# You don't need "AI" forecasting (part 2)

Updated: May 31, 2022

Do this one thing and you will have most of what you get from the AI forecasting engines. In __part 1__ of this post, we showed how to produce weighted forecasts by defining the time frame the forecast probability applies to and by adding a custom field to differentiate between the probability to close this quarter and to close eventually. We also showed that traditional classification forecasts are not very good predictors until the end of a quarter.

If you have not yet implemented the two probability fields, you can approximate the effect by using the opportunity close date to include/exclude the opportunity in your weighted forecast. If the close date is within the current quarter, then include it. This simple approach can yield remarkably good results.

Here again is the example we presented in part 1. The top purple line and bottom black line are the Best Case and Commit classification forecasts respectively. The top of the colored areas is the Funnelcast prediction.

Here’s a weighted forecast (the same dataset as in the previous example) produced by applying the close in quarter rule with the standard Probabilities field to produce a weighted forecast. For reference, that top purple line is the Best Case classification forecast. The black line below that is the weighted forecast, filtered to only include opportunities that are marked to close in the applicable quarter. The top of the stacked area is the Funnelcast forecast.

The weighted forecast (close-in-quarter filtered with Probability) is a very good prediction. Although it is still over-forecasting in this case; probably because the Probability field reflects a long-term likelihood. Unlike the Commit and Best Case classification forecasts which (respectively) under and over-estimate at the start of the quarter, this weighted forecast gives a reasonable early indication of what to expect, and is considerably better than either of the classification forecasts throughout the entire quarter.

It is tempting to think that (at the start of the quarter) the weighted forecast is comparable to the Funnelcast forecast. But the two lines are predicting different things. The weighted forecast is for the open funnel only. The Funnelcast forecast includes a forecast for new funnel (the green area).

The comparable Funnelcast forecast for the open funnel is the top of the red area. Funnelcast was at $14.2 M on day-one of the quarter, the weighted forecast was $17.7M; the final result from the open funnel at the start of the quarter is the top of the dark blue region, $14.0M in this case. So, the day-one Funnelcast forecast was excellent. But the weighted forecast using the simple close-in-quarter filter is OK—a little high, but still better than either of the classification forecasts.

To further improve weighted forecasts, we recommend clearly defining the time-period that the probability applies to. Create and track a custom field for the probability that an opportunity will close in the current quarter. Use the standard probability field to represent the probability that an opportunity will eventually close. If you follow these steps, your current quarter weighted forecasts can approach the quality of the Funnelcast forecasts or any of the other sales forecasting engines on the market.

You may be wondering why a vendor that sells a sales forecasting platform would say you don’t need one. Well, while we think we do a better job than the other vendors, we don’t think that current-quarter forecasts are all that special. We often see very-good forecasts using the simple method we described here and in part one—before applying our secret sauce.

And the “AI” forecasting engines are not all that special either. They offer sales dashboards, basic summary reporting, a different place to enter your forecasts outside of your CRM, and AI scores. The dashboards and summary reports can be replicated in Salesforce and Tableau CRM. Nothing unique there. And they are not using advanced analytic modeling. They are using models that you could replicate in a day after completing a statistics 101 course.

That said, these models are a little better than the simple approach we recommend. Both approaches are quite good at predicting current-quarter results early in the quarter. That can be pretty helpful to set expectations. But how does that help you sell more?

Some of these "AI" vendors use more advanced statistical methods to analyze additional data sources outside of the CRM system (calendars, email, documents, conversations…). (This part may actually involve real AI techniques.) They do this to find signals that add more precision to forecasts. But these signals are most meaningful at the very end of the sales process—when you know what will happen anyway. How does a more accurate forecast at the end of the quarter/sale help you sell more? Is that useful? You decide.

On the other hand, if you try to extend a weighted forecast (using human provided opportunity probabilities or "AI" scores) beyond the current quarter you will get very unreliable results because you need, for each opportunity, the probability to win in the applicable future period. That's more complex than a single number can capture. And the signals that are useful for in-quarter forecasts are mostly meaningless beyond the end of the current quarter. So the "AI" methods (even with their special in-quarter AI signals) are unhelpful for forecasting beyond the current quarter.

To forecast for next quarter and later, you need a different approach. That’s what Funnelcast does for you—in addition to providing current quarter predictions. And with those long-term forecasts, you can answer planning questions like:

What can we expect next quarter, next year?

How much new demand generation do we need to meet our plan?

Where and when should we allocate resources to maximize sales?

Answering these questions helps you set the right goals, hire the right staff, and focus on the right opportunities at the right times. These things actually increase sales and marketing productivity.

We think that is special. You decide.