Placement optimization, which allows advertising budgets to be dynamically allocated across Facebook and Instagram, led to improved campaign delivery over using Facebook alone in a study of 10 brand advertising campaigns. Specifically, placement optimization resulted in 4.1% higher reach and 5.2% lower cost per impression, on average. While there was no statistically significant difference in brand lift between the two approaches, placement optimization achieved the same level of lift at a lower average cost per lift of 10-27% due to its ability to reach more people within the same budget.
3. Optimizing Audience Buying on Facebook and Instagram 1
Running campaigns simultaneously across different
digital platforms should provide advertisers with
extended reach into new audiences as well as the
ability to reach pre-existing audiences in a more
cost-effective manner. However, in practice, it isn’t
immediately obvious what the most optimal way is of
achieving these benefits and how significant they will
be for any given campaign.
Over the past few years, Facebook’s evolution into a
family of apps and services has given advertisers new
ways of extending their campaigns to platforms such
as Instagram and the Audience Network. While some
advertisers and agencies have been experimenting with
manual allocation of budgets, Facebook fully launched
placement optimization in late 2015 as a way of
providing advertisers with an easier way of optimizing
campaign delivery across Facebook, Instagram and the
Audience Network. This paper aims to measure the
effectiveness of optimizing audience buying to deliver
value to brand advertisers.
Introduction
How placement optimization works
Previous Marketing Science research has shown
that over the course of a campaign, cost per
outcome — from mobile app installs to brand
awareness or online conversions — tends to
vary between and within platforms due to
audience behavior and characteristics. While
advertisers and agencies can manually allocate
budgets across platforms to account for these
changes, placement optimization leverages
Facebook’s ad delivery system to dynamically
seek out the lowest cost per outcome at any
given point in time wherever it’s available,
whether it be on Facebook or, as in the
illustrative example in Figure 1, on Instagram. As
a result, placement optimization should provide
lower cost per outcome than advertising on a
single platform or even trying to manually
allocate budgets across various platforms.1
4. Facebook IQ White paper2
Introduction
In this context, placement optimization is both an
important and interesting topic for the Facebook
Marketing Science team to study. Marketing Science’s
mission is to help marketers understand how to
leverage the maximum possible value from their
advertising dollars. Determining whether placement
optimization provides additional value and quantifying
those gains is thus an important and straightforward
goal for the team.
We are also interested in placement optimization,
as it offers a novel opportunity to robustly measure
multichannel, cross-platform campaigns. Cross-
platform and cross-device campaigns have traditionally
been difficult to measure effectively; cookie-based
measurement does a poor job of tracking individuals
across devices and platforms and can tend to
undervalue the true contribution of mobile.2
Not
surprisingly, a recent eMarketer survey showed that
when it comes to cross-platform measurement, 35%
of advertisers are “not using a robust measurement
technique,” while 34% say they “evaluate each channel
individually and optimize based on channel-specific
performance.”3
Rather than cookies, Facebook
measurement solutions instead rely on people-based
identities and conversion pixels, meaning we are in the
unique position of being able to accurately measure
the impact of a cross-platform campaign and better
understand what factors might be contributing to any
efficiency gains we observe.
For this study, we’ve used placement optimization
across Facebook and Instagram as a starting point to
measure placement optimization’s effectiveness for
brand advertisers. In the future, we plan to extend our
methodology to look at direct response objectives as
well as expand it across our family of apps.
Costperoutcome
Time elapsed
Instagram-only avg
cost per outcome
Facebook-only avg
cost per outcome
Placement optimization
avg cost per outcome
Facebook
measurement
solutions
rely on people-
based identity and
conversion pixels
rather than cookies
Figure 1. Graphic illustration of placement optimization mechanism
5. Optimizing Audience Buying on Facebook and Instagram 3
Methodology
Test design
We focused our placement optimization research
on brand campaigns that ran across Facebook and
Instagram and used randomized controlled trials
(RCT) as the primary methodology. We worked with
a total of 10 brand advertisers across a range of
verticals, countries, budgets and target audiences.
While we deliberately chose a variety of advertisers
and campaigns to test, we used the same underlying
methodological approach throughout. Working
closely with each advertiser, we split their budget
and audience size equally between a placement
optimization test cell in which an ad impression
could be either on Facebook or on Instagram and a
Facebook-only test cell where all impressions were
delivered exclusively on Facebook.
Our choice of test cells was deliberate. We
could have potentially added other test cells or
combinations of test cells, such as including an
Instagram-only cell or cells with manually allocated
budget across Facebook and Instagram. While these
are both valid test designs that would have provided
potentially meaningful insights, our consideration
was driven primarily by a desire to focus on the
most common question we get from advertisers
that tend to be comfortable with their Facebook
campaigns and that are looking for answers on how
best to incorporate Instagram into their campaign
strategy. With that question as our priority, we felt
the Facebook-only versus placement optimization
test design was the most logical question to start
understanding this topic.
With the two test cells decided up-front, every
Facebook and Instagram user in a given test was
randomly assigned to either the Facebook-only or the
placement optimization cell prior to the campaign
launch. Each test cell also had its own control group
that did not receive any ads. This setup meant that
the only difference between the two test cells was
that consumers in the placement optimization cell
had the opportunity to be reached across both
Facebook and Instagram whereas consumers in
the Facebook-only cell could only be reached on
Facebook. Apart from that important difference,
every other campaign feature — such as campaign
objectives, optimization method, maximum bid (if
using the auction), audience targeting and creative —
were exactly the same across both test cells. Figure 2
provides an overview of our test design.
This randomization of consumers across rigidly
defined cells is the hallmark of the RCT methodology
and ultimately provides for greater measurement
accuracy compared to observational studies.4
6. Facebook IQ White paper4
Methodology
Measurement of brand lift
Placement
optimization
Randomize groups Potential placement Brand polling
Facebook-
only
Facebook & Instagram
(+ other media)
Facebook
(+ other media)
Note: Campaign budget, target audience, creative and campaign durations are hold constant for both cells.
Figure 2. Test design
7. Optimizing Audience Buying on Facebook and Instagram 5
Methodology
Data collection procedure
Despite some limitations, surveying has traditionally
been the data collection method of choice for brand
advertisers that want to understand the impact of a
given campaign on their brand equity. With that in
mind, we measured the impact of each of the two test
cells (placement optimization and Facebook-only) by
surveying a randomly selected subset of Facebook and
Instagram users from both the exposed and control
groups. We collected on average about 10,000 survey
completions per test. The surveys consisted of three
questions asked sequentially to the same person: while
all the campaigns we measured asked ad recall as the
first question, advertisers were allowed to customize
the other two survey questions they wanted to ask.
These questions generally fell into either upper-mid
funnel metrics around awareness and affinity to lower-
funnel metrics around purchase intent or willingness
to recommend.5
One important decision regarding our test design
was about choosing the platform on which to
administer the surveys. We ultimately elected to
survey on Facebook mobile feed regardless of whether
respondent had seen ad impressions on Facebook or
Instagram. The main reason for this was to control for
the fact that survey respondents on Instagram are not
necessarily the same kind of survey respondents we
see on Facebook; they likely have different baseline
attitudes and perceptions of the brands we were
testing, which could complicate comparing survey
results between platforms and between test cells.
We therefore chose to limit this potential bias by
surveying exclusively on Facebook. The tradeoff
of this approach is that respondents who saw ad
impressions on Instagram may not have been active
on Facebook and thus not eligible to respond to the
survey on Facebook. This in turn could have led to an
outcome where we underestimated Instagram’s true
impact since we might have collected relatively fewer
respondents who were exposed on Instagram.
To address this potential bias, we compared the
distribution of Instagram survey respondents
in the placement optimization cell to the actual
Instagram reach numbers for the campaigns in
question. Overall, we did find that Instagram was
slightly underrepresented in terms of total survey
respondents who were exposed to the Instagram
campaigns when compared to actual Instagram reach.
However, the responses from both groups — those
exposed on Facebook and those exposed on Instagram
— showed no statistically significant differences. Given
that there were no meaningful differences in survey
responses between those exposed on Facebook and
those exposed on Instagram, we concluded that our
survey approach for this campaign did not appear to
introduce any noticeable biases.
8. Facebook IQ White paper6
Methodology
Hypotheses
In thinking about our test design, we considered
three different ways in which placement optimization
could impact a given campaign and formulated a
hypothesis for each.
Hypothesis 1 (Campaign delivery): Placement
optimization is more efficient at campaign delivery
than Facebook-only. By allowing ad spend to flow
freely across Facebook and Instagram, placement
optimization ultimately has access to a wider audience,
which should lead to higher reach with lower cost per
impression (CPM) and/or cost per reach (CPR).
Hypothesis 2 (Brand lift): Placement optimization
generates higher lift than Facebook-only. Lift here is
defined as the difference between the percentage of
favorable answers between the test and the control
groups. One reason why placement optimization might
perform better is that it is more effective at reaching
audiences that are more likely to be influenced by a
campaign. Another reason might be that all things held
equal, seeing an ad on two different platforms might
create a more lasting impact than only seeing it on one
platform. For this study, we compared the brand lift
between placement optimization versus the Facebook-
only cell by using a difference in difference framework
(i.e., is the difference between the control and
exposed for a given brand metric higher for placement
optimization than for Facebook-only?).
Hypothesis 3 (Cost efficiency): Placement
optimization is more cost-effective in driving lift.
If both Hypothesis 1 and 2 hold — e.g., you’ve reached
more people and created more lift for the same
budget — it follows that placement optimization
would offer a more cost-effective way of achieving
a campaign’s objectives. However, it’s also possible
for Hypothesis 3 to be true even if either Hypothesis
1 or Hypothesis 2 is not supported. For instance, it
is theoretically possible to observe the same level
of lift between the two test cells but emerge with
more people being impacted by the campaign via
increased reach. The metric we use for testing this
hypothesis is cost per lift (CPL). It is a concept the
Marketing Science team has previously encouraged
advertisers to consider6
and is essentially a function
of three interdependent factors: campaign spend, ad
effectiveness and reach and frequency. Cost per lift
captures the relationship of these factors:
Cost per lift = f (Reach and Frequency, Ad
Effectiveness | Spend) = Spend/(Reach × Lift)
Simply put, cost per lift estimates how much it costs
an advertiser to change the attitude/perception of a
given person influenced by a campaign.
9. Optimizing Audience Buying on Facebook and Instagram 7
Results
Between December 2015 and May 2016, we
worked with 10 brand advertisers to run placement
optimization tests. These advertisers came from a
range of countries (Brazil, Canada, UK and US) and
verticals including CPG (3), Retail (2), Automotive (2),
Entertainment (1), Politics (1) and Charity (1).
On average, campaigns had an audience size of
13 million and an impression frequency of 1.8 per
week over three weeks (which we consider to be in
an optimal range). Across the campaigns, we also
observed that on average, 83% of impressions were
delivered on Facebook and 17% of impressions were
delivered on Instagram.
For each campaign, we measured the following metrics:
• Cost per reach/impression (Hypothesis 1)
• Brand lift (Hypothesis 2)
• Cost per lift between the placement optimization
and the Facebook-only cell (Hypothesis 3)
Campaign delivery
Result 1 (Campaign delivery): Hypothesis 1 is
supported. On average, placement optimization
campaigns had 4.1% more reach and a 5.2% lower
cost per impression (CPM), which in turn led to a
lower cost per reach (CPR) of 5.8%. For the average
campaign, this translated into an additional half
million people reached for the same budget when
compared to the Facebook-only cells. The increased
reach and lower CPMs were even more pronounced
for campaigns that were booked using Reach and
Frequency (+6.4% for reach and -7% for CPM).
As illustrated in Figure 3, we also found evidence
that campaign spend might be positively correlated
to placement optimization efficiency. For example,
the larger the campaign, the more value placement
optimization delivers.
0.0%
5.0%
-5.0%
10.0%
-10.0%
15.0%
-15.0%
20.0%
-20.0%
Costofreaching1000peopleimprovement
Total spend
Relationship between total spend and cost of 1000 reach improvement
Figure 3: The relationship between total spend and cost of reach improvement
Across the
campaigns,
we observed that
on average, 83% of
impressions were
delivered on Facebook
and 17% of impressions
delivered on Instagram
10. Facebook IQ White paper8
Results
Brand lift
Result 2 (Brand lift): Hypothesis 2 is rejected.
There is not a statistically significant difference in
lift between the placement optimization and the
Facebook-only test cells. The hypothesis is rejected
for both upper- and lower-funnel brand metrics.
At the outset, Hypothesis 2 was going to be the most
difficult one to prove. Even though we collected a
relatively large number of survey responses (~10,000
respondents per campaign), the margin of error
around the surveys means that in some instances
placement optimization would have had to perform
several percentage points better than Facebook-
only for us to register a statistically significant lift. In
essence, even with larger sample sizes, survey is still
prone to Type II error when it comes to hypothesis
testing. Given the above reasons, we are not willing to
unambiguously reject Hypothesis 2 that placement
optimization may indeed be providing higher lift
but have to conclude that any gains it does provide
may be marginal and difficult to fully measure with
conventional survey.
While placement optimization did not provide
additional lift per se, it is important to note that it
did provide equivalent lift, as it reached more people
overall. In other words, placement optimization
produced the same level of lift even though its
budget was ultimately spread across a wider exposed
audience. This is because despite the increased reach
in placement optimization, there was only a marginal
difference in average frequency between the two
cells (2.94 for placement optimization and 2.99 for
Facebook-only). Recent Facebook research7
has shown
a positive and causal relationship between frequency
and brand lift and in the case of our tests. Hence, we
believe brand lift was constant across the two cells
because placement optimization’s greater reach didn’t
come at the expense of less frequency.
Placement
optimization
produced the
same level of lift
even though its
budget was ultimately
spread across a wider
exposed audience
11. Optimizing Audience Buying on Facebook and Instagram 9
Results
Cost per lift
Result 3 (Cost per lift): Hypothesis 3 is supported.
Across the various types of brand questions we asked,
from top-funnel questions like ad recall to lower-
funnel questions like purchase intent, cost per lift on
placement optimization was on average 10% to 27%
lower. As mentioned in the Hypothesis section of this
paper, cost per lift essentially tells us how much an
advertiser paid, on average, to change the attitudes or
raise the awareness level of a given person influenced
by the campaign. In the case of our placement
optimization tests, when we combine the fact that 1)
lift is comparable between placement optimization
and Facebook-only, 2) campaign spend is the same
and 3) placement optimization had a larger reach,
the cost of impacting a person is lower for placement
optimization than the Facebook-only cell.
Illustration of placement optimization cost efficiency calculation
Cost per lift =
x Reach
Spend
Brand lift
Note:
- Spend is the same for the 2 test cells
- Brand lift is the same for the 2 test cells
- Reach is higher for placement optimization
A look at how sales outcomes were
impacted by placement optimization
While all of our campaigns used brand survey
as the primary measurement tool, one of the
10 campaigns also concurrently ran a DLX
study, giving us an opportunity to look at how
sales outcomes were impacted (if at all) by
placement optimization. The overall results
for the campaign (across both test cells) were
statistically significant and return on ad spend
(ROAS) positive. Within the test cells, placement
optimization showed a 5% improvement over
the Facebook-only cell. While inferences cannot
be made based on a single data point, the results
point to the possibility that the incremental reach
offered by placement optimization might also be
able to drive more real-world outcomes.
Figure 4: Illustration of placement optimization cost efficiency calculation
12. Facebook IQ White paper10
Conclusion
What it means for marketers
While our initial research is somewhat narrow in
scope, the results of our research nevertheless
offer a clear answer to a specific question many
brand advertisers have about whether placement
optimization performs better than Facebook-only
campaigns. We plan to build on this initial set of
findings in future research, looking at things like direct
response objectives, creative differentiation and other
apps and services, such as the Audience Network.
Our findings show that advertisers should use
placement optimization for brand campaigns.
Across the 10 brand studies we looked at, placement
optimization consistently reached more people at
lower cost, generated comparable lift to Facebook-
only campaigns and provided better cost efficiency in
moving brand metrics. It is worth noting that all of our
tests used the same creative across both Facebook
and Instagram. Future tests will look at the impact of
customizing creative across both platforms. While
there may be specific instances where it makes sense
for advertisers to think about running Facebook and
Instagram campaigns separately, we believe that, at a
minimum, brand advertisers should think of placement
optimization with identical creative as the default
option over a Facebook-only campaign.
As a starting point, we think this research is useful
both in terms of its insights and the validation it
provided for the multi-cell lift test. We have three key
recommendations for marketers interested in running
brand campaigns across Facebook and Instagram:
1. Focus attention on the overall outcome instead
of guaranteed delivery on a specific platform.
It is normal for campaign delivery to take place
mostly on one platform rather than be evenly
distributed across both platforms. When seeing
skewed delivery, do not interpret that as a sign
of placement optimization underperforming —
rather see it as placement optimization doing its
job and focusing your spend on the platform(s)
where it is finding the most value for a given
campaign and audience.
2. Be mindful of ensuring sufficient campaign
spend and keep a large audience. Placement
optimization appears to deliver better results with
larger audiences and budgets. Smaller placement
optimization campaigns will likely never do worse
than a Facebook-only campaign but the benefits
might not be as evident.
3. Set up your own randomized control tests
to scientifically measure placement
optimization versus Facebook-only. Some
advertisers tend to evaluate a new ad product or
strategy by comparing results to previous
campaigns. Although this might sound intuitive
and can potentially provide some level of insight,
this approach could generate heavy biases based
on the differences, however small, that take place
between campaigns. We therefore encourage
marketers to set up their own randomized control
tests (either using the lift tool or our split tool) to
validate how effective placement optimization
works based on their business objectives.
Placement
optimization
consistently reached
more people at lower
cost, generated
comparable lift
to Facebook-only
campaigns and provided
better cost-efficiency in
moving brand metrics
13. Optimizing Audience Buying on Facebook and Instagram 11
Appendix
Notes & Sources
1
Placement optimization delivers impressions only
where they are most cost-efficient instead of evenly
across platforms. There may be instances when ad
impressions have zero delivery on a platform due
to different audience activities across different
times of day.
2
“People-Based Measurement: In a World of Increasing
Consumer Choice,” Facebook, Feb 2016.
³ “Marketers Still Struggle with Cross-Channel
Measurement” by eMarketer, Nov 11, 2015.
⁴ The specific benefits of the RCT methodology,
compared to more traditional observations
studies, are discussed in depth in a recent joint
white paper between Facebook and the Kellogg
School of Management of Northwestern University,
“A Comparison of Approaches to Advertising
Measurement: Evidence from Big Field Experiments at
Facebook.” See also, “Demystifying Measurement: Why
Methodology Matters” by Facebook IQ, Mar 2016.
5
Ad recall: Do you recall seeing an ad on a mobile
device in the past 7 days?
Awareness: Have you heard of ____ campaign? /
Which brand comes to your mind when
thinking of ___?
Familiarity: How familiar are you with _____?
Favorability: What is your opinion of _____ ?
Intent: How likely are you to consider _____?
6
“Measuring for Success: Facebook on Facebook and
Instagram” by Facebook IQ, Sep 2015.
7
“Efficient Frequency” by Facebook IQ, Jul 2016.