Weighted Average Forecasting

Status: Public Preview — behaviour may change.

What it does

A weighted average forecast builds a demand prediction from a set of historical days you pick and weight, rather than fitting a linear trend over a long history. The model blends the intraday volume and average handle time (AHT) patterns of the selected days by their weights, and produces confidence intervals alongside. It's the third forecast scenario type in Workforce Management, next to short-term and long-term.

Key facts

Produces both volume and AHT as 15-minute intraday patterns for a target channel and queue.
Horizon is set by the forecast start date and duration you enter, not a fixed cap like the linear scenarios.
Data source is External only during preview — you upload the historical file; internal Dataverse data isn't supported yet.
Weighting modes: Manual (edit each weight, normalised percentage updates live), Recency (recent days weighted higher automatically), Equal (every day the same).
Refresh is manual only — no auto-refresh, unlike internal short/long-term scenarios.
Forecast entity is either Conversation or Case.
Confidence bands are shown at 80% and 95%.

When to use / skip

Use it when you have specific comparable days that describe the demand better than a trend line would — promotions, sale events, known seasonal spikes. Base an upcoming promotion on the pattern of past promotion days rather than letting a linear model average them away. Skip it for routine daily staffing where a short-term linear forecast on internal data is less effort and auto-refreshes.

Configuration decisions

Which historical days genuinely represent the day you're forecasting — the whole method lives or dies on this choice.
Weighting mode: Manual gives control but is fiddly; Recency and Equal are quick starting points you can then tune.
Forecast target: channel, queue, and the Apply To date the pattern maps onto.
Whether your external data file is clean and complete for the days you're selecting — there's no internal fallback in preview.

Gotchas

External data is mandatory in preview, so you can't lean on the WFM internal history you may already be using for short/long-term scenarios — you'll be exporting and reshaping it into the upload format.
No auto-refresh: the forecast is a point-in-time artefact and won't move as new actuals land. Someone has to remember to rebuild it.
Garbage in, garbage out is sharper here than with the linear models — a poorly chosen source day skews the whole blend, and the confidence bands won't save you from bad day selection.

Consultant notes

Clients hear "weighted average" and expect a self-maintaining forecast. Set expectations early: it's manual, external-data-only in preview, and best treated as a targeted tool for known event days rather than a replacement for the standard scenarios.
The upload step is where go-lives stall. Scope the export/transform of historical data into the required external format before you promise a demo — the modelling is the easy part.
It pairs well with a short-term linear forecast running in parallel: linear for the baseline week, weighted average layered in for the promotion day. Don't position it as either/or.
Being preview and external-only, don't build a client's peak-season plan solely on this yet. Use it to sanity-check, and keep the linear forecast as the system of record until it graduates.

Preview feature — revisit once weighted average forecasting reaches GA and internal data sources are supported.