Payback window analysis is a complex subset of digital marketing strategy that often goes under-resourced. Marketing payback windows dictate cash flow: the period over which marketing spend is recovered is fundamentally critical to the health of the business. Yet it’s common for me to see companies arbitrarily set payback requirements for their marketing campaigns — and to arbitrarily extend payback windows when marketing performance begins to degrade.
When a marketing team sees its advertising performance diminish, it considers two options:
- Increase marketing efficiency. Find ways to improve the marketing funnel such that performance improves and marketing spend can increase;
- Extend the marketing payback window. Accept a longer payback timeline on marketing spend such that the budget can be increased without addressing underlying performance issues.
Option 1 is, of course, the ongoing, permanent obligation of a digital marketing team. As I discuss in Mobile ad creative: how to produce and deploy advertising creative at scale, I consider marketing for digital products to be an exercise in damage control: as the product ages and the audience becomes saturated, the team, on a medium-term (eg. months) timescale, is likely just trying to maintain performance as a best case scenario. After the product has been live for long enough, merely keeping marketing performance stable is success. Sisyphus would feel at home in a monthly user acquisition review meeting.
And thus the allure of extending the payback window: if marketing performance has been structurally compromised as a result of audience saturation, or of changing market conditions (eg. a new competitor entered the market and is spending aggressively on advertising), or of any myriad factors that could impair advertising efficiency, then simply aiming for a new, later-stage point on the LTV curve creates the opportunity to increase marketing spend.
Imagine an average cumulative monetization curve that is fit across 180 days to a value of ca. $1.20 based on historical actuals.
Projecting the fitted curve to Day 365 yields a value of $1.37 — an increase of 13% on the Day 180 cumulative monetization value. If the marketing team decides to increase the payback window to 365 days (from 180), it can immediately increase its bids by 13% in order to break even on spend.
This line of logic is problematic, for a few reasons.
The data driving the forecast is, by definition, old. The youngest cohorts informing the portion of the curve between Day 180 and Day 365 are at least 180 to 365 days old. The market for the product could have fundamentally changed in that time;
Relatively little data is available to use in building later-stage parts of the curve. Consider a product with a retention curve as below:
The Day 365 retention value for the projection ends up at 8.67%, meaning that, from a cohort of 1,000 new users, 87 would remain in the product by Day 365:
In How much data is needed to predict LTV?, I exemplified the problem with projecting LTV estimates out to endpoints: retention curves reduce the amount of data available for any given cohort age, and it can take a very long time to accumulate the requisite data needed to make valid forecasts at various timelines.
And the problem with using sparse data is apparent from the increasing scope of the confidence intervals in the graphs from the above post: as data becomes increasingly sparse, so too might the predictive power of the curve become weaker. The increase in error margin for later-stage predictions, resulting from decreasing volumes of data points, can eat into the bid gains assumed from extending the frontier of the cumulative monetization curve.
In It’s time to retire the LTV metric, I propose an incremental, ROAS-based methodology for setting bid values. Extending a marketing payback window is a consequential decision given its impact on cash scheduling. Availability of data should be a primary determinant of cumulative monetization forecasting; extending the payback window should be done on the basis of having aggregated enough data to do so and not because marketing performance doesn’t support some level of budget.
Photo by Diego Jimenez on Unsplash
Recent