Does your organization get stuck with attribution and marketing mix modeling? This presentation shows you how companies use both consumer-level and aggregate data to optimize their media allocation with double-digit performance gains. The cases include an online retailer and L'Occitane's multi-channel and multinational optimization of customer segment targeting, which got the 2018 MSI/Informs Practice Prize Award. I also offer a conceptual and modeling framework to your company
2. Agenda
• The New Consumer Journey
• Current State of Attribution Research
• How to Optimally Allocate Media Budgets
• Case 1: Individual Consumers for L’Occitane
• Case 2: Aggregate Data on 11 Ad Forms for Online Retailer
• Implications: Replace Common Attribution Myths
4. Past Metaphor: The linear purchase funnel
Purchase LoyaltyConsiderationAwareness
5. Initial
Consideration
Set
Moment of
Decision
Loyalty Loop
Post Purchase Experience
Ongoing process
Active Evaluation
Information gathering, shopping
McKinsey 2009
The Consumer Decision
Journey
Has the Web made the funnel fat?
Purchase LoyaltyConsiderationAwareness
6. Implications for managers
“Fat Loop” means that consumers
1) Dynamically go back and forth between
sources
2) Hesitate to purchase, then revisit
Which means that marketing actions
1) Can influence consumers on many
occasions
2) Can help consumers gain confidence in
their decision making
7. What do we already know?
• Display ads are effective in early stages; search is effective
throughout the journey (Abhishek et al. 2012)
• Consumer finds choice in mid to late search, then revisits
chosen product (Bronnenberg et al 2016)
• Branded keywords are less effective for large sellers (eBay)
than non-branded, but very effective for small providers of
urgent products, e.g. office furniture (Pauwels et al. 2016)
8. What don’t we know (yet)?
• How responsive are different types of consumers to online
and offline ads ?
• Which online advertising forms are more effective, when
and where in the journey?
• How to (re)allocate ad budgets accordingly?
9. We Need a Sales Response Model
• How does your sales respond to marketing actions, controlling
for outside influences?
• Allows us to calculate ELASTICITY
o Elasticity = % sales lift for a 1% increase in a media’s budget
• Elasticity is the key input for optimal allocation
10. How to Allocate Budget: Ratio of Elasticities
• Analysis shows doubling on comparison ads lift sales by 15%
(elasticity = 0.15) while doubling ‘retargeting’ increases your
sales by only 5% (elasticity = 0.05)
• How should you divide your budget of $ 100,000 ?
10
• Sum up elasticities: 0.15 + .05 = .20
• Ratio of elasticity = 0.15/0.20 = 75%, 0.05/0.20 = 25%
• Result: Comparison spend allocated $ 75 K and
Retargeting spend allocated $ 25 K
11. • Natural ingredients cosmetics and well-being retailer
• €1.3 billion revenues, €168 million profits (2017)
• 8,500 employees in 90 countries with 3,037 stores
• Multichannel: offline and online sales channels
L’Occitane Case Study
11
12. Primary dataset
12
• 84,110 randomly selected customers from 6
countries
• Germany, Spain, France, Great Britain, Italy, USA
• Purchase history: offline sales, online sales, and
discounts
• 4 years of data, 2011-2014
• Marketing actions: direct mail and email
• 2 years of data, 2013-2014
13. Current Approach Follows Conventional Wisdom
13
Marketing Channel
Direct Mail
Email
Customer Segments
High Value
All Segments
$$$
$$$
14. Modeling Approach
Quantify
customer value
Create
customer segments
Evaluate responsiveness
to marketing
Assess
sales variation drivers
Predict
sales
Steps Methodology
RFMC framework
Cluster analysis
Hierarchical linear
model
Cross-Random
Effects model
Forecasting
accuracy
comparison
Description
Quantify the values of R-F-M-C for
each individual customer
Segment the customer base
according to customer value and
country
Evaluate responsiveness to marketing
actions at different aggregation levels:
time, customer value, and country
Assess the extent to which sales
variation can be explained by time,
customer value, and country
Compare forecasting accuracy
of our model to benchmarks
1
2
3
4
5
Period
Calibration
Calibration
Estimation
Estimation
Hold-out
Allocate
marketing resources
Relative elasticities Reallocate marketing actions within
each country keeping budget constant
6
Estimation
Design and implement a field experiment
7
Descriptive
Predictive
Prescriptive
16. • Segmentation: customer value level within each country
• K-means on standardized RFMC metrics
• 4 clusters (36k customers) + dormants (28k) and prospects (19k)
• Dissimilarity measure: Euclidean distance
• Starting points: 20%, 40%, 60% and 80% values of standardized
RFMC
16
Cluster analysis2
17. Customer segments description
17
Prospects Dormants
Non-recent
low value
Recent
low value
Medium
value
High
value
Total
Germany 22% 33% 8% 10% 24% 4% 10,000
Spain 16% 36% 12% 8% 25% 4% 10,000
France 26% 36% 8% 6% 22% 3% 14,111
Great Britain 10% 40% 10% 8% 26% 4% 20,000
Italy 23% 31% 12% 8% 25% 1% 10,000
USA 38% 26% 9% 6% 19% 2% 19,999
Recency (weeks ago) - - 79 9 33 12 38
Frequency (#) - - 1.22 1.34 1.99 8.76 2.15
Monetary value (€) - - 48.6 53.5 82.0 468.1 94.7
Clumpiness (#) - - 0.61 0.69 0.36 0.21 0.47
Note: weekly data. Individual customer RFMC values during calibration period.
2
18. Direct mail: Own-channel effects for
prospects across all countries
18
3
Offline
sales
All countries: Germany, Spain, France, Great Britain, Italy, USA
Elasticities
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
Prospects Dormants Non-recent low
value
Recent low
value
Medium value High value
19. Email: Affects across Channels and
Segments
3
Offline
sales
Online
sales
USAElasticities
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.00
0.10
0.20
0.30
0.40
0.50
Prospects Dormants Non-recent low
value
Recent low
value
Medium value High value
22. The different effectiveness of direct mail and email depending on
the customer type was surprising to us. Rethinking about this
finding, we have a deep and increasing interest in investing in direct
mail activities for customer acquisition and inactive customers.
In the Words of L’Occitane
22
“ “
Ms. Delphine Fournier
CRM Senior Manager, L’Occitane
23. Study: Expand to dozens of ad forms
• “The effectiveness of different forms of online advertising for
purchase conversion in a multiple-channel attribution
framework” International Journal of Research in Marketing
• Compare the long-term effectiveness of a wide range of online
advertising forms, over multiple product categories, and
controlling for offline advertisements.
24. Purpose of our research
• Which form of online advertising (e.g., email, display, search,
comparison) is the most effective?
o Effectiveness as measured by “revenue elasticity”
• When do these effects take hold and how long do these effects
last?
• Where in the conversion funnel are these effects the strongest?
o Bringing in new customers
o Increasing the conversion rate
o Higher average sale
25. Study Dimension: CIC vs. FIC
Firm Initiated
Contact (FIC)
• Companies “pushing” messages
to consumers.
• Used to stimulate prospective
customers.
• Increasingly unwanted.
• TV
• Radio
• Email
• Display
Customer-Initiated
Contact (CIC)
• Advertising triggered by
(prospective) customer action.
• Customer closer to purchase
decision.
• Higher response rates.
• Less intrusive.
• Search (organic & paid)
• Comparison
• Retargeting
• Referrals
26. Study Dimension: Content-Integrated vs. Content-
Separated CIC
CIC - Content-
Integrated
• Advertising that are an integral
part of a medium’s editorial
content.
• Search (organic)
• Price comparison
• Hobby sites
• Referral sites
CIC – Content-
Separated
• Advertising separated from a
medium’s content.
• Search (paid)
• Retargeting
30. • Daily data across 5 product categories
• Spend data for 11 ad activities per category
• Number of visits to different website parts
• Revenue per product category
• Enabled the calculation of revenue elasticity by
activity by category
• Revenue elasticity = % performance lift for a 1%
increase in the media’s budget
11/13/18 | 30
Subject: Major European Online Retailer
31. Current Marketing Allocation
Portals
7% Compare
6%
TV
18%
Radio
14%
Email
1%
SEA-brand
4%
SEA-product
23%
Retargeting
10%
Affiliates
17%
Concentration in Search Engine Advertising (SEA), only 6% to Comparison sites
32. • System of equations: explain each variable
by its own past and the present and past of
all other variables
• (eg TV à search à sale)
• Captures dynamic effects from ads to web
funnel stages and revenue
• wear-in/wear-out
• We allow customer to ‘skip stages’
• e.g. go straight to product page or checkout
page
11/13/18 | 32
Method: Vector Autogression (Chris Sims)
Christopher Sims receiving his
Nobel prize in December 2011
33. 11/13/18 | 33
Structural Vector Autoregression (SVAR)
• Block 1: no current back funnel Block 2: no current skip stages
• Block 3: no dynamic back funnel Block 4: no dynamic skip stages
35. • Comparison sites work best for Electronics
• WHY? Consumers want to compare prices for utilitarian products
• Retargeting and Portals work best for Fashion
• WHY? Hedonic product: fall in love but hesitate to buy à tempt again
• Referral sites work best for Sports & Leisure, Home & Garden
• WHY? Consumers look to authorities’ recommendations
11/13/18 | 35
Matching ad forms with categories
36. Which Ad Forms Were Effective ?
• Only 53% of Firm-Initiated forms: ‘Half of my advertising
is wasted’ still true today!
• 60% of Customer-Initiated Content-Separated
• 80% of Customer-Initiated Content-Integrated
38. When is the largest effect ?
• Same day for ALL Content-Integrated CIC forms
• Same day for 80% of Content-Separated CIC forms
• After 1 or 2 days for most Firm-Initiated firms
39. Why?
• Content-Separated CICs interrupt: an email at work, retargeting
when you are on a different purpose website, ads when organic
search key
• Content-Integrated CICs help with the purpose of visiting site:
price comparison, portals on category with links to other
websites
• Firm-initiated campaigns do not catch customer at time of
interest in purchase
40. How to Allocate: Ratio of Elasticities
• Analysis showed doubling on comparison ads lift sales by
15% (elasticity = 0.15), while doubling ‘retargeting’ increases
sales by only 5% (elasticity = 0.05)
• How should you divide your budget of $ 100,000 ?
• Ratio of elasticity = 0.15/0.05 = 3
• Spend 3x as much on Comparison = $ 75,000
42. But, How About Synergy?
• Synergy means that actions together have a stronger
effect than by themselves
• Sales = 2*OFFline + 5*ONline + 1 *OFFline*ONLine
• In other words, lower-elasticity action helps higher-
elasticity action and should therefore get a higher
allocation
• Example: Old Spice TV + social media campaign
42
43. Does your guy smell like the Old Spice Guy ?
• Focus on key benefit: smell
• Funny, creative, consistent:
1) TV ad gets reach
2) You Tube and fast response
to fan tweets get
engagement
Doubles sales within 1 year
44. What works best for your customers?
• Among Content-integrated ads:
• Affiliate sites if high involvement (e.g., shoes,
fashion, cars, alcohol)
• Paid search, price comparison site if high need
NOW (e.g., refrigerator, travel)
• Among Content-Separated (e.g. retargeting)
• When mood of website matches your offer
44
45. 3 attribution myths
• ‘Bottom funnel’ (retargeting, comparison) ads are overvalued
with last-click attribution
• Switching from Last-Click to other static models fixes your
marketing attribution problems
• Buying & implementing an attribution system is all you need to
do to improve decisions
46. Instead
• Customers switch back and forth between channels & can buy
way after taking ‘bottom funnel’ action
• Model-based attribution quantifies your customized journey with
your own data
• But, it needs your goals and insights to better assist you with
communicating and deciding
47. • Contact me at koen.h.pauwels@gmail.com
• LinkedIn/Twitter handle: koenhpauwels
• My blog: https://analyticdashboards.wordpress.com
• Professional Facebook page:
https://www.facebook.com/pages/Smarter-Marketing-
with-Analytics-Dashboards/586717581359393
• And check out my practical book:
It’s not the Size of the Data, it is How You Use it:
Smarter Marketing with Analytics & Dashboards
Want to learn more?