2. Should I be investing in a metric
+ Set a baseline, what is normal and what’s not for your brand or company.
+ Metric 1- Volumes
+ coca-cola has 40 million fans on twitter
+ Lady gaga has 59 million fans on fb
+ Should we be investing in building social media followings, what’s the
argument to do that?
+ Look at how can we use these metrics , is it that more social media
followers I have , more potential exposure I can have of my marketing
messages with those individuals.
3. Metric – sentiment scoring
+ 0-100
+ TOPSY performs sentiment analysis
+ Takes average of sentiment value across all the
comments
+ Positive=+1, neutral=0, negative =-1
+ Or sentiment score =Positive/negative ratio
4. Social media Vs traditional
marketing research studies
+ Social media sentiment may only represent subset of the
broader population
+ Brand and product sentiment
+ Social media data consist of sentiments across the
products of a single brand
+ Taking care of variation in sentiment across platforms like
blogs, twitter, etc while doing analysis.
5. Volume and sentiment
+ IF both are high, its success
+ If sentiment is low but volume is high, it’s a red flag and
good time to social media monitoring.
+ Understand, how invested are the contributor on social
media, how influential is the contributing group, to avoid
nusance
8. Opinion science and dynamics
+ Should I post? -> What do I Post? -> Where do I post?
+ There is filtering at all levels.
+ Why do people engage in SM activity is for two reasons:
+ Affiliation- want to be a part of group
+ Persuasion goal- I want to influence your decision
making.
+ Implication for companies: Need to be cautious of social
media dynamics to spot deviations.
9. Product ratings change over time and is
a driver for consumer behaviour
+ Variation is contributed by the
Difference in activist and
Low involvement groups, it
Is a general dynamics to be
Kept in mind.
10. Implications of social
media strategy
+ Look at conversations
happening, what kind of
conversations are
happening, what kind of
impact conversations having
on performance, what can
you do to involve in
conversations, can I shape
the conversation that
benefits my brand/org, we
can see long term benefits of
being associated on social
media.
11. Influence of news events is connected to
behavior of stock markets
+ Extracting indicators of economic behavior from SM.
+ Tweets are a perfect indicators of public sentiment.
+ OpinionFinder, an online mood tracking service.
+ GPOMS – google profile of mood states, provides a detailed view of change in
public sentiment.
+ Implementing a prediction model- SOFFNN, self-organizing fuzzy neural network
,has the ability to predict closing stock values, with an accuracy of 87.6% in
predicting the daily up and down
changes.https://www.theatlantic.com/technology/archive/2010/10/predicting-
stock-market-changes-using-twitter/64897/
12. Forecasting models for Marketing
Decision
+ Predict Demand using time series analysis, firstly builds the
model based on demand for a given period of 3 years using
regression model. The predictor variables here are the a)
week numbers b) then try using 12 months dummy variables
(keeping Jan as the reference variable) . Check the absolute
error (difference between the predicted demand and given
demand). Then average the absolute error on the evaluation
data only. And see which model is better.
13. Intro to customer Analytics: Tool
box
+ Types of customer-level data:
Choice-which product (coca-cola, pepsi, etc)
count- how much qty a customer purchase
timing- duration data--when do I become a customer, how long I stay, how
long b/w visits to site, how long b/w purchases.
multivariate- combining above types of data.
Applications of Marketing:
Measuring marketing effectiveness(ROI), click stream data and online
advertising, Loyalty programs and CRM, social media.
14. Regression follows normal
distribution but:
+ Even though the output of regression models is a normal
distribution with N(0,var) , but with customer data it may
not be necessarily the case.
+ Therefore, we use Bernouilli distribution for outcomes
yes and No having prob ‘p’ and ‘1-p’ respectively.
15. Types of univariate data in
customer analytics
+ Continuous (linear regression)
+ Count
+ Choice
-between two options(binary choice)
-between n options (multinomial choice)
Timing
17. Big data-marketing
+ Predicting each customer’s next transaction
+ Predicting requires knowing consumers media
preferences, scrutinizing her shopping habits, cataloging
her interests, aspirations and desires.(It is a short-term
tactical advantage)., this will compete away their
marginal profits, when all competitors have established
prediction, no sustainable adv in learning next buy.
18. Strategic marketing
+ Strategic marketing requires long term customer
stickiness, loyalty and relationships.
+ Not just knowing what will trigger next purchase, but
what will get customer to remain loyal, not just what price
customer willing to pay but what is customer’s lifetime
value., and what will prevent the customer to switching to
a competitor, when they offer better price.
19. Big data can help design
information
+ Ex: recommendation engines create value for customers
by reducing their search time and evaluation cost.
+ Augmenting commodity utilities with customized usage
info. – by Opower.
+ Crowd sourced data – allows consumers to learn from
other consumers, comparing themselves to other
consumers.
20. How big data creates customer
value
+ Answering ques like What info will help customers
reduce their cost or risk?
+ Ex: Uber, eBay, Netflix, Amazon crunch data about
ratings of service providers and sellers to reduce
customer risk. Now, customers are looking at more
granular answers like what other customers like me think
of this product/service.
21. Consumers want more personalized
experience
+ Analytics enable companies to understand their customers.
+ Businesses needs to measure specifically what each customer
want and link their processes and resources to provide it. It is
possible due to advanced manufacturing and distribution
technologies.
+ Consumers are ready to share data in return on receiving
more personalized services or product.
22. Forecasting models for Marketing
Decisions
+ Forecasting Demand of consumer(Marketing
perspective)
+ Forecasting Revenue for products at individual
consumer, store level, national level
+ We need to determine input components for our model.
+ Approaches to forecasting: smoothing methods, Auto-
regressive model, Regression-based method.
23. Smoothing models
+ Recent observations are good predictors of observation in
near future.
+ Time-series contain fluctuations that don’t aid in forecasting.
+ Averaging over recent observation, will smoothen
fluctuations.
+ Simple moving average: of length L, for most recent L
observations to make prediction for next observation.
Yi+1=sum(y0-i)/L
24. Weighted moving average:
smoothing model
+ Putting 50% of weight to most recent observation and
remaining 50% to the rest.
+ This is going to be putting more emphases on recent
observation.
+ yi+1=sum(wiyi)/sum(wi)
+ Smoothing models are good for short term forecasting,
but not with observations where there is a trend.
26. Regression based modelling
+ What would the regression equation look like:
+ Trend component
+ Cyclical component
+ Seasonal component
+ What Assumptions are we making about the
components.
27. Calibration period and forecasting
period
+ Calibration period is the data on which we have built the model
and apply it to forecasting data.
+ Simple linear regression : Y=intercept + slope(Week#)
+ Excel commands:
+ =intercept(y_range, x_range)
+ =slope(y_range, x_range)
+ We are going to predict intercept and slope, using predictor week#
28. If the trend is growing but at a
decreasing rate then apply logarithm of
week#
+ Y=intercept + slope*ln(week#)
+ If the trend is increasing at a fast rate, apply square or
cubic polynomial of week# in your algorithm.
+ If the trend is decreasing, can also use square root of
week#
29. Adding month based dummy
variables
+ Calibrated using 2nd year of data with monthly effects,
yields much better forecast.
30. Customer centric analytics: Bernoulli
Distribution
Modelling binary choice data: yi=1,pi and 0,1-pi
E(Yi)=1*P+0*(1-P)=P
Two common models for binary data:
Logit model, logit(pi)=log(pi/1-pi)=XB
Probit model, pi=exp(XB)/(1+exp(XB))
31. Time duration Models: durations in
Marketing
+ Service research
time until acquisition
+ Behavioral research
response latency
Forecasting
-New product diffusion
Customer Base Analysis
- Is the customer still alive?
32. Timing model
+ Let us assume, customer has Probability of dropping
service: p
+ Probability of keeping service : 1-p
+ Decision made each month are independent of each
other.
+ What is probability of dropping service in month ‘t’=p(1-
p)t-1
33. Managing customer equity
+ Customer acquisition
+ Depth and breadth of customer relationship
+ Customer retention
+ Two approaches to customer acquisition: Direct and Indirect approach
+ Two strategies of customer acquisition: Broad Strategy and selective
strategy
Broad strategy: Telemarketing, list, brokers, detailing
Selective strategy: customer profiling, scoring
34. Customer Acquisition
+ Logistic Regression would be appropriate for any of the
above scenarios, with outcome yes/no, whether the customer
acquired.
+ Companies marketing strategies emphasizes on customer
acquisition process.
+ Challenges in acquiring customers: how much should a
company spend in acquiring the perspective customer, which
perspective customer should they spend on, what will be the
revenue that will be produced with acquiring the customer.
35. Customer lifetime value
+ After acquiring the customer, how much value is the customer
worth more after acquiring for a certain period of time is
called residual lifetime value.
+ CLV=sum( (M*S(t)) / (1+d)t )
Where, M= Margin
S(t)=survival probability
d = discount value