2. Brand Share & Industry Size: Will
the twain ever meet?
Using a portfolio of techniques to improve
accuracy of market volume evolution in
price change scenarios
April 2015
Sreeram Srinivasan, IMRB International
Ranjana Gupta, IMRB International
3. 30
40
Marketer’s pricing dilemma
30 Market
volum
e 100
bn
Marke
t
volum
e
70 bn
With new prices, my share will
grow…
… but will my volume
also grow?
Market share grows….. ….but, market size shrinks
4. Both the consumer and the macro variables needed to
answer the questions..
C
o
n
s
u
m
e
r
M a c r
o
Brand choice
Same brand, switching etc.
In context of price
Category choice
Frequency, consumption,
substitutes etc.
Industry
Past volumes,
substitutes, prices etc.
Economy
Income, affordability,
Inflation etc.
Relies on past trends
Future may be different
Future oriented.
Past learnings not
fully leveraged
5. Hence for accurate volume forecasts, no single
methodology can provide complete answer
Consumer research and macro-economic models should
be integrated
Choice Model
Consumer Research
• Brand shares
• Switching
Econometric
Macro Modelling
• Market size
6. An approach to integration of
choice and econometric model
However, before integration, the individual
tools need to be refined, adapted…
7. …to account for the nature of the category
How to account for occasion?
Is it an impulse or
considered purchase?
Is it a repertoire or non-
repertoire category?
Do number of units matter?
What about frequency of
purchase?
How can we ensure that the
respondent reacts to only
relevant offers?
9. The questioning technique uses FORCE principle
to make the consumer response more realistic…
Familiar
Only brands that the respondent interacts with / likely to interact with shown –
customized for each respondent real time
Evoked set created using respondent’s current repertoire, past usage and future
disposition
Occasio
n
Respondent’s answers using an occasion as a context in occasion led categories
like CSDs, Snacks etc. There can be other household purchase categories where
occasion is irrelevant
Repertoir
e
Respondents allowed to select multiple brands, SKUs and units, as they might in
real life
Channel
Primary channel of purchase identified and specified in the questioning
Event
Recurren
ce
Frequency of purchase
10. Here’s an example…
around 20 such choice tasks shown
Imagine you are doing your monthly grocery shopping from the supermarket and you
have to buy bathing soaps. On the shelf you see the following brands with the given
prices? Which brands are you likely to buy?
You can choose as many brands as you like. Or you can walk out of the shop without
buying any.
Do state the number of packs that you would be buying.
I will not buy any
anything
2
Johnson’s Baby
Soap
75 gms
Rs. 30
Dove
75 gms
Rs. 40
Santoor – pack of 4
100gms X 4
Rs. 50
Now at Rs. 40
Pears
100 gms + 25 gms
extra free
Rs. 25
Rexona
200 gms
Rs. 25
I would buy a
shower gel
Lux
100 gms
Rs. 10
1
11. Consumer choices are converted into utilities – two levels
of calculations
Main Effect
For every level within each attribute at a
respondent level
Example: Utility or preference for Brands like
Dove, Pears etc., for SKUs like 100 gms,, 75
gms and for various price levels
Cross Effect
Interaction between attributes
Example: Utility for Dove by itself and for Dove
at a particular price may be different
The utilities are transformed into
share of preference and
weighted to give the shares in
various scenarios
For all existing brands & SKUs,
the current levels of distribution
are built into the model – to
ensure current scenario shares
are in line with actual market
shares
Respondent level estimation of preference helps in calculation of
gains and losses from one brand to another
12. The output
Current
Scenario
New price
scenario
(Client’s
brands cut
prices by
10%)
Company
Brand X 20.0% 21.8%
Brand Y 17.0% 17.2%
Brand A 10.0% 9.9%
Brand B 5.0% 4.4%
Brand P 8.0% 7.6%
Brand Q 15.0% 14.8%
Brand R 10.0% 9.7%
Brand S 15.0% 14.6%
39%37%
Brand X
Gaining From
Brand B, Brand
S
Net Gain/Loss 1.8%
14. The input
Past volumes Population
Substitute
categories
(Real) Price Purchase frequency
No. of packs(Real) Income Basket size
15. The statistics
Options
Simple Regression
Volume = fn (Price)
Easy but can lead to situations like
increase price to increase
volumes!
Simple Time
Series
Volume = fn (Past volumes)
Builds in past volume trends but assumes that
history will definitely repeat itself
Price
Volume
16. The statistics
Adopted method
ARIMA
(Auto Regressive Integrated Moving
Average)
Volume = fn (Real Price, Past Volume, Real Income, etc….)
Moving average included – accounts for any possible prediction error in
previous time periods
Accounts for autocorrelation
Better prediction accuracy
17. The statistics
Adopted method
ARIMA
(Auto Regressive Integrated Moving
Average)
Deseasonalized data - predict organic change in volumes
Model by major sub-groups to account for different trends – break the
market
Price gap between sub-groups used – inter-movement built in
22. Results validated across markets
0
20
40
60
80
100
Predicted volume accuracy by brands in various markets
(74 data points in this graph)
Average :
88%
The trick: improve accuracy of the individual models
23. Identifying the relative impact of
touchpoints: A tailored statistical
technique for real-time data
Shane Baxendale, Cranfield School of Management
Heval Ceylan-Gilchrist, MESH
MRS ADVANCED ANALYTICS NETWORK
30th April 2015
24. Why are we here today?
Real-time Experience Tracking Methodology
Analysis - using linear mixed-effects
regression
24
25. Our thinking
Consumers experience brands through multiple channels
(not just advertising!)
Brand experiences influence a consumer’s attitude toward
brands
The majority of existing literature focusses on the impact of
one or two types of experience
What impact are different encounters having on
consumer attitudes toward the brand?
25
*Baxendale S., Macdonald E.K., Wilson H.N., (2015), The impact of different
touchpoints on brand consideration, Journal of Retailing, 37(2), 203.
27. Text us whenever you
see, hear or experience
anything to do with the
following brands…
Text framework
27
BRAND: A)Brand A B)Brand B C)Brand C
D)Brand D E)Brand E F) Other
OCCASION: A)TV B)Poster/Billboard
C)Radio D)In store E)Cinema F)Newspaper
G)Magazine H)Conversation
I)Online/Mobile J)Mailing/leaflet K)Me
Purchasing L)Me using M)Someone else
using N)Sponsorship O)Other
FEELING: 5)Very positive 4)Fairly positive
3)Neutral 2)Fairly negative 1)Very Negative
CHOICE: 5)Much more likely to choose
4)Slightly more likely to choose 3)No
difference 2)Slightly less likely to choose 1)
Much less likely to choose
28. Imagine you experienced Brand A Online…
…you would text:
28
a 5i 5
CHOICE:
5) Much more likely to
choose
4) Slightly more likely to
choose
3) No change
2) Slightly less likely to
choose
1) Much less likely to
choose
ENGAGEMENT:
5) Very positive
4) Fairly positive
3) Neutral
2) Fairly negative
1) Very negative
BRAND:
a) Brand A
b) Brand B
c) Brand C
d) Brand D
e) Brand E
f) Other
OCCASION:
a) TV
b) Poster/Billboard
c) Radio
d) In store
e) Cinema
f) Newspaper
g) Magazine
h) Conversation
i) Online/Mobile
j) Mailing/Leaflet
k) Me purchasing
l) Me using
m) Someone else using
n) Sponsorship
o) Other
Which brand was it? Where did you
experience it?
How likely did it make you
to choose the brand next
time?
How did it make you feel?
29. Now tell us more in an online diary…
29
This is an individual’s experience log By clicking on each entry, the experience can
be expanded upon in detail
Wednesday 13th February 2012,11:54
Wednesday13th February2012,11:54
Wednesday13th February2012,10:22
Tuesday512h February2012,18:46
Tuesday12th February2012,13:05
Tuesday12th February2012,08:38
Brand A,Online, Very Positive,
MuchMoreLikely toChoose
Brand C,Conversation, Very Negative,
MuchLess Likely toChoose
Brand E, TV, Fairly Positive,
Slightly More Likely to Choose
Other, Instore, Fairly Positive,
Slightly morelikely toChoose
Brand B, Mailing/Leaflet, Slightly Negative,
Nochange
Brand A
Brand’s website
Very positive
Much more likely to choose
I was looking on the brand website to find out
more information about the product range. Looks
like there are some good options.
Look for product info
13/02/2012, 11:54
Please tell us exactly what yousaw? :
What was the purpose ofyour online activity? :
Brand’s website
Ad from brand
In the news
Social networking site
Price comparison site
Other
For each level of data captured
in real-time we can tailor extra
questions to get more granular
information in near-time
30. Data
For one individual
Brand A Brand B … Brand N
Consideration Wk0
Consideration Wk1
Consideration Wk0
Consideration Wk1
Consideration Wk0
Consideration Wk1
Freq. & Pos.
Brand Ad
Retailer Ad
In Store
WOM
…
Freq. & Pos.
Brand Ad
Retailer Ad
In Store
WOM
…
Freq. & Pos.
Brand Ad
Retailer Ad
In Store
WOM
…
30
+ve -ve -ve
36. The contents of this document are the sole and confidential property of Lieberman Research
Worldwide, and may not be reproduced or distributed without the express written permission
of Lieberman Research Worldwide.
Prepared for CLIENT
TITLE
LRW Europe
BAYESIAN ANALYSIS
FOR MARKETING IMPACT
April
2015
LRW Europe
1, Heathcock
Court, 415, Strand
London
WC2R 0NT
Prepared by:
Adele Gritten &
Graham Williams
for MRS Advanced
Analytics
Conference
When I talk to companies about this most people tell me they expected me to be older
Volume = α + β. price + ¥. Last year Volume (Autoregressive component)+ Ɵet (Moving Average)
Where ¥. Last year Volume is the Autoregressive component – accounts for the fact that Yt is dependent on Yt-1
I = Yt – Yt-1 (The difference in the volumes between two consecutive periods)
Ɵet is the MA term which assumes et is dependent on et-1
Simple regression will not work due to autocorrelation
AR is used to account for autocorrelation whereas MA is used to account for any errors in the previous time period
We do not use the integration factor into our model and only use ARMA due to the fact that we already deseasonalize the input variables. Then there would be an issue of double correction.
Rationale for the analysis
But what data do we have? And what links do we hypothesise?
So what do we want to include in the model, why, and what do we have to consider as a result of this?
How should we include frequency? Which is best from the model? Does this make sense / are we comfortable with choosing this over other options?
How should we include positivity? Which is best from the model? Does this make sense / are we comfortable with choosing this over other options?
What are the results?
Done to a large extent on gut feel.
We will see that there are large money to make by relatively simple steps.
EXAMPLES of categories
Grocery store, like Co-op. Second in Sweden.
Clarify that we are looking at allocating space for each of those product categories.
Done to a large extent on gut feel.
We will see that there are large money to make
Go through the matrix quickly!
Two criteria: Drive margin in categories (+ or -), and be relatively un-correlated to each other