Note: this post is adapted from content from the upcoming Modern Mobile Marketing at Scale workshop taking place in Berlin on April 23. More information here; places are still available for the workshop.
Facebook’s three mobile campaign optimization strategies — MAI, AEO, and VO — are often seen as being mutually exclusive, meaning that one might always, in all cases, be used in place of the others. This is rarely true; most apps would benefit by combining these optimization strategies into a portfolio of campaigns that reach users based on the most effective campaign settings for that audience.
Note that the audience qualifier is important. Facebook traffic is all governed by the same auction: there is no “MAI traffic” or “VO traffic,” just traffic, and so running multiple ad sets with different campaign strategies essentially pits those campaigns against each other for the same users. The shaded overlap in the diagram below represents a group of users, present in both audiences, that would be bid against by the campaigns targeting Audience 1 and Audience 2.
One critical aspect of effectiveness with a composite, multi-strategy campaign portfolio is targeting well-delimited, bounded advertising audiences by each strategy. This sounds obvious, but it can be deceptively onerous to implement in practice, especially when using Facebook’s own tools for audience enlargement like lookalike audiences or lowest-cost bidding.
When I hear teams declare something like, “MAI doesn’t work for us,” or “AEO doesn’t work for us,” I usually interpret that as meaning that the team hasn’t managed to segment and automate audience targeting and creation in a way that allows it to scale spend on Facebook. Facebook is the largest and most effective advertising platform in the world: if a team can’t “make Facebook work,” it reflects poorly on the team, not on Facebook.
Note that this doesn’t necessarily apply to VO campaigns, which are functionally different than AEO and MAI campaigns and might not be compatible with certain monetization mechanics (more on that in this article). Since VO campaigns optimize for magnitude of user “value,” then that campaign strategy tends to not work well for products without long-tail LTV distributions, for example subscription apps or very casual mobile games.
But that aside, for many apps, a collection of campaign strategies, each targeted at different audience segments based on the monetization potential and size of those segments, is optimal. This is really no different than simply segmenting audiences and bidding on them differently, except that each campaign strategy comes with different tools for optimizing traffic delivery.
Very often, this segmentation is accomplished through geography targets: for instance, splitting the total audience into value “tiers” based on ARPU and using non-ROAS-qualified VO campaigns on the top tier of traffic and targeting the remaining tiers with AEO and MAI campaigns. Lowest cost (“autobid”) MAI campaigns can be difficult to make profitable for high-ARPU geographies, for instance, but LTV / CAC deltas can compress or even flip positive for lower-ARPU geographies with autobid as acquisition costs plummet.
One reason that purely geographical audience segmentation is so powerful for targeting is that it can’t be violated: a person is either in a geography or they aren’t, and the advertiser has full control over that targeting distinction. Compare this to multiple sets of lookalike audiences, which are informed by the advertiser but defined by Facebook: targeting multiple lookalikes with multiple campaigns at the same time in the same geographies, even when the seed audiences are compositionally different (eg. tutorial complete vs. payers) engenders a potentially costly risk of audience overlap.
But that’s not to say that audiences can’t be segmented within geographies. For one, lookalikes don’t have to be used: if an advertiser has a large enough list of users to target, it can simply target that list (as in re-targeting / re-engagement campaigns). Device types and platforms (iOS vs. Android) can also be used for audience segmentation, since those qualities are also immutable. The broader point is that, if the audience target is seen as a function of the campaign strategy — that is, if the campaign strategy defines how the audience is targeted because some strategies are more effective on certain audiences — then it’s easy enough to draw impenetrable boundaries between campaigns so that audiences don’t overlap.
The benefit of thinking this way — that campaign construction starts with the selection of a campaign strategy that then defines the audience — is that VO and AEO campaigns almost always work best when they are given maximum audience breadth to experiment. VO, especially, is a “greedy” campaign strategy that performs optimally not only with maximum audience surface area in which to discover potential users but also maximum (unconstrained by a min-ROAS setting) budget with which to compete for the highest-value users in an auction.
By defining the audience as a function of the campaign strategy used, it’s more obvious to make these pairings (campaign strategy to audience) as a matter of value and size. If unconstrained VO is to be tested, it should be with untargeted, high-ARPU segments (eg. US iPhone); other strategies and audience pairings then cascade down the value spectrum for whatever audience segments are not already being targeted (eg. AEO with lookalikes in Germany, etc.). The advertiser covers their entire addressable market by breaking it out into campaign strategy-appropriate segments.