators are in growing demand [2][7].
6
Artificial Intelligence Integration in UBI
6.1
Data Collection and Preprocessing
Data collection and preparation that runs on data
sources are OBD-II devices, smartphones, and
connected car systems. Information includes engine
measurements,
GPS
coordinates,
accelerometer
readings, and surroundings. AI models clean up and
massage the data to handle outliers, missing values, and
high-frequency noise. Data from many sensor sources is
also integrated using data fusion techniques to improve
contextual understanding [4].
6.2
Risk modeling
Supervised machine learning algorithms such as random
forests, Gradient boosting machines, and deep neural
networks, used to identify risky driving behavior. When
these models are trained on the labeled dataset, the
ground truth is the insurance claims. Advanced models
take into account relevant elements such as traffic,
weather, and driver demographics [4].
6.3
Premium Calculation
AI enables dynamic premium pricing by continuously
updating the risk profile of drivers. Models consider
both historical and recent driving behaviors to suggest
profile-based premiums. Reinforcement learning may
also be used to optimize pricing strategies over time
based on observed driver response and claim outcomes
[2]. This helps the insurer in getting the correct
insurance premium based on their risk factor.
6.4
Fraud Detection
Unsupervised learning algorithms like Isolation Forests
and Autoencoders help identify anomalies in telematics
data, e.g., GPS spoofed data or accelerometer signal
manipulation, that are indicative of fraud. Graph-based
The American Journal of Engineering and Technology
29
https://www.theamericanjournals.com/index.php/tajet
anomaly detection techniques can be used to identify
network-level patterns of coordinated fraud [4][8].
6.5
Driver Feedback and Engagement
Reinforcement learning and behavioral analytics can be
used to offer real-time feedback to drivers, encouraging
safer driving through gamification and incentives. AI can
personalize coaching strategies based on individual
driving profiles and behavioral patterns [2]. This helps
the insurer in the right driving pattern adoption.
Fig. 2. Artificial Intelligence integration in UBI
6.6
Robust Pricing Strategy
To
handle
the
data
manipulation,
we
implemented an incentive-compatible pricing contract
•
Contracts include thresholds for safe behavior
and rebate structures.
•
Mixed-strategy equilibria show robustness to
manipulation spikes.
•
Information-theoretic bounds (e.g., KL-
divergence) are used to detect abnormal data
patterns.
The insurer solves the following constrained
optimization problem:
maxP,m E[
ΠI
] subject to
∀
μ, UP(b*, μ) ≤ UP(b*, 0)
,
where b* is the insurer's desired behavior.
7
Architecture and Implementation Framework
A robust AI-enabled UBI system architecture involves
the following layers:
•
Edge Layer
: Data acquisition through telematics
devices installed [6].
•
Streaming Layer
: Real-time ingestion using Kafka or
AWS Kinesis.
•
Analytics Layer
: Feature engineering and model
inference using Spark ML or TensorFlow.
•
Storage Layer
: Scalable storage on cloud platforms
like AWS S3 or Azure Blob.
•
Application Layer
: Interfaces for customers, insurers,
and regulators.
The American Journal of Engineering and Technology
30
https://www.theamericanjournals.com/index.php/tajet
Fig. 3. UBI System Architecture
These are the five key pillars on a high level for an AI-
enabled UBI system.
8
Simulation & Results
We simulate a synthetic dataset of 10000 policyholders
with unbiased and completely random driving profiles
with data on speed, acceleration, and idle time, etc.
Driving behavior is scored using a logistic risk model:
d = 1 / (1 + e^{-
β^T x}),
where x is the feature vector
and
β
are model weights.
Another observation was noted, which was that
Manipulation reduces observable risk scores but
increases variance.
We compare four pricing models-
1.
Flat Pricing
2.
Linear Mileage-Based
3.
Usage-based Pricing
4.
Game-Theoretic Adaptive Pricing
Below were the key numbers we have observed-
•
Game-theoretic model reduced the loss ratio by
15%.
•
Detection rate of manipulative behavior
increased from 52% to 88%.
•
Policyholder satisfaction remained within ±5%
of baseline.
9
Ethical, Regulatory, and Business Implications
9.1
Openness and equality
Although AI improves accuracy, the model requires
openness to justice. Discriminated prices may be caused
by prejudice in training. Techniques like LIME and SHAP
values can be used to explain the model to both insurers
and insureds.[3][7] This way, they understand how the
system works, and they start having faith in the system.
9.2
Data Privacy and Security
Constant monitoring is of significant concern when it
comes to privacy. GDPR and CCPA laws must be catered
to in system design [5]. Differential privacy and
federated learning methods can be employed to train
models without direct access to raw user data. Laws
must be enforced to correct the usage of the data.
9.3
Business Opportunities
AI-enhanced UBI models will be capable of opening up
new customer segments, improving retention, and
reducing
claims
costs.
Insurers
must
weigh
personalization against compliance. Context awareness-
based micro-insurance products and policy bundling
The American Journal of Engineering and Technology
31
https://www.theamericanjournals.com/index.php/tajet
may emerge as new business opportunities.
10
Conclusion and future work
AI reinforces the next generation of the UBI system with
dynamic, fair, and personal insurance products based on
profile and usage. Despite this, there are significant gaps
and opportunities for future research. Below are the few
gaps that we identified in our research.
10.1
Standardization of telematics data
With the proliferation of vehicle sensors and IoT units, a
standard data format for UBI is required. Future
research must focus on the development of oncology
and an interoperability framework [1] [4].
10.2
Explains AI in insurance prices
While current models are predictive and perfect,
regulatory bodies now require interpretation. Research
is required in insurance-focused areas that explain AI
frameworks [7]. This helps to build trust and
transparency in the system.
10.3
UBI model side effects
AI models for UBI are unsafe for unfavorable
manipulation. Future work should emphasize the
strengthening of the model against GPS falsification,
signal driving, and telematics data poisoning [8].
Otherwise, this could lead to distrust in the system.
10.4
Driver coaching using behavior AI
Drivers have the opportunity to develop customized AI-
driven coaching platforms for real-time behavior. Such
platforms can reduce accidents and claims, making
customers satisfied. This will help the insured improve
their driving habits.
10.5
AI Audit for Ethics
An emerging field of study entails designing a framework
for regular revision of AI platforms in UBI to ensure
moral farming and eliminate prejudice [3]. Also, this will
uncover any loopholes in the system.
10.6
UBI in autonomous and divided dynamics
With an increase in autonomous cars and shared
mobility services, UBI products must mature. AI can
figure out shared use context risks and prepare
guidelines for autonomous driving behavior [6].
10.7
Edge AI for onboard risk assessment
Future systems can use Edge AI to reduce the delay in
risk assessment. Real-time inference on in-vehicle
hardware can cause rapid reaction and an increase in
personalization [6].
UBI will continue to develop with progress in AI,
telematics, and the smart mobility ecosystem. A multi-
related approach in combination with computer science,
behavioral economics, legal compliance, and human-
focused design is necessary to realize the full potential
of AI-Powered Insurance. Over time, these stacks will be
optimized and provide the correct data for better
decision-making.
REFERENCES
1.
Tselentis, Dimitrios & Yannis, George & Vlahogianni,
Eleni. (2016). Innovative Insurance Schemes: Pay
as/how You Drive. Transportation Research
Procedia. 14. 362-371. 10.1016/j.trpro.2016.05.088.
2.
Erevelles, Sunil & Fukawa, Nobuyuki & Swayne,
Linda, 2016. "Big Data consumer analytics and the
transformation of marketing," Journal of Business
Research, Elsevier, vol. 69(2), pages 897-904.
3.
M. Hardt, E. Price, and N. Srebro, "Equality of
opportunity in supervised learning," Advances in
Neural Information Processing Systems, pp. 3315
–
3323, 2016.
4.
Oladimeji, D., Gupta, K., Kose, N. A., Gundogan, K.,
Ge, L., & Liang, F. (2023). Smart Transportation: An
Overview
of
Technologies
and
Applications. Sensors, 23(8),
3880.
https://doi.org/10.3390/s23083880
5.
Protection, F. D. (2018). General data protection
regulation (GDPR).
Intersoft Consulting, Accessed in
October
,
24
(1).
6.
Cao, Keyan & Liu, Yefan & Meng, Gongjie & Sun,
Qimeng. (2020). An Overview on Edge Computing
Research.
IEEE
Access.
PP.
1-1.
10.1109/ACCESS.2020.2991734.
7.
Samek, W., Montavon, G., Vedaldi, A., Hansen, L. K.,
& Müller, K.-R. (Eds.). (2019). Explainable AI:
Interpreting, explaining and visualizing deep
learning. Springer. https://doi.org/10.1007/978-3-
030-28954-6
8.
Siva Kumar, R. S., Nyström, M., Lambert, J. M.,
Marshall, A., Goertzel, M., Comissoneru, A., Swann,
M., & Xia, S. (2020, February 4).
Adversarial machine
learning – industry perspectives
(arXiv:2002.05646)
[Preprint].
arXiv.
https://doi.org/10.48550/arXiv.2002.05646
9.
Arumugam, S., & Bhargavi, R. (2019). A survey on
driving behavior analysis in usage based insurance
using big data.
Journal of Big Data
,
6
(1), 1-21.
10.
Che, X., Liebenberg, A., & Xu, J. (2022). Usage-based
insurance
—
Impact on insurers and potential
The American Journal of Engineering and Technology
32
https://www.theamericanjournals.com/index.php/tajet
implications
for
InsurTech.
North
American
Actuarial Journal
,
26
(3), 428-455.
11.
Śliwiński, A., & Kuryłowicz, Ł. (2021). Usage‐based
insurance and its acceptance: An empirical
approach.
Risk
Management
and
Insurance
Review
,
24
(1), 71-91.
12.
Händel, P., Ohlsson, J., Ohlsson, M., Skog, I., &
Nygren, E. (2013). Smartphone-based measurement
systems for road vehicle traffic monitoring and
usage-based insurance.
IEEE systems journal
,
8
(4),
1238-1248.
13.
Carlier, M. (2023, December 18). _ Road accidents in
the
United
States _ [Topic
page].
Statista.
https://www.statista.com/topics/3708/road-
accidents-in-the-us/
14.
Norozpour, Sajedeh & Safaei, Mehdi. (2020). An
Overview on Game Theory and Its Application. IOP
Conference
Series:
Materials
Science
and
Engineering.
993.
012114.
10.1088/1757-
899X/993/1/012114.
15.
McNabb, David & Lee, Chung-Shing. (2020). The
Game
Theory
Framework.
10.4324/9781003094654-10.
16.
Li, Xiaoqin. (2024). Equilibrium analysis in game
theory
using
nonlinear
functional
analysis
techniques. Journal of Electrical Systems. 20. 1099-
1110. 10.52783/jes.3437.
