1. Marketing forecasting
Once we select KPIs, it opens the door to building marketing forecasts!
2. Forecasting on KPIs
Marketing Analysts spend a lot of time analyzing KPI trends. Eventually, we want to answer the question, "is this change normal, or is it exceptional relative to historical trends?"
Forecasts are the easiest way to anticipate changes in marketing KPIs at the channel level, or even for the marketing program.
Analysts will typically partner with finance to build forecasts that use similar assumptions and approaches.
At the most basic level, marketing forecasts should help us understand the future performance of KPIs based on historical data, and how channels are being leveraged to achieve funnel goals.
3. Forecasting accuracy
So what can ensure that we accurately predict future marketing behavior?
To make a forecast accurate, we need lots of historical data! Historical data contains seasonal patterns, anomalous events, and underlying longer-term trends.
Sophisticated forecasts separate each of those components to isolate true marketing patterns without seasonality.
4. Forecasting requirements
There are three main areas to check for before beginning a forecasting exercise.
Number one is: having at least two to three years of historical data for a given channel, ideally at a daily grain. This accounts for multi-year seasonality.
Similarly, we should work with marketing partners to make a list of anomalies that affect historical trends, like pausing channel spend temporarily.
Finally, we need to anticipate recurring seasonal trends. Like educational companies that see a spike in marketing interest that aligns with academic calendars.
5. Forecast planning process
Once we have vetted channel data to ensure it is set up for accurate forecasting, we can work our way through the rest of the forecasting process.
Step one we already accounted for: selecting KPIs!
Step two is determining any relevant dimensions for that KPI forecast. Like forecasting paid search conversions by geographic region.
After we plan for KPIs, channels, and dimensions, we should check available time granularity. Some advertisers aggregate data at a daily or weekly level, and we must forecast at the lowest common granularity.
The next steps are related: deciding how far into the future we will forecast, which dictates how often to refresh results. Just keep in mind, the further out we try to forecast, the more difficult it will be to achieve high accuracy!
6. Forecast modeling options
We may be thinking at this point, "forecasting sounds great, but how do we actually build one?"
Fortunately, forecasting is a common exercise for analysts, so options range from out-of-the-box functions to advanced machine learning models.
One starting place is to use tools like Excel or advertiser reports. While we cannot make many adjustments to these forecasts, we can get a sense of trends quickly.
Over time, we can evolve to statistical models, which we can more finely-tune.
There are many different statistical modeling options like regression, econometrics, time series, and more. Whichever method we select, we can use forecasting accuracy to compare and keep improving!
7. Target setting
Marketers use forecasts for two main purposes: anticipating future trends and setting targets! Good targets will not be unrealistic or too easy to achieve. Forecasts can help marketers find a middle ground in setting channel targets.
The first step is to have an accurate forecast in place to estimate seasonal trends for KPIs.
We can set targets relative to the forecast to make sure targets are dynamic and realistic.
Marketing partners then add in domain knowledge to further adjust targets, like knowing that the budget will be 25% lower than prior years. Forecasting plus target setting provides a powerful way of understanding and growing marketing impact!
8. Let's practice!
Let's review the requirements of an accurate forecast!