Forecasting Alphabet's Revenue

In this post we'll forecast Alphabet's revenue for the next four quarters from June 2025 to March 2026. Alphabet is one of the world's most valuable companies with over 2 trillion in market capitalization, so the stakes for creating a dependable forecast are high. Now with Datamago, the forecasting process doesn't have to be difficult or complex. Our objective is to demonstrate how easy forecasting this type of financial data can be to generate a clear outlook for your own organization and effectively prepare for the future. The dataset we'll be using can be downloaded here.

Forecasting revenue

We'll start by creating a univariate forecast using only revenue. After submitting the file, we see the following forecast a few seconds later:

The forecast achieved a 'good' performance score (for advanced users, the mean absolute percentage error on the validation set of the most 10 recent quarters was 3.27%), predicting 398.474 billion in revenue from June 2025 to March 2026. Given its low error, the forecast could be presented to stakeholders as long as you're comfortable assuming that the underlying conditions that influenced revenue in the past will remain steady in the future. However, if you prefer either more accuracy or more fine-grained control over the future conditions that affect revenue, adding a leading indicator or variable will likely help.

Finding leading indicators

The first thing we'll do is check for leading indicators that can help predict revenue. A leading indicator is an internal or external variable that has a lagged effect on what you're trying to predict. For example, marketing dollars spent today might boost revenue next quarter, while broader economic factors like consumer credit or government spending might trickle down to the bottom line several quarters from now. This makes leading indicators especially powerful because real data from events that have already occurred are guiding the forecast.

When finding leading indicators, we're presented with the options in the screenshot below. We'll leave the default settings while making sure the 'include open data' toggle is on to automatically search hundreds of thousands of variables from open datasets curated by Datamago covering economics, energy, government, and more.

The results are then shown in a table broken down by leading indicator name, lead time, correlation with revenue, and correlation with the residuals. We'll sort the results by correlation with the residuals as those leading indicators are more likely to fill in missing information and improve the forecast. We don't see any significant correlations of .60 or above so we'll use the first open data variable that makes the most sense, which is 'Quarterly Disposable Personal Income per Capita in the United States' (highlighted below) with a lead time of 1 quarter and a correlation of .98, meaning disposable personal income from this quarter is 98% correlated with Alphabet's revenue the following quarter.

The resulting revenue forecast after adding the leading indicator achieved an 'excellent' performance score (1.91% mean absolute percentage error, an improvement of a little over 40%), predicting 402.285 billion in revenue.

We also need to take a look at the leading indicator's forecast. When the lead time is equal to or greater than the forecast window, we know that the leading indicator's 'forecast' is actually real data that's been shifted forward by the lead time. But in this case, the lead time is only 1 quarter, so Datamago automatically generates the next 3 predictions of the 4 quarter forecast window. As seen in the screenshot below, the leading indicator's forecast consists of two colors: the green line representing actual data with the yellow line representing Datamago's auto-generated predictions.

The quarterly disposable personal income forecast achieved an 'excellent' performance score (1.03% mean absolute percentage error on the validation set of the most recent 10 quarters), so we can be confident that it's being modeled correctly. Taking into account the fact that the revenue forecast significantly improved after adding it, we can be confident that it's a good addition.

One thing to note is that the leading indicator's auto-generated forecast shows strong continued growth in the future, which we may or may not want to assume. If we wanted to assume more modest growth and see how that affects forecasted revenue, we can either click the 'adjust forecast' button above the chart, or we can test what the new forecast would look like by clicking the 'Scenario testing' section in the left-hand menu. We'll try the latter: after clicking the 'New scenario' button, a modal appears allowing us to adjust future values either by clicking or dragging them. Here's what that looks like after lowering the forecasted trend line a bit:

After generating the scenario, we see an adjusted revenue forecast of 397.045 billion, 1.3% less than the original (the grey line):

Including marketing and sales spend

Next, we'll create another forecast with default settings, but this time we'll include marketing and sales spend that we extracted from quarterly reports. This variable is included in the file that can be downloaded at the beginning of this post. Here's the resulting revenue forecast:

And the auto-generated sales and marketing forecast:

Both the revenue and sales and marketing forecasts demonstrated good performance over the validation set of the most recent 10 quarters (2.76% and 3.58% mean absolute percentage error, respectively).

At this point the auto-generated sales and marketing forecast can be adjusted to align more closely with assumptions or domain expertise. Let's say we expect sales and marketing to increase and want forecasted revenue to reflect that. We can either set forecast constraints so that future sales and marketing spend stay within given upper and lower bounds, or we can adjust future sales and marketing spend directly, similar to the previous example where we used scenario testing for the leading indicator.

To adjust the data points directly, we went to the 'variable forecasts' section of the forecast configuration options. At that point, we can either click and drag the forecast line or submit an Excel or CSV file with projected values. As seen below, we increased the values by an average of about 3%, which was arbitrarily chosen as an example.

As you can see, adjusting future sales and marketing spend to simulate a scenario in which we have knowledge of what's likely to occur resulted in an increase of forecasted revenue from 385,292 billion to 390,802 billion:

As demonstrated in this post, generating accurate and compelling financial forecasts doesn't have to be complicated or time-consuming. With Datamago, it can be done in seconds due to our powerful prediction engine and by incorporating leading indicators or fine-turning forecasts with your domain expertise.