Forecast Time Granularity

Hi. I’m trying to create a forecast using Quicksight’s built-in forecast function, for example, total revenue for the next 30 days.

For my field wells:

  • X-axis: date
  • Value: revenue

I would like for the forecast graph to show the forecasted revenue for the next 30 days, for each day. When setting up the date with day as aggregate, I will get the following error:

Is it possible for me to still have the forecasting per DAY? If that’s not possible, how do I set the visual to use the detectedTimeGranularity?


Hi @sabmz - this could be because you have too many day-level values in your dataset. Can you try to apply a filter, perhaps for the last 3 months (as an arbitrary test), and then apply these same forecast settings again?

Thanks @crease !

“too many day-level values” - can I know what is the limit for day/ month/ year levels?

Hi @sabmz this is my mistake. When there are too many values (2,000 is the max, regardless of granularity), you will get a more specific error message. When you see an error like the above, it is because our forecasting algorithm has predicted that your data should be visualized at a different granularity. It could theoretically generate a forecast at the granularity you have, but it is likely to be noisy and not very accurate. So rather than just show that, the algorithm instead recommends that you adjust the granularity based on its prediction.

You could also apply a filter, as mentioned above, and try to limit the number of days to something where there is a more discernible daily trend that is well suited for forecasting.

Lastly, it looks like we have an issue where the variables detectedTimeGranularity and userSuppliedTimegranularity are not being translated into their proper values in the second sentence. But they are the same as the first sentence - DAY = user supplied, and MONTH = system detected.