Note: this post is adapted from content from the upcoming Modern Mobile Marketing at Scale workshop series taking place in NYC, SF, and London in October. More information here; places are still available for the NYC and SF workshops.
One of the more confusing aspects of Facebook advertising is cost control: how the various bid strategies impact ad delivery and goal costs. This confusion has been compounded by the surge in popularity of the App Event Optimization (AEO) and Value Optimization (VO) campaign types on Facebook, which, on the surface, appear to do the same thing (optimize campaign targeting around in-app behaviors) but actually operate in fundamentally different ways. In this article, I’ll attempt to highlight the differences between Facebook’s various cost controls, as well as explain how those differences should be managed across the three mobile campaign optimization strategies: AEO, VO, and Mobile App Installs (MAI).
(Before reading: this QuantMar answer about how mobile ad networks function serves as a useful primer for this post. It’s helpful to understand the background of mobile ad delivery optimization before thinking about cost controls.)
Back in April, Facebook introduced a new manual bid strategy called Cost Cap to complement its two existing manual bid strategies: Bid Cap and Target Cost. The Bid Cap strategy sets a maximum absolute bid for a cost per action (eg. app install, landing page view, add to cart, complete tutorial, etc.) in a campaign, whereas the Target Cost strategy establishes an average price to which completed actions should resolve. The purpose of the Bid Cap strategy is to allow an advertiser to maximize volume of optimization events (since Facebook would operate within the entire budget range to find potential optimization events but set a hard bid limit, potentially resulting in volatile per-event prices), and the purpose of the Target Cost strategy is to allow the advertiser to achieve a stable average price per event over time (Facebook would look within a range above and below the target cost for conversion events).
The Cost Cap strategy is somewhat of a hybrid approach between Bid Cap and Target Cost that leans into Facebook’s push into algorithmic, event-based campaign setup: it combines the full-range of price bidding with the focus on a target cost that might include bids above the cost cap to increase volume while also attempting to stabilize prices. This page in Facebook’s documentation provides a good overview of the various bid strategies and when they should be applied:
The crux of the differentiation between these cost controls is that a bid isn’t the same thing as a cost, and an advertiser must decide if it wants to exert control over its bids as it participates in impression auctions. When an advertiser sets a hard bid cap, they are telling Facebook, “I’ll pay up to this amount for an optimization event, but no more.” When an advertiser sets a cost cap or target cost, an advertiser tells Facebook, “This is the business result I want to achieve; you determine the best bid to use in auctions in order to manifest that.” Because Facebook operates auctions with a second-price model, the bid for an impression is almost always higher than the price paid by the winner.
These bid strategies (as well as the Lowest Cost strategy, sometimes called Autobid) apply to mobile app install and app event optimization campaigns — but they don’t apply to value optimization campaigns, and the analytical and operational requirements for scaling VO campaigns are thus different than for AEO and MAI campaigns.
MAI campaigns are designed to maximize install volume to some price standard; AEO campaigns are designed to maximize a volume of in-app optimization events to some price standard. Installs and events are really the same thing here, they just exist at different parts of the funnel: they are discrete, one-time, binary outcomes that a user either completes or doesn’t.
VO campaigns are designed to do something different: they are designed to maximize the magnitude of some series of events — purchases or ad views — that a user completes. The difference between magnitude and binary completion is substantial, and advertisers must think about these campaign purposes differently. In a VO campaign, Facebook is targeting users that will deliver the maximum possible value to an advertiser as evaluated in purchase (or ad view) postbacks, and thus it can’t constrain costs with an absolute per-event bid or target cost: it must calibrate campaign delivery as a function of return on ad spend since there’s no discrete event that can be bid against.
This is why the bid strategies available for VO campaigns are Highest Value and min-ROAS. In the Highest Value bid strategy, Facebook pursues the users that will generate the most “value” without any concern over cost, meaning Facebook will set the bid in the auction dynamically with no cap on what it is willing to spend other than the overall budget. In the min-ROAS strategy, Facebook adjusts its bid on the basis of what the advertiser wants to recoup from users in either one or seven days after the install.
This is somewhat similar to the cost cap bid strategy except that it is dynamic at the level of the user: Facebook is still evaluating that estimated series of value events independently for each user (because the magnitude is the determining factor), and then it is applying a dynamic cost cap which backs into a bid by using the min-ROAS value. For instance, if an advertiser sets the min-ROAS for Day 1 to be 40% for a VO campaign, and Facebook calculates the expected value for a given user to be $10 at Day 1, then Facebook can set the bid for that impression to be whatever it needs to be to achieve a $25 cost per install ($10 / 0.45 = $25).
(Note here again that the bid is different from the cost: the bid might be $50 to achieve a $25 cost because the advertiser will pay one penny more than the second-highest bid.)
An advertiser must deeply understand its ROAS curve in order to scale VO campaigns efficiently. This is not the same thing as understanding per-segment LTV: two ROAS curves could generate the same time-bound LTV at Day 90 and produce completely different outcomes in VO campaigns, as per above. Producing a ROAS curve is wholly different from calculating an LTV for a user segment; it’s also the only lever that advertisers have in constraining costs for VO campaigns.