A Stick in My MAU

How to Model Social Media Network Ad Revenue Opportunities: Mirror The Way Advertisers Buy and How Media Companies Sell

The week before last, Pinterest hosted an investor day which included the company’s overview of the factors they expect will drive their growth.   It was mostly very realistic although perhaps I’d quibble with the characterization of the total addressable market, or TAM for digital advertising in 2022. For Pinterest or any of its peers I’d exclude domestic activity in China, bringing the market size down from $550 billion to more like $430 billion (growth from Chinese advertisers spending in the US and elsewhere would be included in data for each individual market).  Arguably this distinction doesn’t matter: there is much more that will help to support Pinterest’s outcomes beyond the correct TAM, as the company described a wide range of efforts in useful detail during the presentation.

There was one slightly-more bothersome data item they referenced which over the past decade other social media companies and their observers have used with unnecessary frequency: it’s ARPU, or average revenue per user.   ARPUs might make for interesting short-hand comparisons across seemingly similar companies, but I suggest the number is not very meaningful when it comes to anticipating a company’s growth opportunity from advertising.

Let me explain why. 

For starters, I think that philosophically a financial model should be an abstraction of reality.  It should roughly mimic, in a simplified way, the mechanical ways the world works.

ARPUs are absolutely the correct variable to focus on for subscription businesses, where the unit of decision is the individual consumer.  Consumers – individual users or groups of them with or without intermediaries – essentially decide to buy the product or not, and pay more for different tiers of services.  Thus, an ARPU is an important unit of analysis in cable and other consumer-focused telecommunications businesses.

So when it comes to forecasting advertising revenue, why would it make sense to look at ARPU? 

A Media Owner Grows When Marketers Allocate More Budget Share to Them, Or When More Advertisers Become Customers

By contrast, no advertiser sets their budgets on the basis of MAUs or users, and few do so at a global level.  Instead, large advertisers generally establish marketing goals, determine plans to best meet those goals, allocate budgets towards groups of companies that are managed in similar ways, and then optimize across them when they can or within them when they can’t.   They look to diversify spending across multiple partners if they are able to, and subjective factors often play a huge role in making decisions to allocate budgets, even for performance-based marketers who can torture their data enough that it will tell them what they want it to.  Moreover, many of their budgets are based on incremental changes from year-to-year, given the mechanical and relationship issues involved in making big changes. Smaller advertisers are in some ways simpler to model: for them, their spending will mostly go to Alphabet, Meta and Amazon, and possibly only one or two of the three, depending on resources available to manage these investments.   

Whatever the focus in terms of small or large, a company’s ad sales force – the people generating revenue – aren’t able to impact how users grow very well, but they are able to manage revenue per advertiser and to find new segments of advertisers to sell to, much as Pinterest explained that it is doing.

At best a MAU is a measure of potential reach (assuming the figure isn’t overstated), which is useful, to be sure.   But to use MAUs or ARPUs as a basis for anticipating ad revenue is like saying that because the National Geographic channel has as many paying subscribers as ESPN that ESPN’s ad revenue per subscriber should be compared in some way to NatGeo’s.  

Instead, I suggest that what should really matter when comparing similar properties is aggregated time spent by consumers – that is to say time spent by the average consumer who uses a property multiplied by the number of unique consumers – on a channel for a given amount of reach. 

Time Share Leads To Money Share

“Time spent,” as I’m defining it here, is going to be most important for many reasons.  Over time we would expect that roughly comparable platforms will adopt roughly similar commercial tactics in terms of ad loads, methods for segmenting advertisers, sales processes, etc.  We see this in television today and we’ll probably see it more commonly within comparable groups of digital media in the future.

True, there may be some unique advantages that some platforms possess, and to be sure, if a platform is fundamentally different – say, Amazon with its retail media and e-commerce focus or YouTube with its online video focus – the comparison won’t make much sense, especially as a given marketer likely budgets for it through a different pool of spending, anyways.

When they are truly different, time spent across different kinds digital platforms should no more be compared to each other than radio, newspapers or television time should be compared to digital, even if there were such a thing as accurate cross-media time measurement surveys. 

But within any given broadly similar set of media owners or platforms, I think that time-spent share within a given country is possibly the best variable we should be focusing on.   From there, we can spend our time focusing on growth of the medium within each country to derive revenue growth expectations for any given company. 

More Advertisers Also = More Money

There is one other simple way that I think can appropriately reflect the ways media companies generate revenues: to focus on identifying the size of different cohorts of advertisers, and then to look at how the average marketer within a given cohort spends.  As described above, at the most mechanical level, this mirrors how sellers of advertising pursue growth. If we know that a given platform has, say, 2,000 large advertisers with an average budget of $1 million each and 20,000 mid-sized advertisers with an average budget of $100,000 each, and the company manages its sales through a large brand sales channel and a middle market sales channel, how the company expands the number of advertisers in each group and increases average spend per company in those cohorts, we’d also have a superior way to model growth relative to looking at MAUs.    Here, the MAU we should care about is the monthly active advertiser.  Although some companies such as Criteo helpfully provide such variables on a regular basis, few others do.

What’s a Realistic Approach?

Putting aside that data provided by companies may have certain flaws and limitations, I believe it would be helpful if more companies recognized that providing data related to time spent or data related to the number of customers by cohorts on a regular basis might be received positively by a company’s stakeholders.  Greater attention on these kinds of data points from the outside world would probably also help focus everyone inside a company on the metrics that truly matter.

Of course, it’s unrealistic to expect widespread change in the near-term, so what else can those of us looking to model the future of a social network do?

In Lieu Of Better Data, Focus on Qualitative Factors Driving Ad Revenue Share Into The Future

While the long-term opportunity and “fair share” of advertiser spending might best be measured by share of time within a country, at a minimum we can still look at the qualitative factors causing platforms to grow or lose share of consumer time on a relative basis among the most directly comparable peer within a country.  From here we should be able to identify if we think a platform will over- or under-perform in terms of the share it takes.    Alternately, if we have enough clues about some of the relevant data points to start with, we can try to estimate the revenues per advertiser within separate groups, and then model growth per advertiser over time.  

Although either approach is highly imperfect, each at least satisfies the basic principle of modeling that I think is worth repeating here: the driver of a model has to relate to the ways the entity being studied actually operates.   Guessing share of time to share of revenue ratios or growth rates per segments of advertiser will likely be far more accurate than modelling revenues derived from APRUs we’ve been given.