Importing historical volume

Giving the forecast a past to learn from. The fastest way to make a new deployment useful.

Giving the forecast a past to learn from. The fastest way to make a new deployment useful.

Import it

1. Export historical volume by interval from wherever you hold it.
2. Clean it: remove abnormal periods, check the intervals.
3. Import it.
4. Generate a forecast for a period you remember and compare.

Why it is worth the effort

Without history the forecast has nothing to learn from and stays poor for months. With a year imported, it is usable in a week.

That difference decides whether the team trusts the tool or writes it off early.

Interval level, not daily

Daily totals cannot produce the shape of the day, and the shape is the entire reason to forecast. If your history is only daily, the import will be much less useful.

Exclude the abnormal

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

Removing them is the difference between a year of history and a year of noise.

Mark the holidays

Before importing, make sure holidays are configured. Otherwise a quiet public holiday looks like a collapse in demand on that date, every year.

A year is the useful amount

Enough to capture seasonality without importing data describing a business that no longer exists. Two years of a company that has since doubled is not more information.

Check it before relying on it

Forecast a past month and compare it with what actually happened. If that comparison is poor, the import is wrong somewhere and no algorithm will fix it.

See also

• Setting up forecasting
• Why my forecast is far off
• Zendesk WFM product limits