The two-minute guide to understanding and selecting the right Descriptive,

Predictive, and Prescriptive Analytics

With the flood of data available to businesses regarding their supply chain these
days, companies are turning to analytics solutions to extract meaning from the huge
volumes of data to help improve decision making

Companies that are attempting to optimize their S&OP efforts need capabilities to
analyze historical data, and forecast what might happen in the future. The promise of
doing it right and becoming a data-driven organization is great. Huge ROIs can be
enjoyed as evidenced by companies that have optimized their supply chain, lowered
operating costs, increased revenues, or improved their customer service and product
mix.

Looking at all the analytic options can be a daunting task. However, luckily these
analytic options can be categorized at a high level into three distinct types. No one
type of analytic is better than another, and in fact they co-exist with, and
complement, each other. In order for a business to have a holistic view of the market
and how a company competes efficiently within that market requires a robust analytic
environment which includes:

 Descriptive Analytics, which use data aggregation and data mining to

provide insight into the past and answer: “What has happened?”
 Predictive Analytics, which use statistical models and forecasting
techniques to understand the future and answer: “What could happen?”
 Prescriptive Analytics, which use optimization and simulation algorithms to
advise on possible outcomes and answer: “What should we do?”

Descriptive Analytics: Insight into the past

Descriptive analysis or statistics does exactly what the name implies: they
“describe”, or summarize, raw data and make it something that is interpretable by
humans. They are analytics that describe the past. The past refers to any point of
time that an event has occurred, whether it is one minute ago, or one year ago.
Descriptive analytics are useful because they allow us to learn from past behaviors,
and understand how they might influence future outcomes.

The vast majority of the statistics we use fall into this category. (Think basic
arithmetic like sums, averages, percent changes.) Usually, the underlying data is a
count, or aggregate of a filtered column of data to which basic math is applied. For
all practical purposes, there are an infinite number of these statistics. Descriptive
statistics are useful to show things like total stock in inventory, average dollars spent
per customer and year-over-year change in sales. Common examples of descriptive
analytics are reports that provide historical insights regarding the company’s
production, financials, operations, sales, finance, inventory and customers.

Use Descriptive Analytics when you need to understand at an aggregate level what
is going on in your company, and when you want to summarize and describe
different aspects of your business.
Predictive Analytics: Understanding the future

Predictive analytics has its roots in the ability to “predict” what might happen. These
analytics are about understanding the future. Predictive analytics provides
companies with actionable insights based on data. Predictive analytics provides
estimates about the likelihood of a future outcome. It is important to remember that
no statistical algorithm can “predict” the future with 100% certainty. Companies use
these statistics to forecast what might happen in the future. This is because the
foundation of predictive analytics is based on probabilities.

These statistics try to take the data that you have, and fill in the missing data with
best guesses. They combine historical data found in ERP, CRM, HR and POS
systems to identify patterns in the data and apply statistical models and algorithms to
capture relationships between various data sets. Companies use predictive statistics
and analytics any time they want to look into the future. Predictive analytics can be
used throughout the organization, from forecasting customer behavior and
purchasing patterns to identifying trends in sales activities. They also help forecast
demand for inputs from the supply chain, operations and inventory.

One common application most people are familiar with is the use of predictive
analytics to produce a credit score. These scores are used by financial services to
determine the probability of customers making future credit payments on time.
Typical business uses include understanding how sales might close at the end of the
year, predicting what items customers will purchase together, or forecasting
inventory levels based upon a myriad of variables.

Prescriptive Analytics: Advise on possible outcomes

The relatively new field of prescriptive analytics allows users to “prescribe” a number
of different possible actions and guide them towards a solution. In a nutshell, these
analytics are all about providing advice. Prescriptive analytics attempts to quantify
the effect of future decisions in order to advise on possible outcomes before the
decisions are actually made. At their best, prescriptive analytics predicts not only
what will happen, but also why it will happen, providing recommendations regarding
actions that will take advantage of the predictions.

These analytics go beyond descriptive and predictive analytics by recommending

one or more possible courses of action. Essentially they predict multiple futures and
allow companies to assess a number of possible outcomes based upon their actions.
Prescriptive analytics use a combination of techniques and tools such as business
rules, algorithms, machine learning and computational modelling procedures. These
techniques are applied against input from many different data sets including
historical and transactional data, real-time data feeds, and big data.

Prescriptive analytics are relatively complex to administer, and most companies are
not yet using them in their daily course of business. When implemented correctly,
they can have a large impact on how businesses make decisions, and on the
company’s bottom line. Larger companies are successfully using prescriptive
analytics to optimize production, scheduling and inventory in the supply
chain to make sure they are delivering the right products at the right time and
optimizing the customer experience.

Actuarial analysis
Life insurance companies and Annuity firms today
use assumptions to set aside reserves and to
hedge on the market. These assumptions are
typically created with actuarial models that create
cells of customers based on a few factors (usually
less than five) that segment the book of business
into equally distributed groups so that actuaries can
compute aggregate ratios on attrition, withdrawals,
and other factors. Typically, these values are computed
for the previous year and compared against historical
variations to see if the trend is changing its direction. A
more in-depth analysis can step through these changes
year to year moving forward or backward by using
values from the current year to derive dependencies
for the next or previous year.

This process creates several problems in the modern

market place. Since the number of variables are small
emerging trends or trends that are specific to specific
customer populations are hard if not impossible to
create. So, policies must be priced based on broad
characteristics and the burden is shifted to
underwriting in the life insurance industry to
cover for risks. Annuity contracts must be priced
with a significant fee since the risk of withdrawals
cannot be estimate precisely and market hedging
must be done with large pools of money reducing
the flexibility in investment.
These factors limit the ability of life insurance
companies to react to changing market conditions
quickly and to create products that can be better
suited for targeted populations of customers.

Machine learning techniques are changing the way this

process is working in leading life insurers. First,
instead of using about five variables to create
actuarial assumptions, life insurance companies are
adding more variables with a focus on variables that
capture the change in behavior of policy and contract
holders. Policy holders’ payment, withdrawal,
surrender, change in payment mechanisms, and other
variables provide fine grain understanding of who,
when, and what are driving attritions, withdrawals, and
acquisition of related products. Inputs from these
machine learning models are also being
integrated into the actuarial process to estimate
reserves to be established and Monte-Carlo and
similar simulation methods for estimating market
performance of assets. Furthermore, by
incorporating variables that were traditionally
considered to be underwriting variables into the
actuarial processes, leading edge insurers are
reducing the complexity for policy underwriting
and supporting the ability to create insurance
products that are experience focused and target
specific populations (e.g., millennials who are
adventure travelers).

Underwriting
The holy grail in life insurance underwriting has been
to pre-qualify clients similar to credit card pre-
qualification for marketing so that the marketing and
writing of life insurance policies is as smooth as
friction free as issuing a credit card. Insurers have
taken intermediate steps in this direction by installing
rule based execution systems for straight through
processing and other process optimizations. However,
these steps have not made it possible to
underwrite a customer without performing
medical tests to determine their risk profile.
Furthermore, due to the generality of life
insurance actuarial models, it has been harder to
reduce the number of questions that needed to be
answered by the customer.

To reduce the need for the medical test, insurance

companies are now experimenting with deep learning
techniques that use commercially available medical
data on radiology images, clinical trial data including
blood samples and correlation of blood samples with
diseases, text understanding of physician’s notes,
etc. taking thousands of variables and predicting
occurrence of certain diseases in the customer to be
insured and the acute or chronic nature of the disease.
While this brave new frontier needs to be carefully
managed to avoid regulatory and customer backlash,
green lighting policies using deep learning
approaches can provide a way to overcome such
negative reactions in the market place.

The key benefit provided by deep learning is the ability

to include the thousands of heterogeneous variables
in creating new predictive models that can be used to
classify applicants into life insurance pools without
performing medical tests. Consider the number of
variables that blood samples measure. Add to it
variables that radiology images and features that can
be extracted from those images that point to specific
medical conditions. Combine this with key
information about chronic conditions and
treatment plan information from doctor notes
using text understanding. Enhance this information
using life style variables about risky life styles that are
usually ascertained by questions asked but now
determined using external data sources, the number of
variables expand far beyond the ability of conventional
advanced analytics models. Deep machine learning
techniques allow insurance companies permits life
insurance companies to take these multi thousand
variable set and predict the applicant’s risk class.

Product development
One of the key new trends in product development in
life insurance and annuity is to create products for Gen
Xers, Millennials, and Gen Y. From several projects, we
have performed across our life insurance customers,
we have seen that products that appeal to the younger
generation are temporally relevant. While past
generations have bought life insurance products to
protect their family or to protect income by paying
premiums for a long time (typically, not less than 10
years), younger generations are interested in insuring
their life styles over a narrower period. For example, a
millennial may be interested in buying life
insurance as they rent a zipcar to drive around
Boston or may be interested in buying life
insurance for a few years after birth of a child.

While life insurance policies are available for specific

travel (e.g., life insurance for air travel), the idea of life
insurance products that protect for a shorter duration,
ranging from days to not more than a few years is
something that is new to the life insurance industry.
How can advanced analytics and deep learning help
create these types of products? As mentioned earlier,
the actuarial practices work with limited variables
but very long history that is meant to predict and
price mortality during normal course of life with
significant risk in payments spread over many
policy holders. However, products that address life
styles must deal with shorter time frames and events
that are numerous and historical data on their risk is
not available for long periods of time.

However, with new sources of information that are

more real time such as data about life styles
collected from fitness devices, health information
collected as described above, with deep machine
learning’s ability to predict important events that could
occur, life insurance companies can create new class of
products suited to the life styles of new class of
customers opening new avenues of revenue.

Marketing
Life insurers and annuity companies have significant
opportunities in improving their marketing. Most of the
companies today sell their product through distribution
channels (agents, wirehouses, etc.) and perform brand
marketing to establish their firm as a solid company
that will not go defunct in the next few years. However,
given the nature of these products, customers are very
infrequently in touch with the insurance companies
except when they would like to surrender their policy
or contract or when they wish to claim against the
policy. More targeted marketing to retain customers to
increase profitability have typically resulted in
increased mortality risk carried at the book level by the
firms. If effectiveness of marketing is measured in
terms of new customers recruited and customers
retained, then life insurance company marketing has
not been very effective.

Advanced analytics and machine learning can be used

to address this problem. A general industry pattern is
that up to 20% of term life customers attrit in the first
3 years not even paying back the marketing expenses
spent on recruiting them. Another 10% to 20% attrite
one or two years before term end and premium jumps
and up to 50% attrite during so called shock period one
year before and one year after the term end (e.g, 10th
year for a term-1o policy). It should be clear that
retaining customer during the early period and
selectively retaining customers who don’t have
significant mortality around the shock value period can
improve the profitability of an insurer. By creating
predictions on which policy holder will attrit when,
prospectively identifying the signals and drivers for
that attrition, and by better estimating total life time
value for those customers, advanced analytics and
machine learning techniques can help an insurer better
target their existing customers.

In addition, this type of targeted marketing combined

with machine learning refined underwriting and
actuarial practices can create “offers” that are
appropriate for the customer. Better estimation of
mortality through machine learning driven
underwriting and shorter duration life insurance offers
(extensions on current policies) can allow insurance
companies to target and retain late stage customers
whose policy maturity may be imminent. Similarly, for
annuities, a better prediction of withdrawal
probabilities can allow annuity companies to target
contract extensions that will help the customer keep
more of their investments with the firm.

Summary
Advanced analytics and machine learning techniques
can have fundamental impacts on all aspects of a life
insurance and annuity firm. Above, implications of
these approaches for most functions of an insurance
company was provided. For further questions, reach
out to [email protected]