Plan requirement
| Subscription | Any plan |
| Also required | Workforce Management or Workforce Engagement Management |
| Access | Admin |
Predicting how much contact arrives and when, and how many people that needs. Workforce Management picks its own algorithm and looks twelve weeks ahead, per 15 minutes.
What it produces
Expected volume by interval, and from that the number of people needed to meet your service target. Not a daily total: the shape of the day, which is what staffing depends on.
The Required staffing table turns that into full-time equivalents for each 15-minute interval, using a modified version of Erlang C. FTE is a unit, not a headcount: the baseline is typically a 40-hour week, so somebody working 20 hours counts as 0.5. A requirement of 4.5 FTE in an interval is not four and a half people you can point at.
Why the shape matters
A day with two hundred contacts spread evenly and a day with two hundred arriving between nine and eleven need completely different rotas. A daily figure hides that entirely.
It picks its own algorithm, and looks twelve weeks ahead
Workforce Management tests every algorithm against your historical data and selects the best one per workstream by itself. You can override that choice. There are six across three families: two statistical ones (Eight-weeks average and Eight-weeks average with momentum), three machine-learning ones (Prophet, XGBoost, LightGBM) and one neural network (LSTM).
Which one you are on decides how the forecast behaves. On the eight-weeks average, the most commonly chosen, it averages the past eight weeks for the next 12, so something from nine weeks ago has dropped out and last week is an eighth of everything it knows. On the bare minimum of two weeks of data that algorithm forecasts four weeks and repeats them for the remaining eight, which looks like a twelve-week forecast and is not one.
To import history you need at least two weeks of data, and two years is the recommendation. It has to be in UTC, in the template format, at 15-minute intervals.
A new deployment is less bare than that suggests. Activating Workforce Management migrates six months of your Zendesk ticket history into it, so the first forecast is built on six months rather than on nothing. The engine will use up to two years if it holds that much; going back further is a manual import, which we arrange.
One thing to know before you compare figures: Workforce Management counts spam tickets, and Explore probably does not count them the way you do. The two volumes are not meant to match.
The intervals you see
Day view shows 15-minute intervals, week view one hour, and month view a day. A pattern that is obvious in the day view disappears entirely in the month view, so pick the view to match the question rather than the period you happen to be discussing.
What it cannot know
- A campaign nobody told it about.
- A product incident.
- A price change or a launch.
- A competitor's outage sending people to you.
Those need a manual adjustment, and knowing about them in advance is a process question rather than a product one.
Handling time is half the answer
Volume alone does not give staffing. How long each contact takes matters just as much, and that comes from tracking. A forecast built on guessed handling time is a guess with a chart.
Check it against reality weekly
Forecast against actual, every week. That habit is what turns forecasting from a projection into something people trust enough to plan around.
Comments
0 comments
Article is closed for comments.