TikTok’s outage and the digital marketing efficient frontier

A new draft paper from researchers at Columbia Business School finds that the average CPM across Meta’s properties increased by roughly 10% in the US when TikTok went offline ahead of its ban in January. The authors of the paper use a difference-in-differences approach to estimate the increase in US-based advertising prices resulting from TikTok’s inaccessibility for both advertisers and users for roughly 14 hours starting on January 18th. From the paper (emphasis mine):

While daily spend on active ads increased by 12.6% on average after TikTok went dark, there is no significant change in the number of impressions generated. The coefficient on daily impressions is positive (2.7%), suggesting a small yet not significant increase. These results indicate that while demand for ads grew, the supply of impressions did not sufficiently react to meet this demand … On average, the day after the shutdown, CPM on ads targeting the US increased almost 10% relative to ads in other countries. Considering the average pre-ban CPM of almost 21 USD in the US, the above estimate implies an absolute increase of about 2 USD … Figure 7 shows that the increase does last over time, as the CPM is significantly larger over the week after the shutdown, indicating an average daily absolute increase of about 1.5 USD. The figure also provides evidence in favor of the parallel trends assumption, as the changes in CPM of ads in the US vs. other countries were not significantly different before the outage.

The paper is fairly short and worth reading in its entirety. While the Meta CPM effect is likely the paper’s principal headline, another noteworthy observation is that larger advertisers increased both their Meta spend and the number of ads deployed to a greater degree than did smaller advertisers. The paper defines “larger” advertisers as those that spend more than median amount in a country and “smaller” advertisers as those that spend less, where “new” advertisers had not spent at all on Meta in the two months prior to the TikTok outage.

The paper’s model — which includes advertiser, country, and day fixed effects to control for time-invariant correlated error — finds that larger advertisers increased their daily ad spend by 68.4% and the number of ads deployed by 20.8% following the TikTok outage. This compares to increased spend of 26.3% and the number of ads deployed of 7.2% for smaller advertisers. In other words, larger advertisers were able to react more quickly than smaller advertisers in shifting budget to Meta, but in terms of budget and the volume of new creatives launched.

On its face, this dynamic seems obvious: larger advertisers, by definition, had allocated more resources to Meta prior to the TikTok ban than smaller advertisers and should have benefited from more robust measurement and creative production processes for that channel. Larger advertisers would be more prepared, given established infrastructure, to generate new Meta-specific creative (or repurpose existing TikTok creative) to expand budget on Meta; this is simply a function of scale.

But what the change in spending patterns highlights is that these advertisers had room to scale on Meta in the first place. This would contradict popular wisdom about performance marketing budget allocation, which is that advertisers maximize their spend on all channels on the basis of ROAS. This is to say: performance advertisers are constrained by ROAS, not hard budget limits, and they spend however much money they can on every live channel to fulfill those ROAS targets. This can be thought of as something like a digital marketing efficient frontier: that the budget allocation within an advertiser’s portfolio of channels seeks to maximize total, absolute revenue given some ROAS constraint, and that any advertiser’s combination of channel-level budgets can be assumed to be optimized for maximum absolute revenue as an objective function given that constraint.

If large advertisers could swiftly allocate additional budget to Meta, it implies that they either relaxed their ROAS targets to maintain total spend or that they weren’t spending as much as they could have on Meta to begin with. The outcomes observed in the paper suggest that smaller advertisers adhere to the efficient frontier more than larger advertisers, with little ability to incrementally scale spend on any channel given their ROAS constraints. This makes intuitive sense: smaller advertisers are almost certainly more sensitive to short-term ROAS deterioration. The effect would also be less pronounced for smaller advertisers if they saw Meta as a foundational channel that they needed to fully exploit before diversifying or specializing elsewhere simply given Meta’s reach and ease of use. Put another way: Meta might be the first channel that advertisers scale spend with before discovering idiosyncratic, product-specific efficiencies with other channels like TikTok. In cases where that’s true, a small advertiser would have very little “slack capacity” on Meta to grow spend.

I speak to this idea in Opportunity cost and diminishing returns in user acquisition. From that piece:

Until the very largest channels by ad spend showcase negative marginal returns — that is, until the next dollar spent is unprofitable — it often doesn’t make sense to diversify outside of those channels. As uncomfortable as it might seem, most user acquisition teams would be better off operating only on the largest channels — concentrating their spend, developing channel specialization, and minimizing their analytical overhead — versus having their team’s attention and time spread across a wider array of channels with a longer-tailed distribution of ad spend and revenue.

The broader idea here is that the channel diversification and specialization is a luxury that emerges with scale: each additional channel incurs overhead that can serve as a drag on efficiency, and onboarding new channels might not be economically rational until some level of saturation is reached on the channels with the most budget capacity.

This would imply that adherence to the digital marketing efficient frontier is a function of advertiser scale: at lower levels of spend, advertisers are strictly bound by ROAS targets and maximize spend across their portfolio of channels. But as they grow their overall budgets, advertisers gain ROAS flexibility (at least in the short term) and are potentially budget constrained in an absolute dollar sense (thereby creating slack capacity in certain channels), but can be more responsive to exogenous shocks because of team or tool resource availability.