Beruflich Dokumente
Kultur Dokumente
Multivariate testing
Page 1 | Understanding multivariate testing techniques and how to apply them to your email marketing strategies
Multivariate testing
97%
81%
80%
60%
50%
43%
39%
40%
36%
28%
21%
20%
0%
11%
Subject
line
Creative Frequency
Time
of day
Day
of week
HTML
versus
text
11%
11%
Friendly Multivariate
from
Source: Experian Marketing Services Email Market Study, Acquisition and engagement tactics, December 2012
Multivariate testing
The problem is that the company may have missed out on finding the global
optimum by not testing the fourth combination of (I2, C1), which may have
yielded a click-to-open ratio even greater than 9.5 percent.
A/B tests assess one level of one factor versus the control group, but cannot
measure the interaction effect across factors.
The reason this method of sequentially testing one factor at a time fails to find
the optimal factor levels is that an interaction effect exists between factors 1
and 2. That is, the factor effects are not additive, but rather the combinations
among different factors and their levels produce an additional effect (interaction)
when used simultaneously. By not being able to capture interaction effects, this
sequential approach may miss the optimum altogether.
In situations such as these, it is more appropriate to perform a multivariate
test (factorial test to be exact), where all factors are changed together and
all combinations are accounted for. There is a wide array of different kinds of
multivariate tests available, but when the number of factors and the number of
factor levels to be tested are limited, the full factorial approach is the best option,
as it retains the most amount of information about the factors.
Page 3 | Understanding multivariate testing techniques and how to apply them to your email marketing strategies
Multivariate testing
Multivariate testing
The graph below is an example of what may be found through MVT testing:
Factor influence on click rates
42.0%
Layout
39.0%
Factors
Main image
10.8%
Call to action
Size
Color
0%
5.9%
2.3%
10%
20%
30%
40%
50%
Percent effect
Page 5 | Understanding multivariate testing techniques and how to apply them to your email marketing strategies
Multivariate testing
accurately account for the true multidimensional nature of your data, you will be
able to gather deeper analytic insights that just arent possible to collect via A/B or
A/B/C testing.
Applications of multivariate testing
Email creative testing
There are many different dimensions of an email creative that could possibly
affect subscriber engagement: layout, main image, call to action, color, size, etc.
Depending on the number of factors you are interested in testing and the amount
of variations you have to test for each factor, the overall number of potential
creative combinations can quickly become rather daunting. For example, if you
are interested in assessing the effects of two different layouts, two different main
images, five different calls to action, two different background colors and two
different sizes, you are left with 80 different email combinations to test. However,
through the science of experimental design, there exists ways to reduce this
number of combinations to test down to a much more reasonable figure. In fact,
a properly designed multivariate test can reduce the number of combinations to
test from 80 down to as few as 16, a much more manageable situation. Another
benefit to this approach is that the tester loses very little in terms of back end,
actionable insights when reducing the number of creative combinations to be
tested in an intelligent manner. In other words, the ability to identify which factors
have a significant effect (and conversely, which factors show little to no effect) on
user engagement is still able to be captured, along with the identification of what
the optimal settings of the email factors should be in order to maximize subscriber
response/engagement.
Single
factor A/B
Fractional
factorial design
Full factorial
design
Layo
Layo
Buy now
Act now
Get more
info
Cal
l to
acti
on
X X
X
X
X
X
X
X
Cal
X
l to
a
ctio
ut
X
Image
Image
ut
X
Cal
l to
acti
on
Multivariate testing
Volume %
17%
16.6%
2.7%
.12%
$.17
$173
Thuesday 15%
Wednesday 16%
16.8%
2.7%
.12%
$.16
$189
16.7%
2.6%
.11%
$.13
Thursday
15%
16.5%
2.5%
.11%
$.14
Friday
17%
16.4%
2.6%
Saturday
9%
Sunday
10%
17.8%
17.8%
.13%
2.9%
2.9%
0.0%
20.0% 0.0%
4.0%
Unique opens Unique clicks
.16%
.14%
$188
$195
$.16
$179
$.20
$.19
$176
$189
10%
8:00p.m.11:59p.m. 2%
9%
2%
17.6%
16.2%
3.2%
2.5%
.20%
.15%
$.22
$.18
16.1%
17.6%
2.4%
2.8%
.13%
.13%
$.17
$.15
18.2%
21.7%
2.9%
4.2%
.15%
.34%
$.16
0.0% 30%
Unique
opens
$135
$164
$174
$188
$.48
$182
$246
Page 7 | Understanding multivariate testing techniques and how to apply them to your email marketing strategies
Multivariate testing
Looking at these results, we see that Saturday and Sunday had higher open,
click and transaction rates, but much lower volume. For weekdays, Monday
had the highest revenue per mail, but Friday had the better click rate. Volume also
plays a role in time of day, as 8 p.m. to midnight and midnight to 4 a.m. had higher
responses than the more popular 4 a.m. to noon times. However, as every brands
customers behave differently, it is most likely the case that the optimal day of
week and time of day for your email deployments will show a different pattern of
performance. It is important to remember that the data from the aggregate graph
is retrospective, not a controlled study. When designing a time of day/day
of week study for your subscribers, it is imperative that the various day/time
groups are defined ahead of time and in a manner that incorporates random
sampling techniques.
A properly designed MVT will give you clarity into what matters most in terms
of deployment decisions and which day/time combinations have proven
themselves to be statistically significant in their ability to provide improved
metrics for your business.
Frequency and recency testing
The question of How often to mail? cannot be answered without a
comprehensive testing plan. Trying to assess the effects of frequency or recency
on retrospective data has a much higher risk of confounding or misallocating
performance due to other subscriber attributes like engagement or tenure. To
help remove this noise from the process, MVT can be leveraged to create proper
test groups beforehand, as well as identify what sample size requirements need
to be met in order for results to be statistically significant. The use of forwardlooking multivariate test designs versus backward-looking retrospective analyses
will provide you with the assurance that the conclusions drawn from your study
are accurate and statistically meaningful determinations of the optimal delivery
settings for your subscribers.
When testing for optimal frequencies, it is also important to remember that
not all engagement segments should be exposed to the same amount of
experimentation. Your engaged audience is already demonstrating desirable levels
of performance, and you want this behavior to remain. You dont want to run the
risk of disrupting this routine with too many frequency tests. At the other end of
the spectrum, you do not have much to lose when testing against your inactive
audience; there is only upside to be gained. Take advantage of this situation,
and test against a wider range of subscribers to see if any different patterns of
frequency tend to spur added levels of engagement.
Multivariate testing
Email cadence
Test
conservatively
Highly
engaged
Moderately
engaged
Test
liberally
Inactive
Sample learnings
Segment
Current frequency/
recency
Optimal frequency/
recency
Highly engaged
Moderately engaged
Inactive
2x per month/
clustered
Page 9 | Understanding multivariate testing techniques and how to apply them to your email marketing strategies
Multivariate testing
Conclusion
Remember, testing all the various aspects of your email program for optimal
performance is an important tactic for any successful email marketing
campaign. The testing process is also iterative, with no prescribed end. As the
business environment and goals of your brand change alongside the needs and
wants of your customers, it is important to continually test and fine-tune user
experiences on a regular basis to ensure your marketing messages are primed
for maximum effect.