 The incremental CPA is still an average - it contains both, the conversions you acquired cheaply (seen at the bottom of the curve) and the conversions that got more expense (seen at the upper part of the curve). Therefore, depending on the question you need to answer - this CPA might not be the right metric to look at.

This metric makes sense to calculate on both levels: on an overall level (taking into account all sources / channels / campaigns) as well as on an individual media channel level (source, campaign). Note that on an individual channel level the terms CPA and Incremental CPA are often used interchangeably. If the conversions attributed to a channel are considered incremental, you might just say CPA.

Marginal CPA (MCPA)

The marginal CPA answers the question “at the current / a given level of spend, how expensive is the next incremental conversion?”. This mostly makes sense to calculate for incremental channels on a per source / channel level. It takes into account the diminishing returns to scale curve (the saturation curve) and tells you from where you are on that curve, how far you need to go (how much money you need to spend) to get an extra conversion. This metric is extremely useful to see a) what channels / sources are already maxed out and where you want to spend your next dollar.

The calculation is Where “delta spend” is the extra amount of money you spend on top of what you are spending currently and “delta conversions” is the extra conversions you’ll be getting in return. Typically, to calculate it,

− delta spend is fixed at 1, 5, or 10% of current spend, then arriving at the next spend level → spend_new

− predicted conversions are obtained (conversions_new) at spend_new by plugging in spend_new into the saturation curve

− calculate delta_spend as (spend_new-spend) and delta_conversions as (conversions_new-conversions)

− now we can obtain the mCPA by plugging delta_spend and delta_conversions into the formula above

To sum it up:

CPA: All cost divided by all conversions → Gives an insight into the overall and high level effectiveness of all marketing and branding activities.

ICPA: All cost divided by incremental conversions → Gives an insight into the how expensive a customer was acquired through (paid) marketing activities

MCPA: Marginal cost divided by marginal conversions → Gives an insight into how much you must spend to acquire the next customer through a certain source

The relationship in almost all the cases is CPA≤ICPA≤MCPA. Why MCPA is always higher than CPA of a source?

This is an effect due to the concave shape of the saturation function - “diminishing returns to scale behaviour”.

This basically means that the model is considering a saturation behaviour where you get less back the more you spend as your saturation rises.