LTV Series

CAC payback period

Is this a feasible alternative to the LTV/CAC metric?

Paul Levchuk
7 min readJun 2, 2024

In the previous post, I showed you a few cases in which having LTV at hand can help you improve your marketing decisions. The only tricky moment is how to project LTV precisely.

LTV projection precision is the most common concern when people start working with LTV.

There are a lot of reasons why getting reliable LTV projection is a very complex task:

  • projection is always based on past data, but past != future
  • customer lifetime is very sensitive to market dynamics (economic trends, competition)
  • customers are sensitive to changes within the business (prices/discounts, loyalty programs, customer service)
  • customer behavior is not fully predictable as human beings can behave just irrationally
  • the longer the LTV projection time horizon — the more volatile the above-mentioned factors

That’s why marketers are looking for ways to decrease LTV uncertainty and probably find some alternatives. CAC payback is one such alternative.

What is the CAC payback period in a nutshell?

The CAC payback period is equal to the number of periods (days/weeks/months) when [LTV to Date] < [CAC].

CAC payback period for Paid channel.

From the chart above we can learn that, on average, we should wait up to 9 periods (1 period = 30 days) to pay back our marketing investments in User Acquisition. Only after this, we can earn money from the cohort of previously acquired users.

Probably attentive readers noticed that I said: “on average”.

When we are talking about marketing metrics almost in all cases it is recommended to calculate a metric at least on the campaign level.

Below is the table with the [LTV / CAC] metric sliced by campaigns. When the metric <1 it means that the campaign is still in the process of paying back.

CAC payback period by campaign.

As we can see, the situation is very different for different Paid campaigns:

  • 4 campaigns (13, 15, 16, 18) have CAC payback ≤ 3 periods
  • 3 campaigns (14, 19, 20) have CAC payback ≥ 5 and ≤ 9 periods
  • 3 campaigns (17, 21, 22) have CAC payback ≥ 12 and ≤ 24 periods
  • 2 campaigns (23, 24) haven’t paid back at all during the predefined LTV time horizon = 36 periods (see the previous post for more details)

In short, just 7 of 12 Paid campaigns have the CAC payback period below the average.

So, using averages guarantees you almost nothing. Moreover, using external benchmarks to assess your CAC payback period has the same drawback.

I recommend talking to the CFO to figure out which financial leverage your company has that enables you to invest in acquiring new users without waiting for user cohorts to pay themselves back. This exercise will be much more useful than striving to achieve some external averaged benchmark.

Actual CAC payback period vs proxy CAC payback period

So far to figure out the CAC payback period we used [LTV to date]. It means that we calculated the actual LTV at each period and then compared it to the costs we spent on the cohort of users.

However, it doesn’t make a lot of sense from a practical standpoint. We don’t want to wait for 36 periods to figure out that the specific campaigns haven’t paid back.

What if we don’t wait for a long period to calculate actual LTV to date, but instead use the cohort revenue for the 1st period (that is the first 30 days) as a proxy?

CAC payback period vs CAC payback period 30d (sorted by [CAC payback period 30d] ASC).

Is [CAC payback period 30d] = [Costs] / [revenue_30d] a good proxy?

Let’s build a scatter plot and check the relationship between [CAC payback period 30d] and [CAC payback period] metrics.

y ~ x : [CAC payback period] ~ [CAC payback period 30d].

From the chart above we can learn:

  • if [CAC payback period 30d] > 4, it tells us that it is unlikely a campaign will have a payback
  • if [CAC payback period 30d] < 2.5 the relationship is a bit non-linear

Nevertheless, it turned out that even the linear regression model has a good fit, where R2 is ~ 0.86. Not bad.

Is this result useful for us?

  • On one hand, we figured out the relationship between proxy metric [CAC payback period 30d] and real [CAC payback period]. We can predict the real [CAC payback period] after 30 days using our proxy metric.
  • On the other hand, to figure out the real [CAC payback period] we need to wait while each Paid campaign generates enough revenue to recover costs. For some campaigns, it takes so long time.

So, while this result is interesting it’s not practical.

Actual CAC payback period vs LTV / CAC

Probably we should have started with the following question: whether there is a strong relationship between the [CAC payback period] and [LTV / CAC] in general?

If so, then it likely will make sense to use [CAC payback period] as a proxy for [LTV / CAC] or marketing investment efficiency.

Let’s build a scatter plot and check the relationship between [CAC payback period] and [LTV / CAC] metrics.

y ~ x : [LTV / CAC] ~ [CAC payback period].

In general, there is a negative relation between [CAC payback period] and [LTV / CAC].

In other words, the longer the payback period — the lower the LTV / CAC ratio. It makes a lot of sense to me.

As we have fixed the LTV time horizon to 36 months, the longer it takes for a campaign to pay the money back, the less time a campaign has to earn additional money until it reaches the 36th period.

The relationship between [CAC payback period] and [LTV / CAC] is non-trivial, at least it’s not linear. Moreover, to learn regression coefficients you still need to have the real [LTV to Date] to calculate the [LTV / CAC] ratio.

So, it’s returned us again to the point that we need to have reliable LTV projection or wait for a long period for actual [LTV to date].

There is also another critical moment.

Although both linear regressions on the charts above have a very similar R2, this should not create a false sense of accuracy:

  • [CAC payback period] is measured in 30-day periods. From the data, we learned that it varies from 1 to 36. If for some campaigns your estimate of the [CAC payback period] will be shorter by 1 period it’s not critical.
  • [LTV / CAC] varies on a very different scale: from 0.6 up to 3.2. If for some campaigns your estimate of the [LTV/ CAC] will be smaller by 1 it could completely change whether we will consider a specific campaign successful or not. This is a big issue, for example, for 2 campaigns (15 and 18) that I colored orange on the chart above. They have the same [CAC payback period] = 3, but their [LTV / CAC] are very different (2.7 vs 1.6).

So, short payback periods (≤3 periods) tell us little about what LTV / CAC we will get from such campaigns as the spread is really large.

payback_30d WOE as a better alternative

The idea that I will share with you in a moment is related to the Information Value approach used for binary classification.

Instead of projecting LTV for a long period, we can do the following:

  • calculate the share of [revenue_30d] for each campaign
  • calculate the share of [Costs] for each campaign
  • calculate the Weight of Evidence (WOE) coefficient =
    LOG( DIVIDE( _rev_30d_share, _cost_share ), 2 )

The result could look like this:

payback_30d WOE.

The main trick is to compare the performance of each Paid campaign to the overall performance of the Paid channel.

This approach automatically separates your campaigns into 2 groups:

  • above the average performance (WOE > 0)
  • below the average performance (WOE < 0)

Also, it gives you a sense of how the best/worst campaigns could look like within each group. This information might help you define the optimization strategy and shape the range of possible results you can expect from optimizing your marketing campaigns.

Please note, there is no issue with campaigns 15 and 18 which have the same CAC payback period = 3. Their WOEs are quite different: 0.45 vs 0.26.

Beyond the practical aspects mentioned above, [payback_30d WOE] also better explains the [LTV / CAC] metric.

y ~ x : [LTV / CAC] ~ [payback_30d WOE].

If we compare the two last scatter plots, we can learn that:

  • R2 for regression [LTV / CAC] ~ [payback_30d_WOE] = ~0.85, while
  • R2 for regression [LTV / CAC] ~ [CAC payback period] = ~0.79.

Taking into account that the metric [CAC payback period] is based on the whole 36 periods, and [payback_30d_WOE] is based only on one period, I value this approach and use it to make quick (and directionally correct) decisions.

SUMMARY:

  1. LTV projection is a very tricky task.
  2. [CAC payback period] can only approximate the actual [LTV / CAC]. The relationship is not linear, and the margin of error for campaigns with short payback periods could be very large.
  3. [payback_30d WOE] is a better alternative than the [CAC payback period].

In the next post, I will do a crash test of the [CAC payback period] benchmarks.

--

--

Paul Levchuk

Leverage data to optimize customer lifecycle (acquisition, engagement, retention). Follow for insights!