Applying

Machine Learning
to Actuarial and
Pricing Workflows

Applying Machine Learning to Actuarial

and Pricing Workflows
Introduction techniques like penalized regression behave
consistently with standard credibility models.
Applying machine learning to pricing processes This means that machine learning techniques can
is becoming easier due to advances in technolo- limit the noise from thin data and can be used on
gy and increasingly accessible platforms. While datasets that were previously out of scope due to
complex algorithms seem to be at odds with the their small size. Second, the intended purpose of a
transparency and judgment needed in actuarial machine learning analysis can be expanded from
and pricing practices, these concepts can be highly creating final implementation recommendations to
complementary if machine learning is implemented providing teams with the best data-driven point to
correctly. In this paper, we will explore the necessity start additional analysis. Making this adjustment
and possibilities of machine learning applications allows machine learning techniques to be applied
in the insurance industry, and provide guidelines to everyday work and provides teams with the best,
for incorporating machine learning techniques into most data-driven starting point for analysis. Finally,
pricing applications. the increased ease of applying these credibility-
based machine learning techniques will allow
Limited Data and the Intended Purpose pricing teams to adopt these new tools without
of Machine Learning significant investment. The use of machine learning
algorithms can begin when a team is working with
One common pushback to employing modeling or small datasets and will evolve as the team and
machine learning techniques is the lack of sufficient company grow.
data. This viewpoint is often based on antiquated
assumptions about machine learning algorithms For small datasets, machine learning algorithms
and the intended purpose of their output. First, may be used to identify the starting point for
a traditional method of addressing thin data is to continued pricing analysis. Machine learning can
apply credibility considerations to it in order to avoid determine the characteristics, directionality, and
overreacting to noise in the data. Machine learning approximate magnitude of the changes required
 APPLYING MACHINE LEARNING TO ACTUARIAL AND PRICING WORKFLOWS

to correct for poor experience in a rating plan. assumptions ranging from most to least restrictive
The algorithm will provide a shortcut by quickly from a business perspective. The result will be a
identifying potentially hidden problem areas without chart similar to the one below.
hours of dashboard analysis.

As lines of business grow, analysis of medium-

sized datasets can also evolve. By using the
existing rating plan as a starting point, machine
learning techniques can be applied effectively on
datasets that would previously be out of scope.
While the data may still not be credible enough to
build a rating plan from scratch, it is likely sufficient
to fully support the direction and magnitude of the
proposed changes.

Finally, with large datasets, machine learning will

fully support the creation of a new rating plan and Business decisions are rarely black and white.
can be used for advanced analysis. This is typically Usually, there is a range of reasonable actions that
the step that people jump to when considering the can be implemented to address a problem, and
application of machine learning techniques, but it is the “best” solution is a matter of opinion. Similarly,
not the starting point! the efficient frontier approach provides teams with
the opportunity to identify a “range of reasonable
Machine learning has too much potential for pricing models.” Within this range, actuarial and business
departments not to take advantage of it in every judgment can be applied to determine the best
applicable analysis. On increasingly small datasets, course of action. A point estimate does not allow
analysts will benefit by starting with a purely data- for this judgment to be easily applied.
driven machine learning model that immediately
identifies signals coming from the data, and informs
future analysis.
The Changing Actuarial Experience
Moving from a manual process to a machine
Using Machine Learning for an Efficient learning modeling process will change the working
Frontier experience for an actuary. Often, this results in a
perceived - but not actual - loss of control due to
As an extension of the exploration of machine automation. The separation of purely data-driven
learning models, we should reconsider our decisions and business decisions is key when
viewpoint on a model’s output as well. It is assumed adopting a machine learning approach.
that the output of a model is always a point
estimate - but this is not necessarily true! Machine For example, the tedious process of deciding where
learning techniques can address different levels to cap a characteristic such as policyholder age
of constraints and achieve different outcomes. based on limited credibility is often a judgmental
When business decisions need to be applied to decision. However, the optimal placement of this
a model, it may be best for the machine learning cap is a purely data-driven decision that should
technique to produce an efficient frontier of outputs be automated to remove human bias from the
based on various assumptions. To do this, teams process. After seeing this data-driven result, a user
can take advantage of the automation provided by can change this data-driven selection if necessary
machine learning and run a series of models with for business decisions. This is not a lack of control
 APPLYING MACHINE LEARNING TO ACTUARIAL AND PRICING WORKFLOWS

- but rather a more informed starting point to may make explanations to internal and external
exercise control. stakeholders difficult, and their questions on the
calculation of an insured’s score may not have an
The majority of the day-to-day actuarial experience intuitive or helpful response. Transparent machine
will shift from manual creation of data analysis to learning algorithms allow an actuary to easily
interpreting the output of automated data analysis. interpret the actuarial soundness of a model and
This is not to say that actuarial teams will require greatly speeds up the internal and external approval
less technical knowledge - to fully understand the process.
assumptions of these machine learning algorithms,
an actuary will still need sufficient technical skills.
The removal of this busywork will simply allow an Equity, Fairness, Bias, and Transparent
actuarial team to use their technical skills in more Algorithms
meaningful and interesting ways.
Machine learning cannot be discussed without
also acknowledging the topic of algorithmic bias.
The Necessity of Transparent How to consistently avoid disparate impact while
Algorithms for Actuarial Analysis setting equitable and competitive rates is a current
discussion and research topic in the US insurance
This paper has discussed the necessity of and regulatory environment, and this paper will
interpreting automated decisions and creating an not propose a specific methodology to address it.
efficient frontier of actions to evaluate, but so far Transparent machine learning technology will also
has neglected to mention their prerequisite, which not miraculously solve this problem, but it will allow
is transparent machine learning. When selecting rates and models to be comprehensively analyzed
between similar models, a user needs the ability to and their reasoning fully understood. For each
determine which model is the most appropriate for rating variable, the exact impact is understood and
implementation given business considerations. easily measured. “Why did this happen?” is an easy
question to answer with transparent models.
This differentiation is quite difficult with black box
models, as it may not immediately be clear what is On the other hand, black box machine learning
happening within the black box to cause the models techniques make the already difficult problem of
to produce different scores for similar risks. When equity and bias even harder to solve by obfuscating
using black box models, an actuary must also the calculation and justification for rates. There
take into consideration the possibility for hidden are increasingly sophisticated ways to look at the
behaviors that are unintuitive at best and dangerous exports of a black box, but in the end, a modeler
at worst. Within a black box model, it is nearly needs to trust that they’ve explored everything.
impossible to prove that unintuitive behavior is What happens when an unanticipated edge case
not happening. An actuary would need to perform appears in practice? How can we be sure that the
sufficient testing to prove that the likelihood or black box is unbiased in both the aggregate and at
effect of unintuitive behavior is immaterial or at an all individual levels? With transparent models like
acceptable level of risk. This is a time-consuming penalized GAMs, questions on model behavior are
and necessarily imperfect process that can be easy to answer. With a GBM on the other hand, the
avoided with transparent algorithms. explanation of a rate calculation is opaque.

Due to the less interpretable nature of black box While it is uncertain how the industry will move
methodologies, the likelihood of identifying an forward to address issues of equity and disparate
unintuitive behavior late in the work cycle will impact, transparent machine learning clearly has
increase. Additionally, black box methodologies
 APPLYING MACHINE LEARNING TO ACTUARIAL AND PRICING WORKFLOWS

an advantage over black box methodologies in the

ability to be explained and validated. Transparent ABOUT AKUR8
machine learning algorithms will still be as biased
Akur8 is revolutionizing insurance pricing with Transparent
as the data going into them - but this bias is not
Machine Learning, boosting insurers’ pricing capabilities with
hidden by black box methodology. The effect
unprecedented speed and accuracy across the pricing process
of all variables will be visible to the actuary and without compromising on auditability or control.
regulator and can therefore be properly addressed.
Transparent machine learning algorithms are Our modular pricing platform automates technical and
necessary for insurance companies and regulatory commercial premium modeling. It empowers insurers to
bodies that want to address possible biases in compute adjusted and accurate rates in line with their
business strategy while materially impacting their business
insurance models.
and maintaining absolute control of the models created, as
required by state regulators. With Akur8, time spent modeling
Conclusion is reduced by 10x, the models’ predictive power is increased by
10% and loss ratio improvement potential is boosted by 2-4%.
Machine learning is quickly entering the insurance
space, and actuaries can benefit greatly from its Akur8 already serves 50+ customers across 20+ countries,
including AXA, Generali, Munich Re, Tokio Marine North
adoption in pricing processes. Expanding the use
America Services (TMNAS); specialty insurer Canopius and
and purpose of machine learning analysis will open
MGA Bass Underwriters; consulting partners Xceedance and
exciting new opportunities. Adopting an easy-to- Perr & Knight; and insurtechs Manypets and wefox. Over 700
use platform for machine learning will be key in actuaries use Akur8 daily to build their pricing models across
allowing actuaries to use machine learning to its all lines of business. Akur8’s strategic partnerships include
fullest potential. An analyst will not frequently use Milliman, Guidewire, Duck Creek and Sapiens.
machine learning techniques if they need to set up
a codebase from scratch for every application and
maintain the codebase across multiple platform
updates. Machine learning algorithms should be as
accessible to actuaries as spreadsheets.

The proper application of machine learning will

meaningfully change a team’s work experience,
but it will not dilute the importance of industry
knowledge. Machine learning algorithms will
quickly provide data-driven insights so users can
apply their industry knowledge to the results and
drive company strategy. Machine learning is no
longer just a forward-looking buzzword - it is now
ready to be broadly applied to pricing analysis.

www.akur8.com /company/akur8 contact@akur8.com