Upload
doanque
View
223
Download
2
Embed Size (px)
Citation preview
Google Confidential and Proprietary 1 Google Confidential and Proprietary 1
Online & Offline Media ROI Measurement and Media Mix Optimization
January 2015
Food brand
Google Confidential and Proprietary 2 Google Confidential and Proprietary 2
Contents
Introduction: Marketing levers and modeling process
Project objectives and methodology.
Sales models Key drivers of sales.
Online models Key drivers of Google Paid Search, CTR and queries.
Media mix optimization Optimization of advertisings contribution to sales.
1
2
3
4
Google Confidential and Proprietary 3 Google Confidential and Proprietary 3
Contents
Introduction: Marketing levers and modeling process
Project objectives and methodology.
Sales models Key drivers of sales.
Online models Key drivers of Google Paid Search, CTR and queries.
Media mix optimization Optimization of advertisings contribution to sales.
1
2
3
4
Google Confidential and Proprietary 4 Google Confidential and Proprietary 4
Project objectives Unilever wanted to identify the main marketing levers and measure the impact of online and offline media on sales of two products of a food brand.
Their key objective was to optimize media investment in order to maximize sales.
1
Google Confidential and Proprietary 5 Google Confidential and Proprietary 5
Sources of information The analysis was carried out by Conento, with the support of BIG, between July and December 2014.
1 *Note: We only considered distributors data and the presence or absence of promotions, because we did not have any more data available.
VARIABLES BY CATEGORY PERIODICITY
Sales by product Monthly (sell-out) - Since 2011 Weekly (sell-in): Since Jul 2012*
Web traffic: Total and MicroSites Weekly - Since July 2013
Media investment (online and offline) Offline: Monthly – Since 2011 Online: Aggregated data since
March 2013
Competitors’ Media investment Monthly – Since 2011
Distribution, Price, Promotions* by product Monthly – Since 2011
Queries Quarterly – Since March 2011 Monthly – Since March 2013
Paid Search: Impressions / Clicks / Position / CTR by word group Daily – Since March 2013
Display: Impressions / Clicks / CTR Daily – Since December 2012
Macro-economic Variables: Unemployment Rate / consumer confidence index / GDP evolution Yearly – Long historic
Socio-demographic data (population, Internet penetration, etc) Yearly – Long historic
Temperature, Holidays etc. Monthly – Since 2011
Google Confidential and Proprietary 6 Google Confidential and Proprietary 6
Methodology
– We tested both linear and log-linear econometric models in order to find the best methodology. Finally, we chose linear models because, although the results of both analysis were quite similar, the interpretation of the coefficients of the variables is far more complex in the case of log-linear models.
– Using the linear methodology, we built the sales, Google Paid Search clicks and the brand queries models. In the case of the latter, we also used Loess time series decomposition in order to disaggregate the series into their trend, seasonal and remainder components.
– Using the learnings from the models and an optimization algorithm, we were able to recommend optimal levels of investment for TV, Paid Search and Display, in order to maximize their contribution to sales.
1
Google Confidential and Proprietary 7 Google Confidential and Proprietary 7
Econometric models help to quantify the ROI of offline and online media, measuring the impact of the main marketing and business levers.
1
SALES
OFFLINE MEDIA ADVERTISING
OTHER (SEASONALITIES,
ETC) DISTRIBUTION PRICE PROMOTION
ONLINE MEDIA ADVERTISING
* In this study data of awareness are not available .
Google Confidential and Proprietary 8 Google Confidential and Proprietary 8
We identified the main communication drivers for each step of the consumer purchase process.
1
CLICKS SEM SALES
3 4
BRAND QUERIES
2 MEDIA
ADVERTISING SPEND
MEDIA
Television
Magazines
Internet
1
BRAND
Google Confidential and Proprietary 9 Google Confidential and Proprietary 9
Contents
Introduction: Marketing levers and modeling process
Project objectives and methodology.
Sales models Key drivers of sales.
Online models Key drivers of Google Paid Search, CTR and queries.
Media mix optimization Optimization of advertisings contribution to sales.
1
2
3
4
Google Confidential and Proprietary 10 Google Confidential and Proprietary 10 2
The final sales models are very robust, with levels of fit to actual sales over 94%.
R2: 99.3%
Sale
s (‘0
00 K
g)
R2: 98.8%
Sale
s (‘0
00 K
g)
Product A
Product A references
Product B
Reference 1 Reference 2 Reference 3 Reference 4
98.3% 94.6% 99.5% 99.6%
R2 Product B references
Reference 1 Reference 2 Reference 3
94.7% 95.9% 94.5%
R2
0
1
2
3
4
5
6
Mar
-13
Apr-
13
May
-13
Jun-
13
Jul-
13
Aug-
13
Sep-
13
Oct
-13
Nov
-13
Dec
-13
Jan-
14
Feb-
14
Mar
-14
Apr-
14
Sales JellyFitProduct 1 salesFitProduct A salesFit
0
2
4
6
8
10
12
14
Oct
-11
Dec
-11
Feb-
12
Apr-
12
Jun-
12
Aug-
12
Oct
-12
Dec
-12
Feb-
13
Apr-
13
Jun-
13
Aug-
13
Oct
-13
Dec
-13
Feb-
14
Apr-
14
Sales Double PouchFitProduct 2 salesFitProduct 2 salesFitProduct B salesFit
Google Confidential and Proprietary 11 Google Confidential and Proprietary 11
-‐40
-‐30
-‐20
-‐10
0
10
20
30
40
Oct-‐11 Dec-‐11 Feb-‐12 Apr-‐12 Jun-‐12 Aug-‐12 Oct-‐12 Dec-‐12 Feb-‐13 Apr-‐13 Jun-‐13 Aug-‐13 Oct-‐13 Dec-‐13 Feb-‐14 Apr-‐14
-‐6
-‐4
-‐2
0
2
4
6
8
Mar-‐13 Apr-‐13 May-‐13 Jun-‐13 Jul-‐13 Aug-‐13 Sep-‐13 Oct-‐13 Nov-‐13 Dec-‐13 Jan-‐14 Feb-‐14 Mar-‐14 Apr-‐14
2
The models are capable of isolating the effect of each variable, in order to achieve the optimization of media investment.
Sale
s (‘0
00 K
g)
Weighted distribution creates the baseline of
sales.
Media investment contributes to sales
Sales are price sensitive: The higher the price per unit, the more sales decrease
Promotions make sales grow
Product A sales - Key drivers
Sale
s (‘0
00 K
g)
Growth levels: Possibly advertising + promotions extra
effect
Media investment contribution
Possibly advertising + promotions extra effect
Note: Media investment explains 4.21% of real sales.
Contribution of variables
Note: Media investment explains 8.20% of real sales.
Product B sales - Key drivers
60.9%24.0%
4.12%5.3%
4.8% 0.8%
Weighted SellingDistribution
Price
Media Investment
Promotions
Rest of positivevariables
Rest of negativevariables
55.1%41.9%
0.66%0.4% 1.6% 0.3%
Google Confidential and Proprietary 12 Google Confidential and Proprietary 12
Contents
Introduction: Marketing levers and modeling process
Project objectives and methodology.
Sales models Key drivers of sales.
Online models Key drivers of Google Paid Search, CTR and queries.
Media mix optimization Optimization of advertisings contribution to sales.
1
2
3
4
Google Confidential and Proprietary 13 Google Confidential and Proprietary 13
-‐2,000
-‐1,000
0
1,000
2,000
3,000
4,000
5,000
6,000
31/03/13
14/04/13
28/04/13
12/05/13
26/05/13
09/06/13
23/06/13
07/07/13
21/07/13
04/08/13
18/08/13
01/09/13
15/09/13
29/09/13
13/10/13
27/10/13
10/11/13
24/11/13
08/12/13
22/12/13
05/01/14
19/01/14
02/02/14
16/02/14
02/03/14
16/03/14
30/03/14
13/04/14
27/04/14
11/05/14
25/05/14
08/06/14
22/06/14
-‐10,000
-‐5,000
0
5,000
10,000
15,000
31/03/13
14/04/13
28/04/13
12/05/13
26/05/13
09/06/13
23/06/13
07/07/13
21/07/13
04/08/13
18/08/13
01/09/13
15/09/13
29/09/13
13/10/13
27/10/13
10/11/13
24/11/13
08/12/13
22/12/13
05/01/14
19/01/14
02/02/14
16/02/14
02/03/14
16/03/14
30/03/14
13/04/14
27/04/14
11/05/14
25/05/14
08/06/14
22/06/14
3
Impressions and Average position are the main drivers of Google Paid Search clicks models.
Contribution of variables
More than 90% of Product A clicks are explained by 3 main variables: Impressions, Average position and Brand queries .
Goo
gle
Paid
Sea
rch
clic
ks
Goo
gle
Paid
Sea
rch
clic
ks
More than 80% of Product B clicks are explained by 3 main variables: Impressions, Average position and Brand queries.
Brand Queries have a positive effect on clicks.
Brand Queries have a positive effect on clicks.
As average position rises, number of clicks decrease
As average position rises, number of clicks decreases
Product A clicks - Key drivers
Product B clicks - Key drivers
The growing trend may be due to the internet penetration increase over time in Russia
See Search explanation
55.0%
20.0%
17.0%
3.2% 4.8%
66.1%12.4%
6.6%
12.4%
2.5%
Impressions
AveragePosition
Queries Knorr
Trend
Rest ofvariables
Google Confidential and Proprietary 14 Google Confidential and Proprietary 14
0%
5%
10%
15%
-‐5,000 15,000 35,000 55,000 75,000 95,000
Shawarma Ryba Burritos StroganoffCaesar Schnitzel Zhulien
3
Synergy between online and offline media. In general, products with higher offline investment also have higher CTR.
Position increases clicks. Increasing / decreasing average position increases / decreases clicks in a non linear way.
Average CTR vs. Offline Investment
Average
position
% of clicks gained /
lost
1 +35.2%
1.3 * -
2 -29.9%
3 -45.3%
Average
position
% of clicks gained /
lost
1 +58.4%
1.4 * -
2 -23.5%
3 -41.2%
*Note: Paid Search average position.
Product A Product B
Offline investment
CTR
In this case average position for Product A was 1.3. If average position had been constantly 1, total clicks should increase by 35.2%. If average position had been 2 (or constantly 2), total clicks should decrease by -29.9%.
CTR
0%
5%
10%
15%
20%
25%
1 1.5 2 2.5
Burritos Stroganoff Zhulien Ryba Caesar Shawarma SchnitzelAverage Position
CTR vs. Average Position
Google Confidential and Proprietary 15 Google Confidential and Proprietary 15
-‐1,000
-‐500
0
500
1,000
1,500
2,000
2,500
3,000
Jan-‐12
Feb-‐12
Mar-‐12
Apr-‐12
May-‐12
Jun-‐12
Jul-‐1
2Au
g-‐12
Sep-‐12
Oct-‐12
Nov-‐12
Dec-‐12
Jan-‐13
Feb-‐13
Mar-‐13
Apr-‐13
May-‐13
Jun-‐13
Jul-‐1
3Au
g-‐13
Sep-‐13
Oct-‐13
Nov-‐13
Dec-‐13
Jan-‐14
Feb-‐14
Mar-‐14
Apr-‐14
Queries Base level GRPs Total Knorr Extra GRPs Advertising Rest of positive variables Decrease queries levelBrand
80.1%
10.3%
9.6%
3
Offline advertising of the brand helps increase Queries (brand online interest), especially when new products are launched.
Advertising helps queries to rise
Goo
gle
Bran
d qu
erie
s
Trend of queries
New references launch period
New references launch period
In some periods, queries are lower than expected.
Key drivers Brand Queries
Contribution of variables
10.3% of Brand queries come from offline investment.
See Queries explanation
Google Confidential and Proprietary 16 Google Confidential and Proprietary 16
Contents
Introduction: Marketing levers and modeling process
Project objectives and methodology.
Sales models Key drivers of sales.
Online models Key drivers of Google Paid Search, CTR and queries.
Media mix optimization Optimization of advertisings contribution to sales.
1
2
3
4
Google Confidential and Proprietary 17 Google Confidential and Proprietary 17
Based on their contribution to sales, it is possible to calculate the ROI for each media:
4
% ROI
Media contribution (‘000kg) Product A
Product A
Product B
Product B
TV contribution to sales is 3.63 thousand kilos in its total period (Mar 13 – Apr 14)
TV contribution to sales is 9.33 thousand kilos in its total period (Oct 11 – Apr 14)
For each million Rubles invested on TV, Product A sales rise 0.76% over average.
3.63
0.02
0.14
TV
Search
Display
9.33
0.23
0.49
TV
Search
Display
0.76%
0.57%
0.59%
TV
Search
Display
1.04%
1.69%
0.86%
TV
Search
Display
Google Confidential and Proprietary 18 Google Confidential and Proprietary 18 4
TV is the media with the highest contribution to sales. However, in the case of lower investments, Search and Display have bigger contributions.
Maximum investment
Product A Product B
Note: As online investment is low and the market is growing, we estimated the sales contribution of Search and Display for higher investments using Conento’s benchmark for these media (see dotted lines)
For small budgets, online investment has a bigger contribution to sales than TV
0
0.5
1
1.5
2
2.5
0 10,000 20,000 30,000
Con
tribu
tion
('000
Kg)
Monthly Investment ('000 Rb)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 10,000 20,000 30,000
Con
tribu
tion
('000
Kg)
Monthly Investment ('000 Rb)
Google Confidential and Proprietary 19 Google Confidential and Proprietary 19
0%
20%
40%
60%
80%
100%
11 12 13 14 15 16 17 18
% In
vestmen
t by med
ia
Total investment (M RB)% Search % Display % TV
SC1 SC2 SC3 SC4 SC5 SC6 SC7
0%
20%
40%
60%
80%
100%
17 18 19 20 21 22 23 24
% In
vestmen
t by med
ia
Total investment (M RB)% Search % Display % TV
SC1 SC2 SC3 SC4 SC5 SC6 SC7
4
Optimizing media investment, Unilever should increase advertisings contribution to Product A sales by 2.3%, and to Product B by 8.1%.
See optimization methodology
89.3%
1.7%9.1%
93.8%
1.0% 5.2%
93.6%
0.4% 5.9%
96.4%
0.3% 3.3%
Product A Product A monthly average investment
Optimized investment distribution by scenario Optimized investment
+2.3% Increased contribution
due to advertising
TV: -2.8 pp Search: +0.1 pp Display: +2.6 pp
TV
Search Display Search Display
TV
Product B Product B monthly average investment
Optimized investment distribution by scenario Optimized investment
+8.1% Increased contribution
due to advertising
TV: -4.5 pp Search: +0.7 pp Display: +3.9 pp
TV
Search Display Search Display
TV
Product A average investment
Product B average investment
Optimized vs. real investment
Optimized vs. real investment
Google Confidential and Proprietary 20 Google Confidential and Proprietary 20
0%
20%
40%
60%
80%
100%
28 30 32 34 36 38 40 42
% In
vestmen
t by med
ia
Total investment (M RB)% Search Jelly % Search DP % Display DP + J% TV Jelly % TV DP
SC1 SC2 SC3 SC4 SC5 SC6 SC7
% Search Prod. A
% TV Prod. A
% Search Prod. B
% TV Prod. B
% Display Prod. A + B
If we optimized media investment between the 2 products, advertisings contribution to total sales should rise by 23.7%, as Product B sales response is higher.
4
49.3%46.2%
0.7%0.3% 3.6%
37.4%
60.0%
0.4%0.2% 2.1%
+23.7% Increased
contribution due to advertising
Product A + Product B average investment
Product A + Product B
TV Prod. A: -13.8 pp TV Prod. B: +11.9 pp Search Prod. A: +0.1 pp Search Prod. B: +0.3 pp Display: +1.5 pp
Product A + Product B monthly average
investment Optimized
investment Optimized investment distribution by
scenario
Search Prod. B
Display Search Prod. A
TV Prod. A
TV Prod. B
Search Prod. B
Display Search Prod. A
TV Prod. A TV Prod. B
Optimized vs. real
investment
Google Confidential and Proprietary 21 Google Confidential and Proprietary 21
• Main drivers of sales are Structural Variables The most important contributions to Product A and Product B sales are generated by structural variables (weighted distribution, price, and promotions) though advertising has a clear extra effect on sales. The total impact of media on short term sales is 8.20% for Product A and 4.21% for Product B.
• Both TV advertising and a better position in Paid Search increase clicks Products that have TV investment, have a higher CTR than those that don’t.
• TV Advertising has a positive impact on Brand Queries 10.3% of Brand queries are due to TV advertising. Queries also have a significant impact on clicks: they account for 17.0% of Product A clicks and 6.6% of Product B clicks.
• Increasing Online investment will increase sales Currently TV is the media that has the highest contribution to sales due to its higher share of investment. There is an opportunity to increase sales through higher investment in Search and Display.
• There are opportunities to optimize media investment in Product A and Product B and increase sales If we maintain the average levels of investment in Product A and Product B and optimize media mix (decreasing TV and increasing search and display investment), then advertisings contribution to sales could improve by 2.3% and 8.1% respectively (and 23.7% with a global optimization of investment between the 2 products).
Summary of Main Learnings.
Google Confidential and Proprietary 22 Google Confidential and Proprietary 22
Thank you
Google Confidential and Proprietary 23 Google Confidential and Proprietary 23 3
A web user: Enters in www.google.ru In the Google Query String makes a search: · Brand · Non-brand Clicks the impression related to the brand (in the Paid Search Area - SEM) and goes to its webpage.
1
2
1
2
3
We explain these clicks
3
Go back
Brand food
Google Confidential and Proprietary 24 Google Confidential and Proprietary 24 3
Brand Queries (related to the brand)
keywords:
Brand name Misspelled Brand name
Brand web page Brand + other words
…
Queries
Model
• When working with Brand queries we added up all keywords with representative volume regarding the brand, as showed above.
• The main difficulty was that we had quarterly data until March 2013. For the rest of the period we had monthly data.
How do we consider the brand queries?
Go back
Google Confidential and Proprietary 25 Google Confidential and Proprietary 25 4
Total monthly Budget (in RB)
Monthly optimal investment by
media (in RB)
Effectiveness response curves
Input Output Optimization
Once we have defined the data set, parameters, variables and
initial restrictions in the optimization software, we set
the budget to optimize.
With a non linear algorithm, we try to get the highest
contribution to sales using the calculated curves.
We obtain the budget share by media that guarantees the
highest response for sales for the given budget.
How do we do the optimization to allocate spending by media?
Go back