Measuring marginal organic contribution from paid marketing

One trap that’s common for mobile-first companies to find themselves in is believing that organic installs are truly the result of totally un-directed, spontaneous, aimless discovery. This is a trap because it’s unlikely; it’s rare for a truly static, stable baseline of organic installs to exist for a company that is independent of paid marketing.

Of course, exceptions do exist:

  1. Companies with prominent existing brands (either developed in the early days of mobile, eg. Candy Crush, Angry Birds, etc. or borrowed from desktop-era dominance eg. Facebook, YouTube, Netflix, etc.);
  2. Companies operating in red ocean categories that are tightly associated with household-name keywords (eg. Solitaire games, various utility apps);
  3. Companies that realized first- or second-mover advantage in a new mobile-first category and benefited from powerful earned media campaigns.

For these companies, paid traffic might not actually influence whatever baseline of organic traffic exists. People who work for companies with these profiles tend to very aggressively object to the idea that paid marketing is the primary driver of organic installs for their apps.

Organic baselines can be tested — the most brute force approach to establishing an organic baseline is to shut off paid marketing completely. If, when total marketing spend declines to $0, organic install volumes increase and total install volumes don’t change at all, then paid marketing was merely cannibalizing organic traffic. If total install volumes drop to 0, then organic traffic was totally dependent on paid marketing. What usually happens is that organic install volumes decrease to some level, which is the organic install baseline.

(Famously, Uber shut off all marketing spend when it suspected its agency partners of committing fraud and saw its total install volumes remain unchanged. I imagine that this wasn’t a very pleasant revelation for the marketing team.)

Shutting off marketing spend across the board isn’t practical for most companies, which intuitively understand that paid marketing is their primary source of revenue generation and thus can’t afford to stop marketing just to test an organic baseline. A company might choose a geographically representative but not commercially critical geography to use in organic baseline discovery with marketing cessation (eg. Canada if the company’s primary market is the United States), but this introduces potential problems around results applicability given historical levels of spend, total DAU, etc.

And marketing doesn’t need to be shut off completely: even the degree to which organic install volumes change when paid marketing levels do is valuable, and so a company might reduce marketing spend by 10-20% and observe changes in organic install levels. Again, if paid marketing decreases by 20% and organic install volumes increase to keep total install levels flat, the company learns something. These types of changes, monitored in isolation against all other channel settings — eg. increase the Facebook budget by 5% while leaving everything else the same — are how companies ultimately build media mix models.

There are innumerable permutations of this kind of testing: each comes with risks, caveats, and uncertainties. It’s folly to think that any specific test could reveal, clearly and categorically, an exact number for an organic baseline in a product and marketing dynamic as complex as virality; that’s impossible, and it’s the kind of uncertainty with which successful teams are comfortable. Descending into measurement rabbit holes with the goal of defining some ratio or metric absolutely is a horribly inefficient use of time and brainpower. Marketing teams are better off when they operate under the premise of continuous model refinement as their product and user base change over time.

And anyway, marginal organic contribution ratios are far more valuable than some abstract measure of baseline organic install volume. Even if one believes that their product benefits from a hidden yet constant and reliable current of organic installs, how does that help them? If those install volumes are truly static, then they likely won’t increase meaningfully over time, and the company will need to rely on paid marketing for growth, anyway. Which means that the company still needs to look to paid marketing to increase their user base and revenues.

Thinking about Organic installs through the perspective of marginal organic contribution ratios is helpful because it orients analysis around what the developer experiences at their current level and mix of spend. Marginal organic contribution ratios effectively describe how an additional dollar of spend on a given channel impacts volumes of Organic traffic — essentially, the degree to which spending produces incremental installs. Values of marginal organic contribution ratios can be:

  • Positive: Marginal organic contribution ratio > 0 means that every additional dollar spent on a channel contributes to organic install levels;
  • Zero: Marginal organic contribution ratio = 0 means that every additional dollar spent on a channel has no impact on organic install levels;
  • Negative: Marginal organic contribution ratio < 0 means that every additional dollar spent on a channel reduces organic install levels.

Putting this framework to practice is really just a matter of measuring levels of organic installs as a channel’s budget changes from moment to moment.

In the case where an advertiser has an existing portfolio of traffic, they can measure the organic contribution of a single channel at the margins by reducing spend on that channel and observing any attendant change in organic install volumes. Using Facebook as an example, when a channel has positive marginal organic contribution, ie. every new dollar spent at some level increases the amount of organic installs generated, then the number of organic installs would decrease as Facebook’s budget decreases (note that total installs decrease to a greater extent than Facebook installs decrease):

If Facebook’s marginal organic contribution ratio is 0, then organic installs would remain unchanged as Facebook spend decreases (note that total installs decrease by the same amount as Facebook installs decreases):

And where Facebook spend is actually cannibalizing organic installs — that is, Facebook ads are being clicked on by people that would have installed the app anyway — the number of organic installs actually increases as Facebook installs decrease (note that total installs remains constant as Facebook installs decreases):

The cannibalization case raises a new question: can the advertiser reduce Facebook spend to some level that doesn’t cannibalize organic installs? This can be tested through further decreases in Facebook spend to discover a level at which organic installs stop increasing. Extending the above example, imagine the advertiser reduces Facebook spend yet again to find that Organic installs still increase, albeit to a smaller degree than in the previous week:

In this case, Facebook installs decrease by 15,000 week-over-week, but Organic installs only increase by 7,500 (meaning that only half of the foregone Facebook installers would have otherwise installed the app). The advertiser might cut Facebook spend yet again to arrive at a level of budget that produces no marginal organic cannibalization:

Here, the advertiser has reduced Facebook spend and found that Organic installs remain exactly the same, meaning that at this level of spend, Facebook installs are incremental and do not cannibalize Organic traffic.

Note that these values can change over time, and they tend to be geographically-scoped based on historical levels of marketing spend and saturation. A sophisticated media mix model would also take into account every channel’s install volumes as a single channel’s budget was changed; the more changes being observed per period, the more sophisticated the model needs to be to accommodate those simultaneous changes. But this approach to measuring organic install volumes is more durable and philosophically defensible than trying to calculate a concrete organic baseline that can’t be attributed to paid media.

Photo by Anastasia Dulgier on Unsplash