How to Predict Customer Acquisition Costs

How do I know how much my customers will cost to acquire next month?

That’s the question that burns in every forecaster and marketer’s heart, because Customer Acquisition Cost (CAC) impacts everything.

If only CAC fluctuations weren’t so difficult to predict.

The number of variables that affect CAC, and the interactions between those variables, lead many marketers to simply throw in the towel and guess.

But if you’re wrong, and next month’s CAC is wildly off from your predictions, you’re looking at potentially devastating misses.

Guessing won’t get you there. So we set out to find a better way to answer that big, burning question.

And it starts with another question:

What factors cause CAC to rise and fall?

1. Spend Volume
2. Consumer Confidence
3. Past Performance

Factor 1: Spend Volume

Most experienced marketers know intuitively that CAC rises with an increase in spend and vice versa, but is that really true?

Turns out, the answer is a clear “yes.”

Our models found a reliable linear relationship between Total Ad Spend and Weighted CAC without accounting for other variables.

This is a good starting point for CAC modeling.

For this particular brand, every $10.5k of Spend increase correlates to a $1 rise in CAC.

Factor 2: Consumer Confidence

Most ecommerce data analysts have an enormous blind spot: the impact of macroeconomic signals on individual brand performance.

This makes sense: less confident consumers take more convincing, and that costs money.

Factor 3: Previous Month’s CAC

While individual ad performance changes in the long run it’s rare for the CAC associated with that ad to fluctuate wildly month-over-month:

Again, there’s an intuitive sense here: most brands don’t delete all of their ads and cancel all of their campaigns at the start of a new month, so the aggregate performance of an ad account will look pretty much the same on February 1st as it did on January 31st.

Bringing it Together: Predicting Aggregate CAC

The 3 factors above, plus an additional variable to account for seasonality (CAC drops in November and December in aggregate) give us a solid basis for CAC prediction.

We used that resulting model to predict CAC for November of 2022 and put it up on Twitter to hold us accountable to that prediction:

Well, we were 6.3% off, but it was a good learning experience. We said $31.34 for the monthly CAC and it came in at $33.46. That was still within the 90% confidence interval of $28.77 to $33.91.

But predicting aggregate CAC on its own is simply a cool trick … unless you can reliably predict CAC for your brand.

Making Short-Term, Brand-Specific CAC Predictions

Turns out, what works in the aggregate also works on an individual level … with some notable caveats.

To illustrate: At the end of last year, we modeled December CAC for a single brand using the information that was available to us in November — October CAC, October CCI, forecasted Spend, and an adjustment for seasonality.

Unlike the aggregate model, however, the results were not as accurate as we had hoped …

We predicted CAC to be $83.54 and it came in at $71.15.

We were off by 17% and outside of the 90% confidence interval.

It turns out, however, that the reason was fairly straightforward: Ad spend for that month was 20% under target — a huge miss.

If we’d used this figure in our model instead, our prediction would have been $76.09 - much closer to the actual value of $71.15 and within the confidence interval.

Making Long-Term, Brand-Specific CAC Predictions

While predicting next month’s CAC has its uses, meaningful predictions have to extend further than that.

To do so, consumer confidence and last month’s CAC are no longer relevant variables.

Instead, we’ll need to use historical data to predict seasonal fluctuations.

We produced this model for a Home Goods brand using data from both Statlas.io (and reported in our DTC Index) and historical data from the brand to make predictions for March, April, May, and June of 2023.

Here’s what we’re predicting:

In other words, we’re going out on a four-month limb, putting our model to the test in full view of the public.

Our promise to you: We’ll follow up at the end of each month to assess the quality of our predictions.

For more data insights check out our weekly data newsletter and on-demand data service, the DTC Index.