Why Smart Prediction?
The Fathom Smart Prediction rule allows you to use linear regression to forecast the future values of your forecast. This linear regression draws a line of best fit through the data to forecast future periods. We built this functionality into Fathom to enable you to leverage your business's historical performance, without leaning on methods, like last year's actuals, that are more fixed in nature.
The core principles of our Forecasting tool are auditability and transparency of forecast data. Which is why we have opted out of developing any black box machine learning methods that are often difficult to recalculate. Instead our Smart Prediction functionality uses mathematical formulas you can trust.
Smart prediction is just one value generation method that can be used to calculate your forecast. To read more about all of our value rule methodologies, please reference this article.
Smart Prediction short term
When applying a Smart Prediction value rule, you can tell Fathom how much prior period data to use in the linear regression model to predict your future values. You can select from the following date ranges in the short term:
- 3 months
- 6 months
- 12 months
- 24 months
We built in some flexibility to these rules in order to allow you to select the data that you believe to be most representative of your future periods. This can be handy when looking to forecast out future periods based on your most recent results - particularly relevant during volatile economic conditions, like those seen in 2020!
Smart Prediction long term
If you choose to use 24 months (2 years) of actuals data, Fathom is able to factor seasonality into the statistical model. This allows you to account for any numerical trends and changes to the chart of account line items, that are tied to seasonal conditions.
FAQ: Why are my forecast values negative?
Smart Prediction finds the line of best fit according to your historical data and continues that trended line forward into the forecast. If the historical data has a downward trend, then this will eventually result in negative values in your forecast.
To correct for this, you may want to choose a different date range of historical data (prior 3 months, 6 months, 12 months, or 24 months) to base the Smart Prediction calculation on and change the calculated trend. Or you may want to choose a different value rule for the account or the time period for which the forecasted values are negative.