[MUSIC] Welcome back.

Now that we have laid out the data and

analytic framework, we need to answer the question, what analytic techniques
can I use and what tools can help me. In this video, we're going to talk
about five types of data analytics, and the tools used to help build our analysis.
We categorize the types of analytics
that one can do into five types. Descriptive analytics, diagnostic
analytics, predictive analytics, prescriptive analytics, and
adaptive and autonomous analytics. These are the broad categories that
will help you deliver on the data and analytics framework we discussed
in the previous video. The framework is abstract, but these types of analytics will
help you operationalize it. As we discussed earlier, one of
the common uses of analytics in marketing is in cross-selling multiple
products to customers. When we looked at the outcome of
increasing the number of cross product holdings of our customers,
we started with the descriptive analytics. We looked at their data to profile
their customers with respect to one product holders,
two product holders, etc. These profiles included
the sociodemographics of the customer, their online and offline behaviors,
their attitudes, life stages at which they bought
these products, and so on. Descriptive analytics helps you understand the current
state of
affairs in an organization. It lets you look at what
is happening today and it's what has happened in the past. This type of analytics
typically
provides summarized information to understand currently existing
sales patterns or customer behavior, customer profitability,
past competitor actions, etc. Specific techniques might
include simple box plots, histogram charts with means,
minimums, and maximums. Plotting the data in quartiles or deciles
across a number of different variables. Or computing statistical measures like
mean, mode, standard deviation, etc. Descriptive analytics is very powerful for
understanding the current state of
affairs and for developing the hypothesis to anticipate where business problems and
opportunities may lie. It helps us answer the question,
what happened? For example,
from the descriptive analytics, it was clear that a large proportion of
the customers of the insurance company we talked about in the previous video,
had only one product, and a very small number of customers
had four products or more. In addition, there were three
products that all had more or less equal share within
the one product customers. After we had a good description of what
the cross product holding of the different customers were, we started investigating
why the cross product holding was so low. It turned out that the insurance company
had multiple channels to sell their products, captive agents, independent
agents and telemarketing staff. Each of these channels sold a specific
type of product well and this resulted in an equal number of one product
customers across all three channels. Given the expertise of these channels,
there was very little cross selling. Multiple product holdings was more
a result of customers demanding the additional products, as opposed to
any deliberate cross selling process. Diagnostic analytics helps you
understand why it happened. It provides the reasons for
what happened in the past. This type of analytics typically tries
to go deeper into a specific reason or hypotheses based on
the descriptive analytics. While descriptive analytics cast a wide
net to understand the breadth of the data, diagnostic analytics goes deep,
probing into the causes of issues. For example, we might look at creating
a decision tree analysis of the cross product holdings to reveal the types of
customers who have bought these products, the channels they use, the products
they've bought and when they bought them. Once we knew what some of the major
issues
in a low cost product holding were, we started formulating
hypothesis on what we can do to increase cross-sale of products. We built
predictive model to rank
customers on their propensity to buy a specific second or
third or fourth product. This predictive model was built
based on the understanding we gained from the two previous steps of
descriptive and diagnostic analytics. Next we have predictive analytics. Unlike
descriptive or diagnostic analytics, predictive
analytics is more forward looking. Predictive analytics lets you envision
what could happen in the future. This type of analytics can help
the client answer questions like, what are my customers
likely to do in the future? What are my competitors likely to do? What will the
market look like? How will the future impact my product or
service? Predictive analytics typically
predicts what could happen based on the evidence we have seen. In our insurance
case study,
once we've built our propensity model, we were able to identify some of
the high potential targets for cross sale, and
what product they should be cross sold. Given the different propensities to buy,
we computed the next best offer for each and every customer and how the cross
sell message should be personalized for each customer and distribution channel.
Prescriptive analytics goes beyond
providing recommendations to actually executing the actions or
taking the decisions that are right for a particular situation. It does this by
looking at
what happened in the past, the present state and
all the future possibilities. Prescriptive analytics provides
answers to the question, what steps or interventions need to be taken
to achieve the desired outcomes? Often the intervention might be an optimal
solution given the circumstances. Or the best possible action given
the uncertainty in the environment and the limited information available. It
frequently involves scenario analysis
and or searching for optimal solutions. Prescriptive analytics is powerful
in understanding the right actions needed today to address future
possibilities and put an organization the best possible position to take
advantage of future conditions. While we built a one off solution for the client to
increase the cross-product
holding of their customers, what they really needed was an adaptive
and continuous system that learns from the behavioral interventions and
actions taken by customers to automatically change the
recommendations and try out new measures. Such an always-on insights
platform where the system builds a model of the real world, takes actions,
learns from the environment, and continuously adapts itself is the ultimate
adaptive, autonomous solution. Adaptive and autonomous analytics
is still in it's infancy, most systems today are either
predictive or prescriptive. Very few of them are completely adaptive,
or autonomous. However, there are a number of companies
that are building more adaptive, or autonomous analytic solutions where we
are eliminating the human in the loop. Autonomous car driving is a great
example of an adaptive or autonomous analytic solution. Adaptive and autonomous
analytics
provides answers to the question how does the system adapt to changes? How can we
run analytic
solutions on a continuous mode? Constantly learning and correcting its
behavior to optimize its performance. We may not want to build adaptive and
autonomous systems in all cases. There may be instances where we may
want to retain the human decision maker. But there may be other situations
where the speed of decision making, is such that having a human in
the loop may be counter productive. Algorithmic trading might
be one such example. Now that I've given you a high level
overview of the different types of analyses, let's look at the tools
available to help you perform each one. Descriptive and diagnostic analytics
usually rely on analytic tools that can handle manipulation
of large sets of data or that help visualize and
interact with summarized information. Examples include SQL,
Oracle database or Oracle DB, Hadoop/Spark, Tableau, QlikView, Microsoft Access,
SAS, R, Python and
various statistical packages within them. Predictive and prescriptive analytics
have traditionally relied on analytics tools that have significant
mathematical modeling capabilities or scenario planning or
simulation capabilities. Examples of these tools include SAS,
R, SPSS, Python, and
various packages associated with them. Optimization tools like Garrobi,
ILOG, RiverLogic, etc. Simulation tools like Vensim,
AnyLogic, STELLA. Machine learning and deep learning tools, like Scikit,
TensorFlow, Caffe, Theano, etc. Natural language processing
tools like NLTK or Natural Language Tool Kit or OpenNLP. In this video, we talked
about five different types of
analytics you can use to analyze data. Descriptive analytics helps you understand
the current state of the problem and answer the question, what happened? Diagnostic
analytics helps you
understand why it happened or the underlying causes for
the observed data. Predictive analytics helps you understand what could happen in
the future
given certain conditions. Prescriptive analytics helps you
understand the right course of actions needed today to address future concerns.
Adaptive and autonomous analytics
helps you answer the question of how to continuously adapt to change. We also took
a very high level look
at the tools available to help you perform these analytics. Later in the course, we
will take a much
closer look at what some of these tools do and how they provide value. In the next
video, you'll have
the opportunity to hear some of our PwC professionals talk about data and
analytics at PwC. And how it plays a role in over tax,
assurance and advisory or consulting practices. [MUSIC]