Plan requirement
| Subscription | Any plan |
| Also required | Workforce Management or Workforce Engagement Management |
| Access | Admin |
The forecast does not match reality. Six causes, and five of them are the data rather than the method.
Why is my forecast so far off?
- Volume consistently over or under what arrives.
- The shape of the day is wrong even when the total is close.
- Particular days are always wrong.
One: not enough history
Less common than it sounds. Activating Workforce Management migrates six months of your Zendesk ticket history, so a fresh deployment starts with half a year rather than with nothing. The case that bites is a workstream that is genuinely new, or a Zendesk account that is itself young.
Where it does apply it shows up hardest on the statistical algorithms. Eight-weeks average needs two weeks at a minimum, and on that minimum it forecasts four weeks and repeats them for the remaining eight, which reads as a twelve-week forecast and is not one.
Fixed by importing history rather than by changing anything: at least two weeks, ideally two years, in UTC and at 15-minute intervals.
Two: abnormal periods in the history
An outage, a campaign, an unusual month. The forecast learned that those are normal and predicts them again.
How sharply depends on the algorithm your workstream landed on. On Eight-weeks average one bad week is an eighth of everything the forecast knows, which is why a single outage can distort a quarter. Exclude the period and regenerate.
Three: a public holiday
This produces "particular days are always wrong", and there is nothing to configure: Zendesk WFM has no holiday calendar, so a national holiday is a quiet day it never saw coming, every year.
The handle you do have is a volume adjustment on the forecast scenario, made in advance for the days you already know about.
Four: business hours wrong
If the hours do not match reality, contact arriving outside them is either ignored or attributed oddly. The shape goes wrong before the total does.
Five: something changed in the business
A new product, a price change, a marketing push, a channel added. The forecast is predicting the old business faithfully.
This is not a fault; it is the forecast needing to be told.
Six: the method does not suit the pattern
Last, not first, and remember you did not choose it: Workforce Management tests all six algorithms against your history and picks one per workstream. Overriding that is a deliberate act, and it is only worth doing after the five above are ruled out, because a different algorithm on bad data produces a differently wrong answer.
And one that is not a fault at all
Workforce Management counts spam tickets. If your own volume figure comes from Explore and excludes them, the two will never agree, and the forecast is not what is wrong.
How to find it
Look at which intervals are wrong. Whole days wrong points at a holiday or at the hours. Peaks wrong points at seasonality. Everything wrong by a similar proportion points at a change in the business.
Comments
0 comments
Article is closed for comments.