This is the subhead for the blog post

Every marketer – or at least the boss of every marketer – wants to know how much tangible revenue translates from their ad spend. Sure, web traffic or brand awareness is extremely important, but sometimes clients just want to know how X dollars lead to Y sales.

This simple question is a major source of tension between marketing teams and CEOs, particularly because it has a complicated answer. Marketing ROI depends on a host of different factors like channel mix, seasonality, offline marketing events, latency, virality, and a lot of other lurking variables. All of this makes it very difficult to define marketing profitability with a simple causal statement.

Benefits to Clients

CPA Projections use multivariate analysis to calculate the range of possible revenues given a range of ad spends. It controls for factors such as seasonality and latency with dummy variables and raw data modification — all to help inform marketers how much they can earn through ad spend on key marketing channels.
And while CPA Projections are not full-blown forecasts (which we can also do), they do provide helpful context on questions like “What happens when I push budgets to the limit?” or “When do diminishing returns kick in on this channel?”

Figure 1: This graph shows the range of possible Clicks (green) and Conversions (blue) given ad spends between $100-$600 on a specific marketing channel. The dotted lines are the upper and lower bounds of the prediction with a 95% confidence interval.

CPA Projections can also help marketing teams manage their bidding strategies more effectively. Once we predict the number of conversions across a range of spends, we can use a similar model to find the click levels needed to hit those goals, giving you your optimal Cost-Per-Click (CPC).

For instance, if the client wants to generate $50,000 in revenue over the next two months, the CPA Projection tells you how what Cost-Per-Click (CPC) levels you need to maintain to achieve that goal.

Finally, the model can also predict spend ranges that the client has never tried before. Want to do a spend surge on PPC non-brand? Here’s what you might see if you do it – though don’t expect miracles when it comes to Hail Mary strategies.

What the Client Needs

So what does a client need for a CPA Projection?

As is always the case in data science: the more data, the better. Having rich historical data that extends beyond a year is useful to model out seasonal events such as holidays or ‘hot’ and ‘cold’ quarters.

It also enhances the model to have Lifetime Value (LTV), or the estimated total revenue of an individual customer. Having this dimension can better capture the longer-term aspects of a product, like subscription sales or customer loyalty. If that’s not handy, we can rely on the conversion of your choosing.

Conversion or sales volume for each day is also important. Conversions that happen infrequently means the model will suffer from zero lower bound problems (conversion levels that are almost indistinguishable from zero). Volume creates the variance needed for a model to pick up on trends within the data.

Finally, supplying us with as much offline ad intel is crucial. For example, if the client runs a massive TV ad campaign in August, we need to know that to control for unexpected surges in income. When it comes to CPA Projections, context is extremely important.

How Do I Get Started?

Not a 3Q client? Reach out to the 3Q Decisions Sciences team here. Already working with 3Q? Let your account team know and we’ll see how we can help.
A final note: these models aren’t magic wants, but they are easy to interpret and can be built across channels, countries, and conversion types of all shapes and sizes.