accelerated testing for retention and crm programshearst...overview of retention / crm programs •...
TRANSCRIPT
Accelerated Testing
for Retention and CRM Programs
Brian Fichtel
Senior Director, Customer Retention
Hearst Magazines
Gordon Bell
President, LucidView
Managing Partner, Artestry gbell @ lucidview.com
1
Overview of Retention / CRM programs
• Multiple effort subscription renewal series utilizing a mix
of direct mail, e-mail, telephone and on-magazine efforts
• Generally segment by original acquisition source
(e.g. Direct Mail, Internet) and subscriber tenure
• Series runs for as long as 9 months
• Retention efforts are highly responsive and highly
profitable but have small quantities compared to
Acquisition Direct Mail
• All of the above makes testing a long, challenging
process with limited testing opportunities
2
Control Renewals / Retention Program
4 months
before expiration
- 3 months
- 2 months
- 1 month
Subscription
expiration
1 month after
expiration
+ 2 months
+ 3 months
+ 4 months
3
CRM Testing: Real-World Challenges
4 months
before expiration
- 3 months
- 2 months
- 1 month
Subscription
Expiration
+ 2 months
+ 3 months
+ 4 months
Subscription segments
Direct mail Internet Prior
Renewals
20 U.S. Magazines
Car
and D
river
Cosm
opolita
n
Countr
y L
ivin
g
ELLE
E
squir
e
Food N
etw
ork
M
agazin
e
Good
Housekeepin
g
Mari
e C
lair
Opra
h
Magazin
e
Popula
r M
echanic
s
Redbook
Road &
Tra
ck
Sevente
en
Tow
n &
Countr
y
Vera
nda
Wom
an’s
Day
Monthly expire groups
Month 1
Month 2
Month 3
Month 4
4
“Business Case” for Multivariate Testing
• Many creative, offer and segmentation test ideas but
small retention quantities that limit testing opportunities
• Started with
41 ideas to test…
Envelope size Stronger copy pushing payment
Postage Laser magazine title on all letters
Return slip position More aggressive in promotion of RAB discount
Branding Increase urgency around expire
Format Attached mail
Offer and payment info Same form with all efforts
Urgency Postcard with drive-to-web
Magazine benefits copy Offer expression - $0.97 vs. rounded up (9.97 vs. 10)
Guarantee No discount for late effort High segments
Payment chit Incentive - extra issues vs. pricing
Highlight discount Credit card renewal option
Change price 2- and 3-year renewal option
Incentive for early renewal Hearst Rewards (earn points)
Late-series incentive Renew now for digital copy/app
Multi-year offer Special offer upon cancel request
Compress contact stream Add a chit - gift/cross-sell subscription
More e-mail Premium with 2- or 3-year term
Start renewal cycle earlier Shorter term offer (6 issues) offer
Add telemarketing efforts Sweepstakes entry for early renewal
E-mail beforehand Magnitude of price change (small/medium)
Decrease time between efforts
5
“Business Case” for Multivariate Testing
• Multivariate testing provided opportunity to test many
ideas in a short period of time
– Enabling results in 9 months instead of 3 years
• Multivariate testing also provided an opportunity to
gain insights about test element interactions
that’s not possible with traditional A/B testing
6
Test Design: Universe of Possibilities
2-level
Test elements
Optimal designs • Complete freedom
• Cost = complexity
and error if assumptions
are wrong
Taguchi designs • Use many levels
• Bold assumption that
interactions don’t exist
Orthogonal test designs
• Full factorial
• Fractional factorial
• Plackett-Burman
• Centerpoints
• CCD (etc.)
7
A/B split-run
tests
Selected designs
“Mosaic” Test Design
• Two test designs
1. 11 creative changes
2. 6 elements of the offer and contact strategy
• Why?...
a) Number of test elements
b) Magnitude of impact (creative changes tend to have a smaller effect)
c) Value of interactions – mix of contacts and offer elements tend to have
larger, more valuable interactions
• Choices
1. 12-recipe test for 11 creative elements ( = fewer package versions)
2. 16-recipe test for 6 offer and creative elements ( = more interactions)
8
CRM Testing: Real-World Challenges
4 months
before expiration
- 3 months
- 2 months
- 1 month
Subscription
Expiration
+ 2 months
+ 3 months
+ 4 months
Subscription segments
Direct mail Internet Prior
Renewals
20 U.S. Magazines C
ar
and
Dri
ver
Cosm
opolita
n
Countr
y L
ivin
g
ELLE
E
squir
e
Food N
etw
ork
M
agazin
e
Good
Housekeepin
g
Mari
e C
lair
Opra
h
Magazin
e
Popula
r M
echanic
s
Redbook
Road &
Tra
ck
Sevente
en
Tow
n &
C
ountr
y
Vera
nda
Wom
an’s
Day
Monthly expire groups
Month 1
Month 2
Month 3
Month 4
Test ideas 4
segmentation models
11 creative elements
(2,048 possible combinations)
6 elements of the offer and
contact strategy (96 combos)
9
Test Elements
Test Elements - Creatives (-) Control (+) New idea
A Envelope size Large Small
B Postage Metered Indicia
C Return slip position Top of form Bottom of form
D Branding Magazine branding Hearst corporation branding
E Format Letter (with signature) Statement (no signature)
F Offer and payment info On right side Move to left side
G Urgency Current urgency copy Different urgency copy
H Magazine benefits copy Control Stronger benefits
J Guarantee None Add guarantee copy
K Payment chit No Add chit
L Highlight discount No Yes
Test Elements - Offer and Strategy (-) Control (+) New idea
A Change price Control +/- 20%
B Incentive for early renewal No Yes, offer incentive in 2nd effort
C Late-series incentive Discount Premium
D Multi-year offer Only once Add multi-year option to additional efforts
E Compress series timing Control (every 4 weeks) Compress (every 2-3 weeks) and add efforts at end
F More e-mail efforts Control More e-mail following DM efforts
Brainstormed 41 new ideas selected 17 test elements (split into 2 test designs)
10
“Mosaic” Test Designs
Envelo
pe s
ize
Posta
ge
Retu
rn s
lip p
ositio
n
Bra
ndin
g
Form
at
Off
er
and p
aym
ent
info
Urg
ency c
opy
Magazin
e b
enefits
copy
Guara
nte
e c
opy
Paym
ent
chit
Hig
hlig
ht
dis
count
(late
r
Recipe A B C D E F G H J K L Sample
1 – – – – – – – – – – – 205632 + + + – + + – + – – – 205883 + – + + – + – – – + + 204804 – – – + + + – + + – + 204715 + + – + + – + – – – + 204666 + – – – + + + – + + – 204577 + – + – – – + + + – + 205818 – + + – + – – – + + + 206339 – – + + + – + + – + – 20569
10 – + – – – + + + – + + 2045511 – + + + – + + – + – – 2051612 + + – + – – – + + + – 20528
Creative Test
Change p
rice
Incentive f
or
early r
enew
al
Late
-series incentive
Multi-year
off
er
Com
pre
ss s
eries t
imin
g
More
e-m
ail
eff
ort
s
Recipe A B C D E F Sample
1 – – – – – – 20,4372 + + – – – – 19,8803 – – – + – + 20,5854 + + + – + – 19,9455 + + + + + + 20,0816 – + – + + – 20,4677 – – + – + – 20,3908 + – + – – + 19,9799 + – + + – – 20,083
10 + – – + + – 20,09011 – + – – + + 20,19012 + + – + – + 19,98813 – + + + – – 20,51614 – – + + + + 20,55715 + – – – + + 19,95116 – + + – – + 20,170
Offer & Strategy Test
11
Creative Test (control)
12
Creative Test (2 versions)
E+
F+
D+
C+
G+
H+
13
Mosaic Creative Test – Results
-0.10%
-0.16%
-0.23%
+0.39%
+0.58%
-0.73%
-0.81%
-1.22%
-1.22%
-1.48%
-3.59%
0.0% 1.0% 2.0% 3.0%
B-: Postage
H-: Magazine benef its copy
C-: Return slip position
F+: Of fer and Payment Info
K+: Payment Chit
E-: Format
G-: Urgency
L-: Highlight discount (no)
A-: Envelope size (large)
J-: Guarantee (no)
D-: Branding (magazine)
Effect (avg RR of test as % change vs. control)
Hearst Magazines CreativeTest: Renewal Rate|Effects| as % change versus the control
Elements: sign of effect (optimal setting)
Four significant effects:
D-: Brand letters with the magazine title
(not Hearst Magazines)
J- and L-: Skip the guarantee language
and highlighting the discount
A-: Keep the larger envelope
14
Mosaic Offer & Strategy
Test – Design
3-levels of price means that each A+
Recipe is divided into two groups:
Half with a 20% drop in price
Half with a 20% price increase
Change p
rice
Incentive f
or
early r
enew
al
Late
-series incentive
Multi-year
off
er
Com
pre
ss s
eries t
imin
g
More
e-m
ail
eff
ort
s
Recipe A B C D E F Sample
1 – – – – – – 20,4372 + + – – – – 19,8803 – – – + – + 20,5854 + + + – + – 19,9455 + + + + + + 20,0816 – + – + + – 20,4677 – – + – + – 20,3908 + – + – – + 19,9799 + – + + – – 20,083
10 + – – + + – 20,09011 – + – – + + 20,19012 + + – + – + 19,98813 – + + + – – 20,51614 – – + + + + 20,55715 + – – – + + 19,95116 – + + – – + 20,170
15
Recipe A B C D E F Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4
1 – – – – – – 20,4372 + + – – – – 19,8803 – – – + – + 20,5854 + + + – + – 19,9455 + + + + + + 20,0816 – + – + + – 20,4677 – – + – + – 20,3908 + – + – – + 19,9799 + – + + – – 20,083
10 + – – + + – 20,09011 – + – – + + 20,19012 + + – + – + 19,98813 – + + + – – 20,51614 – – + + + + 20,55715 + – – – + + 19,95116 – + + – – + 20,170
Change p
rice
Incentive f
or
early r
enew
al
Late
-series incentive
Multi-year
off
er
Com
pre
ss s
eries t
imin
g
More
e-m
ail
eff
ort
s
Segment 1 Segment 2 Segment 3
Total Sample Size
Mosaic Offer & Strategy Test – analytical plan
16
Offer & Strategy Test –
Execution: recipes 1 vs. 5
- 4 months
- 3 months
- 2 months
- 1 month
Expiration
+ 1 month
+ 2 months
+ 3 months
+ 4 months
+ 5 months
$ $
$ $
$ $
$ $
$ $
$ $
$ $
$ $
$ $
$ $
17
Mosaic Offer Test – Results
0.07%
0.11%
0.12%
0.17%
0.25%
0.35%
0.36%
0.65%
0.67%
+0.86%
+1.22%
+1.56%
-2.43%
+3.04%
-5.86%
0.0% 1.0% 2.0% 3.0% 4.0% 5.0% 6.0%
AE
ACF
AC
ACD
CD
AB
AD
AF
CF
E+: Compress contact stream (yes)
F+: More e-mail (yes)
A+: Change price (yes)
D-: Multi-year of fer (no)
B+: Incentive for early renewal (yes)
C-: Late-series incentive (discount)
Effect (avg RR of test as % change vs. control)
Hearst Magazines Billing & Renewals Test: Renewal Rate|Effects| as % change versus the control (optimal = 6.7% lift)
Elements: sign of effect (optimal setting)
All 6 elements are significant
(but no interactions)
C-, B+: Keep the later discount,
but add and early-renewal incentive
E+, F+: Add more contacts – more e-mails
and compressed timing
D-: Skip the multi-year offer
A+: Change price (but how?)
18
-50%
0%
50%
100%
150%
200%
250%
-20% Control +20%
% c
ha
ng
e in
re
ve
nu
e v
s. c
on
tro
l
A: Price
Effect (A): Price change(change in Revenue versus control)
Magazine #1
Magazine #2
Magazine #3
Mosaic Offer Test – Effect of Price…
To maximize revenue
• For magazine #1,
increase price 20% (Revenue increases 19%
but response is 9% lower)
• Increase price 20%
for magazine #2 (Revenue increases 23%
but response is 22% lower)
• For magazine #3, remain
at the control price
19
-50%
0%
50%
100%
150%
200%
250%
-20% Control +20%
% c
ha
ng
e in
re
ve
nu
e v
s. c
on
tro
l
A: Price
Effect (A): Price change(change in Revenue by Segment)
#1 - Cancel risk
#1 - Loyal
#2 - Cancel risk
#2 - Loyal
#3 - Cancel risk
#3 - Loyal
… Depends upon segmentation
Differences by
Segmentation Model
• “Loyal” customers (defined by a combination
of statistical models)
Respond well
to higher prices
• Subscribers who are a
“Cancel risk” are better
served at the control or
lower price…
20
Because… Need to consider response
(not shown) along with revenue
Renewal rate drops up to 26%
for this high-risk segment
And we need to deliver promised rate base
Plus renewals are lower cost
than new subscribers
Summary of Mosaic Tests
• Tested 17 marketing-mix elements at once
• Quantified 10 significant elements
– Analyzing main effects, interactions, and “curvature”
in the price effect
• Pinpointed differences by segment, title, and
statistical model
• Tested in 9 months what would require more
than 3½ years with A/B splits
21
Conclusions & Insights
• Benefits of multivariate testing for Hearst retention marketing
Identified creative, offer, segmentation elements with most leverage
with some rolled out immediately, others fine-tuned via more tests
Low-leverage test elements removed from future test consideration
Additional tests to “fine-tune” high-leverage elements led to roll outs
New control series that outperformed existing control series of the
last 7 years (response lifts as high as 15%)
New model-based offer and segmentation strategy based on a
subscriber’s likelihood to renew or cancel
22
Risks & Recommendations
• Risks of multivariate testing for Retention efforts
Testing more elements increases risk of negative impact
Normally low risk test elements could be high risk when mixed with
other test elements to form counter-intuitive recipes needed for test
Control creative may not accommodate variety of recipes required
for testing, which may force additional creative changes and
response risk
• Recommendations on when to use multivariate testing
Statistically significant A/B test volumes can’t accommodate
all test elements
There’s a lot to test and results are needed quickly
The organization supports bold, high risk/reward testing
23
Accelerated Testing
for Retention and CRM Programs
Brian Fichtel
Senior Director, Customer Retention
Hearst Magazines
Gordon Bell
President, LucidView
Managing Partner, Artestry gbell @ lucidview.com
24