Much criticism has been levied towards marketing-mix modeling recently. This article shows innovations and proposes solutions for reinventing this powerful marketing measurement tool

1. Next Generation
Marketing-Mix Modeling

2. Content
• Marketing-Mix Modeling at the Crossroad & its Shortfalls 4-5
• Next-Generation Marketing-Mix Models 5
– It starts with the basics of blocking and tackling 6
– No Silos for marketing and Marketing Synergies 7-9
– Measuring the Long-Term Effects of Advertising 10-12
– Addressing Multi-Touch Marketing Attribution & Digital ROI 17-22
– Building in the Voice-of-the-Customer into MM Models 23-36
– Simulation and Real-Time Marketing-Mix Models 37-38
• Analytics Portfolio 39
• Global-Analytics Partners 40-42
• Contact Us 43
2

3. Marketing-Mix Modeling at a Crossroad: Marketing-
Mix Models Falling Short
• Marketing-mix models have been referred to as the de facto standard for marketing measurement.
• Yet, with all the dramatic change in the marketing landscape, there has been little to no change in how
these models calibrate and measure marketing ROI over the past 30 years. The tools have failed to adapt
and this has stimulated a chorus of criticisms, wondering if Marketing-Mix Modeling is obsolete
• In a recent AdAge article entitled “Marketing-Mix Models Get Pushback As Media Landscape Changes”
(Apr. 2013), we hear a growing chorus of critics.
• “Some critics believe the models have been wrong all along, and work even worse after three decades of
change in the media landscape. They say the models underestimate the impact of advertising, particularly
of broad-reach network TV; overstate the value of price promotion, mislead marketers into buying thinly
rated programming; wrongly downplay risks of going dark for weeks on end; and fail to account for how
online search has made all advertising more effective”.
• Marketing-Mix Modeling has been criticized for its focus only on short-term marketing response and, for
the most part, not adapting to more advanced methods for measuring long-term marketing impacts and
customer loyalty and repeat-purchase dynamics. Because MM models mostly focus only on short-term
effects of adverting, this has relegated the ad investment to negative ROI in about 85% of cases. Without
consideration of the larger long-term effects of advertising, there is a major under-estimation of the true
value of advertising and marketing and this has caused an incorrect focus on short-term only marketing.
• There seems to be a consensus that a new paradigm needs to be developed for marketing measurement.
However, most MMM vendors have not put forth answers to some of these known short-comings.
3

4. Marketing-Mix Modeling at a Crossroad
 A lot of the underlying method of MM models relies on statistical assumptions which assume
complete independence of the marketing drivers. Marketing can not be relegated to silos. All of
this ignores the synergistic and more complex symphony represented by multiple marketing
activities acting together.
 Marketing is all about a brand’s “relationship” with customers. Yet, through all of this, “the voice-of-
the-customer” remains silent in the MM modeling exercise. The proper step for MM modeling
requires developing a means for measuring the various aspects of the customer-brand-experience
and clearly understanding the brand value proposition directly from the customer’s perspective
 MM modeling focuses pretty much exclusively on marketing channels. Media effectiveness is seen
through the lenses of TV, radio, print or digital channels. All of this ignores that marketing is really
about “message and communication”. MM Modeling needs to change its focus and measurement
towards the effectiveness of ad message and creative. This is a major missing piece that limits MM
Modeling from being an effective and powerful tool for forming marketing communications
strategies.
 MM modeling has also been criticized for its inability to accurately address the issue of digital multi-
touch attribution. Single equation econometric models often yield biased solutions, with extreme
solutions favoring the media or activity closest in proximity to the sales conversion and giving no or
little credit to key touch-points along the customer journey path. This means that more advanced
“multi-equation” econometric solutions need to be employed which will better account for and
accommodate the actual pathways and media touch points of the customer journey.
 We think that it is time for something different: a paradigm shift.
4

5. Incremental Contribution from marketing
Return on Investment per £1 spent Optimize spend, maximise sales
Develop relationship between sales and drivers
Next-Gen Marketing Mix starts with blocking & tackling
5

6. …pushing the boundaries.
Next Generation Marketing-Mix Models
TV
RADIO
NEWSPAPER
PAIDSEARCH
6
Effectiveness Modeling (econometrics) has not changed a great deal over the last 30 years.
We fundamentally believe that marketing and media channels do not operate in silos; but most
statistical models treat them as such. We employ advanced non-linear methods which account
for direct and indirect effects from marketing drivers.

7. Marketing Synergies
-
500
1,000
1,500
2,000
2,500
3,000
3,500
4,000
Separate Impact Simultaneous Activation
Marketing Synergy
Radio
Digital
Radio and Digital Media are most effective when executed together. There is about a
+45% synergy or additional lift due to simultaneous activation.
7

8. Marketing Synergies
-
1,000
2,000
3,000
4,000
5,000
6,000
Separate Impact Simultaneous Activation
Synergy
TV
Digital
TV and Digital Media are most effective when executed together. There is about a +66%
synergy or additional lift due to simultaneous activation.
8

9. Synergistic
Effects
Copy
Quality
Effects
Social
Media
Effect
Long Term
Effects
Innovations: Multi-dimensional Media
Measurement
Bottom Line Analytics is at the
forefront of innovations in media
measurement. We have developed
ways to measure media synergies,
copy quality, long term ad effects and
social media engagement.
9

10. Measure long term ad effects
Most advertising creates an initial short term lift in sales and a prolonged long term
impact. This is generated through repeat purchase and customer loyalty.
Long Term Effect
10

11. Many MM Vendors do not know how to measure Long-Term ad
effects and counsel clients to simply multiply short-term media
by 2. This is wrong! For this brand, where 90% of sales were
repeat buys, Long-Term ad effects were 15x
66.05%
7.29%
1.82%
0.50%
0.21%
0.02%
1.45%
22.67%
33.95%
Incremental Contributions to Total Neutrogena Sales
Baseline Dist.@Min Mdsg.Ftr Mdsg.Displ
Mdsg.F + D ST Adv.Spend ST Adv.Creative Long-Term Adv 11

12. The financials of advertising dramatically
change from -$100K to 2 .7 million dollars!
($500)
$0
$500
$1,000
$1,500
$2,000
$2,500
$3,000
Short-Term Effect Only With Long-Term Effects
($100)
$2,720
HBA Brand Annual Ad Net Returns
* Assumes 20% EBITDA/Sales Margin
12

13. Media copy quality measurement
0
50
100
150
200
250
300
Radio Ad 1 TV Ad 2 Digital Ad 3 TV Ad 4
GRPs/Spend
Creative
13
Media content and copy quality can be separated and measured. This has
implications for design, content and message mix. 60 to 80 percent of short-
term advertising is message or creative and not spend/GRPs
Note: we can apply this technique to digital media also.

14. In the ideal world, spend on an ad-by-ad basis would align
perfectly with ad sales lift, but this is not the case. The
true situation is most often highly inefficient and
wasteful!
Correl.=0.298
0
20
40
60
80
100
120
140
160
3 5 7 9 11 13 15
SPENDPERADSQRT
Ad SALES LIFT SQRT
12

15. However, when we use copy test scores from ABX, it has
proven to align and correlate extremely well with individual
ad sales lifts. This provides a great resource for allocating
marketing funds across individual ads!
Correl.= 0.743
4
5
6
7
8
9
10
11
12
13
3 5 7 9 11 13 15
ABXINDEXCREATIVESCORESQRT
AD SALES LIFT SQRT
Using the ABX metric to allocate media budget by ad would
have generated + $21MM (2%) in incremental revenue for
Neutrogena
13

16. Effective Measurement of Digital Media
Annual Marketing Contributions
86.2%
0.0%
2.9%
0.7%
0.9%
1.9%
1.2%
1.2%
3.0%
0.2%
1.8%
13.8%
Baseline
Display Premium
Display Network
Paid Search
SEO
Branded TCP TV GRPs
Sponsorship (ITV Weather )
Cinema Ad
Radio
Press
GDP Effect
1.4 million in marketing spend generated almost 13.4 million pounds in revenue
sales. Total media accounts for about 12% of total sales. Radio, Digital Display and
TV were the largest drivers of car sales.
16

17. Addressing the issue or “multi-touch”
marketing attribution bias
• A common and well-known criticism of current MM models is their failure to
accurately cover and reflect the influence of different channels, especially digital
ones, that reflect the customer journey towards sales conversions. Frequently,
what we find is a bias that favors the specific channel closest to the sales
conversion, attributing most of the impact to that single last-touch point.
• MTA or multi-touch attribution covers the challenge of attributing accurate impact
of our marketing and advertising efforts across multiple devices (desktop, laptop,
mobile, TV) and/or channels.
• On the next slide, we illustrate how different model methodologies generate
results from the same data case study.
• Our findings reveal that common “single equation” econometrics yields extreme
results, assigning near full credit to one single channel.
• By using more advanced and multi-equation econometric methods, we are able to
develop models which simulate a path-solution rather then point-in time
responses. These more advanced multi-equation models allow for each
marketing channel to assume its more accurate impact based on the true
consumer purchase path. 17

18. Percent Contributions Single Eqtn. OLS 2SLS SUR Nested NNet
Digital Website Page Views [lag 3] 4.27 0.83 0.89 0.56
Display Ads 3.44 3.19 0.88
Digital.Pd.Search 0.51
Mass.TV 0.44 0.44 0.44 0.43
Mass.Print 0.09 0.16 0.17
Trend (4.05) (4.05) (4.06) (0.85)
Final LongLTVariable .KalmanFilter 28.40 28.40 28.34 4.49
Base 70.94 70.85 71.04 76.91
Total 100.00 100.00 100.00 100.00
Synergy 16.90
6.0 5.9 6.1 1.8
Conventional single –equation models are
biased with respect to multi-touch attribution
18
(5.00)
(4.00)
(3.00)
(2.00)
(1.00)
-
1.00
2.00
3.00
4.00
5.00
6.00
Single Eqtn. OLS 2SLS SUR Nested NNet
Model Percent Contributions
Trend
Mass.Print
Mass.TV
Digital.Pd.Search
Display Ads
Digital Website Page Views [lag 3]
Multi-Equation approaches are more balanced and overcome MT Attribution Bias
Single equation solutions often biased in
favor of activity closest to sales conversion

19. Single Equation Regression Model
Sales
Web
Page
View
TV
Print
BaseTrend
Lag 3
1) Sales is attributed to Web Page
Views, TV, Base and Trend
2) Numerous variables
are not significant +
multicollinear
19
Long
Term
Media
3) Very high attribution
on 1 variable (last
touch?)
Competitor
Ads
Display
Ads
Paid
Search
19
The most common approach to MM modeling is single-equation models, which have a high
likelihood of generating biased attribution due to “last-touch attribution bias”.

20. Two Stage Least-Squares Model
Sales
Web
Page
View
TV
Print
BaseTrend
Lag 3 Lag 1
1) Uses two equations, also referred to
as instrumental variable approach
2) Sales and Web Page Views have a
reciprocal relationship which is
lagged
3) Display ads are indirectly
contributing to sales via Web Page
Views
4) It appears that competitor ads are
also driving some traffic to own
Home Page.
5) Direct & Indirect Effects
Paid Search not
influencing sales or
web page views.
20
Paid
Search
Display
Ads
Competitor
Ads
Long
Term
Media
20Multi-equation models are better suited for discovering the nature and direction of complex
indirect or reciprocal relationships and interactions within the data models

21. Seemingly Unrelated Regressions
(SUR)
21
Display
Ads
Sales
Web
Page
View
TV
Long
Term
Media
Print
Trend
This variables is not
significantPaid
Search
Lag 1Lag 3
1) Like SEM for time series
modeling. SUR is a true Multi-
equation system
2) When errors are correlated,
solution is a path rather than
discrete data variables. This path
can be assumed to be the
attribution path.
3) But when the regression errors really
are unrelated, then we are just
generating single eqtn. OLS results
Competitor
Ads
21

22. Nested Neural-Network Model
22
Sales
TV Print
Display
Ads
Paid
Search
Long
Term
Media
All
MarCom
Competitor
Ads
Web
Page
Views
Trend
1) All MarCom Variables pooled
into meta-variable and
dynamically weighted
2) Good for discovery of non-
linearity, interaction and synergistic
effects without a priori knowledge
3) Have undeserved reputation for
being black-box & can be trained to
be stupid.

23. Measuring the “Voice-of-the-
Customer” via Social Media
A breakthrough in measuring social media engagement
23

24. Can social media be measured?
1 The Growing Importance of Word of Mouth, www.boundless.com
24
 Social Media really isn’t Media as we know it. It doesn’t have “inventory”
and it’s not meant to deliver “ads” like traditional “media”
 Marketing was once seen as a one way relationship, with firms
broadcasting their offerings and value proposition.
• Now Marketing is seen more as a conversation between marketers and customers.1
• Social media is a key and critical channel for this two-way communication
 Current social media metrics are expressed in terms of “sentiment”
• Positive and negative commentaries about brands
• These metrics do not seem to explain or predict purchase behavior
 Many have given up and say social media can not be measured

25.  If we remember that social media is a form of word-of-mouth, then words
matter!
• The semantics, linguistics and context of the conversation matters
 Our Social Media analysis is based on Stance-Shift Analysis
• Uses the Social Media conversations about your Brand as input
• Apply linguistic principles of sentiment and tonality
• Results in an engagement score that is a translation of a customer’s “personal” and “emotional”
relationship with brands, as revealed through language & semantics….Social Engagement Index (SEI)
• Academically published, peer reviewed & validated.2
 Stance-Shift Analysis translates the consumer’s qualitative emotions into
quantitative metrics.
Our approach to measuring Social Media
2 Stance Analysis: social cues and attitudes in online interaction, Mason, P , Davis B, In E-Marketing Vol. II . 2005.
25

26. Developing the Social Engagement Index (SEI)
Net Positive SEI Index
1. Mine all brand related social media
reviews and commentary.
2. Parse into positive & negative
review groups
3. Apply Social Engagement Index
algorithm to “score” reviews
4. Time code by period and aggregate metrics
Positive
Reviews
Negative
Reviews
Positive
Scores
Negative
Scores
LOW MEDIUM HIGH
HIGH 0 5 7
MEDIUM -5 0 5
LOW -7 -5 0
Emotional Effect
Personalisation
26

27. SOCIAL
ENGAGEMENT
INDEX (SEI)
Conversations are scored on personal
and emotional content
“I HAD A DIET COKE FOR LUNCH TODAY”
“THE WARM DIET COKE WAS RATHER BLAND”
27
“I REALLY LOVE MY COKE WITH PIZZA”
“I LIKE THE TASTE OF SPRITE WITH LEMON”
“MY COKE HAS LOST ITS FIZZ AND TASTES AWFUL”

28. SEI shows superior correlationsto brand sales compared
with other SocialSentiment Metrics
82.9%
14.8%
9.9%
7.7%
5.9%
2.8%
-3.2%
-20% 0% 20% 40% 60% 80% 100%
SOCIAL ENGAGEMENT INDEX POS/NEG RATIO
METRIC 5 POS/NEG RATIO
METRIC 1 POS/NEG RATIO
METRIC 4 POS/NEG RATIO
METRIC 6 POS/NEG RATIO
METRIC 2 POS/NEG RATIO
METRIC 3 POS/NEG RATIO
Comparison of correlation to sales for the SEI versus the six leading sentiment metrics
28

29. The correlation* to sales over time shows the SEI has Predictive Power
29
ACID TEST: SEIsm has proven linkage with brand sales
Correlation = 86.4%
Correlation = 84%
Correlation = 81.1%
Correlation = 83%
Correlation = 83%
* Lead lag analysis has confirmed that causation is only one way – the SEI to a large degree is able to drive hard commercial metrics.

30. Applications of the SEISM
Packaged inside a media mix model, the SEI
acts as the key indicator for social media
‘word of mouth’.
We are able to determine the return on
investment for social media and provide
steer around the most effective channels and
spend.
SEI to help uncover
market insights
The SEI is also the primary tool used to
understand the degree of brand engagement
as it transpires through the use of language.
• Understand drivers to positive engagement.
• Measure the efficacy of individual campaigns.
• Develop content strategy that has cut through.
• Enhance the execution of sporting events.
• Assess brand perception in a competitive sense.
• Understand consumer discourse and manage crises.
SEI to measure social
media ROI
30

31. 31
SEI to measure social media ROI
 We find that conventional advertising has both a “direct” and “indirect” impact on sales due to
its influence on social media conversations and the SEI.
 The large contribution from the SEI support the notion that this is a “word-of-mouth” effect
67%
8%
3%
2% 2%
10%
5%
11%
20%
Marketing Contributions
Base Sales Direct Alpha Brand Mass Media Direct Alpha Brand Digital Media
Direct Social Media Social Media on SEI Mass Media on SEI
Digital Media on SEI SEI Base
Net driven by media
SEI
Engagement
Sub-model

32. 32
The impact of Social Media sentiment
 A key insight we uncovered across clients is the difference between “positive” and “negative”
brand conversations
 Negative-toned conversation have a significantly greater net impact on brand sales
+4.4%
+16.5%
0%
5%
10%
15%
20%
Positive Sentiment Negative Sentiment
The absolute impact from positive &
negative consumer reviews
Marketers need to develop strategies and tactics to immediately mitigate “Negative News”
and prevent them from going Viral.

33. Much like other marketing and media metrics, we can deconstruct the different elements of
the SEI metric into the channels driving social engagement and brand sales.
Source: Nielsen BuzzMetrics data as of November 27, 2011
Social channels driving consumer
engagement and sales
33

34. Most Important Drivers to
Positive SEI.
Using this insight, the
client developed a ‘bring
a friend, and get one
coffee free’ to drive store
level sales.
Positive SEI
3.93 = 100
Place2HangOut
>5.46= 211
9.1%
Place2HangOut
<5.46 = 83
91.9%
ToMeetPeople>
9.43 = 325
2.6%
ToMeetPeople<
9.63 = 188
6.5%
Atmosphere
>14.0 = 466
0.6%
Atmosphere
<14.0 = 288
1.9%
To Meet People
>5.4 = 229
3.8%
To Meet People
<5.4 = 85
85.5%
Beverage A
>6.4 = 271
7.7%
Beverage A
<6.4 = 74
77.8%
Place2HangOut
>3.6 = 126
5.9%
Place2HangOut
<3.6 = 76
71.9%
Beverage B
>5.2 = 211.1
1.6%
Beverage B
<5.2 = 67
70.3%
Note: Separate analysis - Classification & Regression Trees (CART)
The tree starts with an average SEI score of 100; and each level indicates a higher or lower SEI based on
an SEI score for a topic. The percent represents the percent of the sample in each segment.
Develop In-Market strategies based on
“Why” consumers use your brand
34

35. Alpha_P1
Beta_P1
Note: Separate analysis - Adapted Statistical Correspondence Analysis
Example: Global Coffee Chain
Bubble size represents the buzz/volume of chatter (SEI Conversational Clusters)
Alpha_P2
Beta_P2
Gamma_P1
Gamma_P2
Net Chatter around value
and price
Net Chatter around coolness, funky,
style, Décor
Net Chatter around taste and
product quality
Net Chatter around in-store
customer experience
Delta_P2
Delta_P1
Good value
Coffee Price
Food prices
Staying in
Seating/chairs
Toilets
Richness
Latte
Amazing taste
Like no other
Cool brand
Funky
Stylish Artwork/Decor
Visualize social media brand conversations
35

36. Introducing…
 Bottom-Line Analytics & GAP is a full service consulting group focusing on
marketing effectiveness and brand performance analytics.
 Our modeling experts have a total of over 150 years of direct experience
with marketing mix modeling with direct experience in over 40 countries
 We are dedicated to the principles of innovation, excellence and
uncompromising customer service.
 Everything we do is geared towards improving commercial performance.
36
&

37. Play out marketing What-if scenarios
An interactive dashboard allows you to simulate different marketing mix/spend
scenarios and assess the resultant impact on sales and profitability.
1. Set marketing
budgets.
2. Set your
spend levels
across media
channels
3. Assess the
resultant
impact on sales
& profit
37

38. Continuous model updates
allow for real-time simulation
and planning
An interactive dashboard allows you to
simulate different marketing mix/spend
scenarios and assess the resultant impact
on sales and profitability.

39. Why
Impartial and
Independent
Full Service
Analytics
Capability
VOC
Measurement
With Social
Media
Marketing Mix Modelling 3.0
Ad Copy ROI Measurement
Multi-Touch Attribution Models
Marketing Synergies
Long-Term Ad Effect
Pricing Optimisation
Radial Landscape Mapping
Key Drivers Analysis
Demand Forecasting
Customer Satisfaction Modelling
Performance Analytics Dashboards
Segmentation Analysis
Marketing Decision Support Tools
Our proprietary approach
to social media
measurement is unrivalled.
Objective approach to
media measurement.
39
&

40. Full Service
Analytics
Capability
Social Media ROI
Marketing Mix Modelling
Pricing Optimization
Radial Landscape Mapping
Key Drivers Analysis
Demand Forecasting
Customer Satisfaction Modelling
Digital Performance Analytics Dashboards
Segmentation Analysis
Decision-Support Systems
40
BLA is a trusted advisor to a wide array of clients
We believe in the continuous innovative application
of analytics to advance customer centric decision
making for improved business performance.

41. It’s all About Results
Company Results
Coca-Cola
Brought marketing ROI modeling to company for first time in 1996. In first year developed models for
Coca-Cola, Coke Light, Fanta and Sprite in 12 Countries. Year two sales gains over prior year exceeded
$300 million.
Starbucks
Developed measure of customer-brand experience using social media. Discovered that Starbucks main
strength lies in its in-store experience. Successfully developed brand positioning for Frappucino and
Via Coffee. Sales growth improved from +7 to +11 percent
McDonald's
Identified significant upside growth opportunity to drive higher restaurant sales by investing
significantly more in "dollar-value meals" one year after launch in 2005. Per recommendation, major
& higher marketing investment in dollar value meals made McD's the growth leader in its competitive
segment for 2 years thereafter.
L'0real
Developed models which measured the ROI across 12 different "Celebrity Spokespersons" in L'Oreal
Commercials. Recommended reducing number from 12 to 5 Celebrities, leading to growth
improvement from +3 to +5%.
Hyatt Hotels
Developed SEI to quantify measure of "customer satisfaction" derived from measures of Trip Advisor
hotel reviews across 300 different properties. This lead to a 5% improvement in customer satisfaction
in subsequent year and a +6% growth in total bookings
AT&T
Identified and quantified impact from the launch of iPhone. By identifying which ad copy messages
were most effective, AT&T managed to increase it's wireless telecom market share from 28 to 30%.
Johnson and Johnson
Developed analytic system for measuring and evaluating ad copy for Splenda brand. Enabled brand to
reduce ad production from 8 to 4 commercial executions, saving $6 million
41

42. Global Analytics Partners
37
Global Analytics Partners is a consortium of advanced analytics, marketing technology
& strategy firms bringing together extensive global experience in all phases of marketing
science, decision support & advanced analytics. Collectively, we have the scale and the
tools to assume any challenge & have over 150 years of direct experience covering over 40
International markets

43. Bangalore, IN Office:
No. 141, 2nd Cross, 2nd
Main,Domlur, 2nd Stage, Bangalore
560071Phone: +91 80 40917572,
+91 80 40916116
info@therainman.in
Contact Us US Office:
Suite 100, 1780 Chadds Lake Dr, NE
Marietta, Georgia, 30068-1608
Atlanta, USA
mjw@bottomlineanalytics.com