Setting up forecasting

Getting a first forecast out of the product, and the handful of decisions that determine whether it is any use to you.

Getting a first forecast out of the product, and the handful of decisions that determine whether it is any use to you.

Set it up

1. Check the operational hours on your locations, because the forecast's own business hours default to them.
2. Decide what is being forecast. A workstream is the unit here, so the split you choose between channels and queues is the split you will read and staff.
3. In Workforce management, click Forecasting in the sidebar and select Forecast scenarios. Choose the workstream and set the data source in the Historical volume panel: your Zendesk data, or Imported data if you are bringing your own.
4. Generate a forecast and compare it against a period you remember.

You do not choose the algorithm

Zendesk WFM tests the available algorithms against your history for each workstream and uses whichever performs best there. So two workstreams on the same account can be forecast by different methods, and neither is a setting you tune.

What that means practically: a forecast that is poor is almost never the wrong algorithm. It is the history, the workstream split, or the hours.

Hours first

A forecast built on the wrong opening hours is wrong in a way that is hard to spot afterwards, because the chart still looks plausible. Ten minutes on the locations before anything else.

Forecast per channel

Phone and messaging arrive in sharp peaks and need people at that moment. Email arrives all day and can be absorbed. One combined forecast overstaffs one and understaffs the other.

Check it against a period you know

Generate a forecast for last month and compare with what happened. That single check tells you more than any amount of configuration review.

Exclude the abnormal history

An outage week, a campaign, the fortnight a competitor was down. Leaving those in teaches the forecast that they are normal, and it will predict them again.

The setting for it sits in the same Historical volume panel. With Zendesk data selected you choose between Use all available data from the last 2 years and Ignore data prior to a specific date, and the second is how you cut off a period you would rather the forecast did not learn from.

Handling time has to be real

Staffing comes from volume times handling time. Where tracking is not yet accurate, the forecast inherits that, so this is a reason to get tracking right first.

Expect to adjust it

A first forecast is a starting point. Reviewing it against actuals weekly for a couple of months is what makes it good, and there is no shortcut.

See also

• The forecast algorithms and when to use which
• Setting business hours for forecasts
• Why my forecast is far off