In over a decade managing Google Ads accounts for many clients in many industries, I've found that there are always similar questions that surface. The most consistent question I often receive, distilled down into one essential concept is:

'How can we forecast results for a different advertising investment?'.

Providing a great answer to this question has been my mission over the last 2 years.

Seasoned Google Ads professionals develop an instinct for answering questions like this. After spending time in an account, you get a feel for its potential.

Despite being able to provide an intuitive answer for what I felt was a good prediction of performance, I always came out short when trying to provide justification.

Budget Optimize is the tool I developed to solve this problem. It provides a visual plot of campaign trajectory & a mathematically based forecast for performance at different spends. It allows us to produce a solution & also see the reasoning behind it.

How can I forecast PPC spend accurately?

This question usually arises when taking over a new account but also over the years continues to pop up as the accounts mature. While client questions on this topic do range, they converge at a similar point:

• I have plenty of budget if we can make it work, can we hit ROI of X:1?
• What will happen to CPA if we increase the spend in the account by X?
• What should I really be spending in my account to get the most out of it?
• I need to pull back the budget, what will my CPA (or ROI) be if we drop budget by X%?

It's important to understand what's behind this question and who is asking it. These questions are high level, the person asking it is not looking at details, such as why a certain ad is written one way, or why this keyword is matched with that ad.

They are the questions being asked by decision-makers and key stakeholders, and this means that ultimately they are the questions that matter to a business. This is why it is such an important question to answer well, a well-grounded answer will impress those that matter &influence key decisions in the business.

At this point, how would you answer this question? A common approach is to extrapolate based on impression share. We could look at each campaign and estimate the change in spend & conversions.If we double impression share, we assume that spend and conversions will also double. This will provide a limited answer, but it assumes linear campaign performance as we increase investment which doesn't factor in diminishing returns.

We can see how this looks visually and how unrealistic it is for continued spend increases. In the example below, as we double impression share (I.S.) from 30% to 60%, we are assuming linear increases in spend & conversions, with both of these metrics doubling.

Image 1:Using Impression Share to estimate increased spend potential creates a linear projection that is unrealistic.

We need a better model which can more accurately map our campaigns and account for diminishing returns.

Visualising a Google Ads account

A better approach is to build a mathematical model based on past performance. The best way to explain this model is to visualise it on a graph.

Think of this as being able to look at yourGoogle Ads account from another dimension. We are used to looking at campaigns, impressions & clicks as well as keyword & ad performance.

These are all vital to understanding and optimising components of the account, but focusing on these metrics doesn't provide a holistic view of account trajectory.

Account trajectory is a new dimension that allows us to answer the underlying question. It provides us with a visual view of account performance which we can use to project & forecast performance.

Below is an example of how we can plot and visualise account trajectory.

The x-axis shows spend per day, and the y-axis shows conversion volume per day. Each dot is the spend and conversions for a particular day in a six month period. There are approximately 182 dots on the graph mapping each day in this six month period.

We can already get a feel of the account trajectory just by looking at the graph. We can see there is a relationship between spend and conversions. As spend increases, conversions increase at a particular rate.

It's the rate of change of this relationship, which we can describe mathematically that is our key insight.

Regression analysis

Regression analysis is a type of machine learning model that can represent this relationship between spend and conversions over this period mathematically. In the same example, we have now added a regression line to map this relationship.

The regression curve allows us to predict the corresponding conversions at different levels of spend. Furthermore, we can move beyond the limits of the graph, up to any daily spend using the regression formula of the curve.

The cost vs CPA dimension

We previously charted cost vs conversions. We can also chart another potentially more insightful dimension: cost vs CPA, where CPA (or ROI) is the measure of performance that matters most.

In the screenshot below, the graph on the left is a plot of cost vs conversions, while the graph on the right is the same account plotted with cost vs CPA.

The cost vs CPA graph shows us graphically that there is an optimal CPA point at approximately $1,400 spend per day, which is where CPA will be lowest. As we increase spending from that point, we can visualise how CPA starts to rise.

We can now see the account trajectory from two visual aspects (dimensions). Both regression graphs are useful to predict either conversions or CPA at different levels of spend. (or alternatively revenue & ROI). These are the metrics that matter, and we have a formula for forecasting them.

Understanding the potential of the account and answering the original question is now possible. It's no longer a guessing game or an intuition, now we can forecast based on a sound mathematical model grounded in past performance.

Budget Optimize value proposition

While regression analysis can be performed in Excel, Budget Optimize is able to add extra capabilities for superior analysis. The advantages include:

Fit different regression models: Different accounts have different trajectories, and therefore different regression models offer more accuracy. We look at r-squared and mean squared error as measures to autofit the best model and come up with the most accurate predictions.

The below example shows how different models are able to represent the relationship between cost & conversions. Some models more accurately reflect the trend than others.

What-if analysis: When we visualise an account, it makes it easy to see the point of optimal CPA or ROI. The tool is also able to work this out mathematically using what-if analysis

Advanced filtering: Running and rerunning these models is time consuming when you need to filter out certain account metrics and look at different campaign combinations. You might want to review non-brand campaigns only or change historical time periods or review only mobile campaigns. The tool makes this possible within seconds rather than manually taking hours.

Plot multiple regression lines (advanced): While not included in the current functionality, the tool provides regression formulas to allow you to plot lines and measure performance in graphing tools.

Remove outliers: Easily filter out outliers with a click. You might have had a sale day or some other unusual activity that skewed results. Budget Optimize allows you to filter out this skewed data by automatically detecting it.

Compare against actual results: Budget Optimize allows you to see actual results for the period against forecasted projections moving forward. Making it easy to compare and forecast on the same screen.

Limitations

Budget Optimize and regression analysis does not claim to provide a 100% accurate forecast. While we believe it is a sound method to predict performance, it's accuracy will vary for each account, and it only should only be seen to serve as a prediction.

The main limitation is that results are based on historical data. Things can happen in the future that are not accounted for in the historical data. Some examples include:

• Natural occurrences like a spate of bad weather. (This would be great if your business sells umbrellas)
• New changes to the account itself like a new account manager who is better than the previous one.
• Market-based changes like a new competitor entering or exiting.

In terms of seasonality, we recommend using data from a period that is similar to the period you are trying to forecast. Also, choose a time period that is long enough with enough data points. It is a balancing act to select the most accurate time period and have enough data to work with.

Solving the big question

As the famous saying goes, 'The only thing that is constant is change'. Google Ad accounts are dynamic, marketing budgets will change, and that is why clients always want to know what the predicted outcomes of a change in budget will be.

My answer is a mathematically based solution grounded in machine learning regression algorithms. While it does have limitations & shouldn't be relied upon for 100% accuracy, it is a sound approach to estimate future account performance.