UNIT-5

APPLICATIONS
Retail Analytics
 Retail analytics is the process of providing analytical data on inventory levels, supply chain
movement, consumer demand, sales, etc. that are crucial for making marketing, and procurement
decisions.
 To understand the roles analytics plays in retail, it is useful to break down the business decisions taken
in retail into the following categories:
1. Consumer
2. Product
3. Human resources
4. Advertising
Retail Analytics
1. Consumer: Personalization is a key consumer-level decision that retail firms make.
 Personalized pricing by offering discounts via coupons to select customers is one such decision.
 This approach uses data collection to better understand a customer’s purchase patterns and
willingness to pay and uses that to offer personalized pricing.
 Such personalization can also be used as a customer retention strategy.

2. Product: Retail product decisions can be broken down into single product and group of product
decisions.
 Single or individual product decisions are mostly inventory decisions: how much stock of the product to
order, and when to place the order.
 At the group level, the decisions are typically related to pricing and assortment planning.
 To make these decisions, predictive modeling is called for to forecast the product demand and the price-
response function, and essentially the decision-maker needs to understand how customer reacts to price
changes.
 A fine understanding of consumer choice is also needed to understand how a customer chooses to buy a
certain product from a group of products.
Retail Analytics
3. Human resources: The key decisions here are related to the number of employees needed in the store
at various times of the day and how to schedule them.
 To make these decisions, the overall work to be completed by the employees needs to be estimated.
 Part of this is a function of other decisions, such as the effort involved in stocking shelves, taking
deliveries, changing prices, etc.
 There is additional work that comes in as a function of the customer volume in the store. This includes
answering customer questions and manning checkout counters.
4. Advertising:
 In the advertising sphere, companies deal with the typical decisions of finding the best medium to
advertise on (online mediums such as Google Ad words, Facebook, Twitter, and/or traditional
mediums such as print and newspaper inserts) and the best products to advertise.
 They may also purchase other items which have a greater margin.
Retail Analytics
Complications:
There are various complications that arise in retail scenarios that need to be overcome for the successful
use of retail analytics. These complications can be classified into
(a) those that affect predictive modeling and
(b) those that affect decision-making.
Methodologies:
1. Product-Based Demand Modeling- exponential smoothing and ARIMA models are typically focus on
forecasting sales and may require un censoring to be used for decision making.
2. Incorporating Consumer Choice in Demand Modeling- This directly motivates modeling customer
preferences over all the products carried by the retailer.
3. Business Challenges and Opportunities- A good example of such an approach is the “buy online, pick
up in store” (BOPS) approach that has become quite commonplace.
4. Retail Startups - In terms of data collection, there are many startups that cater to the range of retailers
both small and large.
Retail Analytics-Examples

 Analytics has revealed that a great number of customer visits to online stores fail to convert at the last
minute, when the customer has the item in their shopping basket but does not go on to confirm the
purchase. Theorizing that this was because customers often cannot find their credit or debit cards to confirm
the details, Swedish e-commerce platform Klarna moved its clients (such as Vistaprint, Spotify, and 45,000 online
stores) onto an invoicing model, where customers can pay after the product is delivered.
 Amazon has proposed using predictive shipping analytics6 to ship products to customers before they even
click “add to cart.” According to a recent trend report by DHL, over the next 5 years, this so-called psychic
supply chain will have far reaching effects in nearly all industries, from automotive to consumer goods. It
uses big data and advanced predictive algorithms to enhance planning and decision-making .
Marketing Analytics
 Marketing analytics can be defined as a “high technology enabled and marketing science model-
supported approach to control the true values of the customer, market, and firm level data to enhance
the effect of marketing strategies”.
 Basically, marketing analytics is the creation and use of data to measure and optimize marketing
decisions.
 Marketing analytics comprises tools and processes will help in various aspects of marketing such as
target marketing and segmentation, price and promotion, customer valuation, resource allocation,
response analysis, demand assessment, and new product development.
 These can be applied at the following levels:

1. Firm
2. Brand/product
3. Customer
Marketing Analytics
1. Firm: At this level, tools are applied to the firm as a whole. Instead of focusing on a particular product
or brand, these can be used to decide and evaluate firm strategies.
2. Brand/product: At the brand/product level, tools are applied to decide and evaluate strategies for a
particular brand/product.
3. Customer: Tools applied at customer level that help in segmenting and targeting customers.
Marketing Analytics
We can segment marketing analytics into the following processes and tools:
1. Multivariate statistical analysis: It deals with the analysis of more than one outcome variable.
2. Choice analytics: Choice modeling provides insights on how customers make decisions.
3. Regression models: Regression modeling establishes relationships between dependent variables and independent
variables.
4. Time-series analytics: This section consists of auto-regressive models and vector auto-regressive models for time-
series analysis. These can be used for forecasting sales, market share, etc.
5. Nonparametric tools: They are used when the data belongs to no particular distribution.
6. Survival analysis: It is used to determine the duration of time until an event such as purchase, and conversion
happens.
7. Sales force /sales analytics: This section covers analytics for sales, which includes forecasting potential sales,
forecasting market share, and causal analysis.
8. Innovation analytics: It deals specifically with new products.
9. Conjoint analysis: It covers one of the most widely used quantitative methods in marketing research.
10. Customer analytics: These can be used for segmenting customers and determining value provided by customers.
Social Media and web Analytics
 Social media has created new opportunities to both consumers and companies.
 Companies can analyze data available from the web and social media to get what consumers want.
 Social media and web analytics can help companies measure the impact of their advertising and the effect of mode of
message delivery on the consumers.
 Companies can also turn to social media analytics to learn more about their consumers.
 Social media analytics involves gathering information from social networking sites such as Facebook, LinkedIn and
Twitter in order to provide businesses with better understanding of customers.
 It helps in understanding to creating customer profiles and evolving appropriate strategies for reaching the right
customer at the right time.
 Search engine optimization (SEO) is another technique to acquire customers when they are looking for a specific
product or service or even an organization.
 Advantages:
 It enables businesses to identify and encourage different activities that drive revenues and profits and make real-time.
 It can help businesses in targeting advertisements more effectively and thereby reduce the advertising cost while
improving ROI.
Social Media and web Analytics
1. Display Advertising in Real Time: The Internet provides new scope for creative approaches to
advertising.
There are different types of display advertisements. The most popular one is the banner advertisement.
This is usually a graphic image, with or without animation, displayed on a web page. These
advertisements are usually in the GIF or JPEG images if they are static, but use Flash, JavaScript or video
if there are animations involved.
2. How to Get the Advertisements Displayed?
There are many options for getting the advertisements displayed online. Some of these are discussed
below.
One of the most popular options is placing the advertisements on social media.
You can get your ads displayed on social media such as Facebook, Twitter and LinkedIn.
In general, Facebook offers standard advertisement space on the right-hand side bar.
These advertisements can be placed based on demographic information as well as hobbies and interests
which can make it easy to target the right audience
Social Media and web Analytics
Programmatic Display Advertising:
 Programmatic advertising is “the automation of the buying and selling of desktop display, video, FBX, and mobile ads
using real-time bidding.
 The main components of programmatic display advertising are as follows:

(a) Supply-Side Platform (SSP) The SSP helps the publishers to better manage and optimize their online advertising space
and advertising inventory. SSP constantly interacts with the ad exchange and the demand-side platform (DSP).
(b) Demand-Side Platform (DSP) The DSP enables the advertisers to set and apply various parameters and automate the
buying of the displays. It also enables them to monitor the performance of their campaigns.

(c) Ad Exchange Ad exchanges such as Facebook Ad Exchange or DoubleClick Ad Exchange facilitate purchase of
available display inventory through auctions. These auctions are automated and take place within milliseconds, before a
web page loads on the consumer’s screen. These enable the publishers to optimize the price of their available inventory .
(d) Publisher Publishers are those who provide the display ad inventory.

(e) Advertiser The advertiser bids for the inventory in real time depending on the relevance of the inventory.
Social Media and web Analytics
Flowchart for real time bidding:
Health care Analytics
 Health care analytics is a subset of data analytics that uses both historic and current data to improve
decision making, and optimize outcomes within the health care industry.
 Health care analytics is not only used to benefit health care organizations but also to improve the
patient experience and health outcomes.
 Predictive analytics is the use of historical data to identify past trends and project associated future
outcomes. In the health care industry, predictive analytics has many impactful uses, such as
identifying a patient’s risk for developing a health condition, streamlining treatment courses, and
reducing a hospital’s number of 30-day readmissions (which can result in costly fines for the hospital).
 Prescriptive analytics is the use of historical data to identify an appropriate course of action. In the
health care industry, prescriptive analytics is used to both direct business decisions and to literally
prescribe treatment plans for patients.
Health care Analytics

Benefits:
 Improved patient care, such as offering more effective courses of treatment.
 Predictions for a patient’s vulnerability to a particular medical condition.
 More accurate health insurance rates.
 Improved scheduling for both patients and staff.
 Optimized resource allocation.
 More efficient decision-making at the business and patient care level.
Financial Analytics
 Data analytics in finance is a part of quantitative finance.
 Quantitative finance primarily consists of three sectors in finance—asset management, banking, and
insurance.
 Across these three sectors, there are four tightly connected functions in which quantitative finance is
used—valuation, risk management, portfolio management, and performance analysis.
 Data analytics in finance supports these three sequential building blocks of quantitative finance,
especially the first three— valuation, risk management, and portfolio management.
Q-Quants:
 In the Q-world, the objective is primarily to determine a fair price for a financial instrument, especially
a derivative security, in terms of its underlying securities.
 The price of these underlying securities is determined by the market forces of demand and supply.
 The demand and supply forces come from a variety of sources in the financial markets, but they
primarily originate from buy-side and sell-side financial institutions.
 The Q-quants typically have deep knowledge about a specific product.
Financial Analytics

P-Quants:
 The P-world showed that the conventional investment evaluation criteria of net present value (NPV)
needs to be explicitly segregated in terms of risk and return.
 P-world was the capital asset pricing model (CAPM) converted “business school” framework to an
“economics department” model.
Financial Analytics

Stage I: Asset Price Estimation:

 The objective of the first stage is to estimate the price behavior of an asset.

Step 1: Identification
 The first step of modeling in the P-world is to identify the appropriate variable which is different for
distinct asset classes.
 The most common process used for modeling a financial variable x is the random walk:

where εt is the error term and is random.

 A stationary process is one whose probability distribution does not change with time.
 As a result, its moments such as variance or mean are not time-varying.
 Choosing the right financial variable is as important as the modification made to it.
Financial Analytics
Financial Analytics

Step 2: i.i.d process

 The second step is to transform the identified variable into an independent and identically
distributed (i.i.d.) process.
 Once the financial variable that is of interest is identified, the next step in data preparation is to obtain
a time series of the variables that are of interest.
 These variables should display a homogenous behavior across time.
 The models used would depend on the features displayed by the financial variable.
 Techniques like autoregressive conditional heteroscedasticity model (ARCH) or generalized
autoregressive conditional heteroscedasticity model (GARCH) are used to factor out volatility
clustering.
 If the variable displays some kind of mean reversion, one might want to use autoregressive moving
average (ARMA) models if it is a univariate case or use vector autoregression (VAR) models in
multivariate scenarios.
Financial Analytics

Step 3: Inference
 The third step in estimation after the financial variable is identified and after we have gotten to the
point of i.i.d. shocks is to infer the joint behavior of i.i.d. shocks.
 From the daily index levels, the 1-day returns are calculated as follows:

 This return rt itself is not distributed in an i.i.d. sense.

 we model variance as follows:
Financial Analytics

Step 4: Projection
 The fourth step is projection.
 A commonly used technique for this is the Monte Carlo simulation. We first pick a random number
from the standard normal distribution say x and returns R.

 We project the financial variable in the Qspace using risk-neutral parameters and processes. This also
helps us to understand how the P- and Q-worlds converge.
Step 5: Pricing
 The fifth step is pricing which logically follows from projection.
 What pricing allows us to do is arrive at expected profit or loss of a specific instrument based on the
projections done in Step 4
Financial Analytics

Stage II: Risk Management

 The second stage of data analytics in finance concerns risk management.
 It involves analysis for risk aggregation, risk assessment, and risk attribution.
 The framework can be used for risk analysis of a portfolio or even for an entire financial institution.

Step 1: Aggregation
 The aggregation step is crucial because all financial institutions need to know the value of the portfolio
of their assets and also the aggregated risk exposures in their balance sheet.
Step 2: Assessment
 Assessment of the portfolio is done by summarizing it according to a suitable statistical feature.
 Assessment is done by calculating the risk of the portfolio using metrics such as threshold persistence
(TP) or value at risk (VaR).
 VaR is a measure of the risk of a portfolio under normal market conditions over a certain time horizon.
 TP is a horizon over which the cumulative return remains below the threshold β
Financial Analytics

Stage II: Risk Management

Step 3: Attribution
 Once we have assessed the risk of the portfolio in the previous step, we need to now attribute the risk
to different risk factors.
 What financial institutions typically do is to attribute risk along a line of business (LoB).
 Stand-alone capital is the amount of capital that the business unit would require, if it were viewed in
isolation.
 Incremental capital measures the amount of capital that the business unit adds to the entire firm’s
capital.
 Component capital or allocated capital measures the firm’s total capital that would be associated
with a certain line of business.
Financial Analytics

Stage III: Portfolio Analysis

Step 1: Allocation
 After having aggregated the portfolio, assessed the risk, and then attributed the risk to different lines of
businesses, we move on to changing the portfolio for the entire firm, for a division or an LoB for
optimal allocations.
Step 2: Execution
 We have to execute the respective trades for us to be able to get to the desired portfolio risk levels.
 Execution happens in two steps.

i) Order scheduling involves deciding how to break down a large trade into smaller trades and timing
each trade for optimal execution.
ii) order placement which looks at execution of child orders, and this is again addressed using data
analytics.
Execution is almost always done programmatically using algorithms andis known as high-frequency