Weekly Forecast-vs-Actual Review (generic process)

Weekly Forecast-vs-Actual Review (generic process) is the generic process card, GP-602, for the weekly thirty-minute review at which the forecasting, scheduling and real-time seats of a workforce function compare the week's forecast of record with what happened by driver (volume, handle time, shrinkage, attendance), attribute each variance that is signal rather than noise to the function that owns it, and decide what changes. WFM Processes names this one meeting, held consistently, as the highest-return process improvement available to a workforce function, because it is the only point at which the three pillars' feedback loop closes on a calendar rather than by complaint, and the contact-center management literature has long treated the forecast-to-actual review as a standing practice rather than an investigation. No existing page documents it as a process, so this card owns the method. It is part of Process Standardization and sits among the cross-cutting sections of the function catalog in Anatomy of a ROC Standard.

The review is not the accuracy measurement: the stairstep that scores each forecasting layer is the method of Forecast Value Added in Workforce Management, and the error measures are on MAPE WAPE and Forecast Bias; the review reads those reports and decides on them. Nor is it the reforecast, which it may trigger and which GP-103 runs. Its ancestor in demand planning is the weekly review that forecast-value-added practice recommends, so that adjustments carry a recorded reason and are scored against the forecast they replaced.

Process profile

The profile extends the L0 identity fields of Process Decomposition (L0–L3). Local values are held by the deploying function and marked SET LOCALLY. Authority tiers follow Decision Authority Tiers in Workforce Operations.

Field	Generic value
Card	GP-602 · local card id assigned by the standards committee at deployment
Class	both, in the catalog's four-class vocabulary (strategic · real-time · both · supporting): it reads the week's steering and changes the plan
Trigger	the weekly calendar: the first business day after the week's actuals close
Cadence	weekly; thirty minutes [A]; the closed week is reviewed against the forecast version in force when its schedule was built, and the coming two weeks against the latest version
Scope in	every queue and client portfolio the function plans; the four drivers; the noise test; attribution; the decisions the review may take; the action log
Scope out	computing the accuracy measures (GP-106); the reforecast itself (Reforecast and Rolling Forecast Methodology); incident learning (Post-Mortem and RCA for Workforce Operations); intraday steering (GP-301)
Inputs	the forecast of record by vintage (Forecast Vintages); actuals by driver on the register's definitions; the real-time event log; the post-lock variance log (Forecast Lock Process); the adherence summary; the week's stairstep where kept; last week's action log
Outputs	the attributed variance table with grades; the decisions; the action log with owner seats and dates; items routed to the reforecast, schedule maintenance, lever policy or the collision calendar, or logged out of scope with the reason
Upstream	real-time (the event log, GP-301); scheduling (the published schedule and its changes, GP-202); forecasting (the vintage, GP-103); the definitions register
Downstream	GP-103 Weekly Rolling Reforecast (the trigger); GP-202 Schedule Maintenance; GP-302 Real-Time Lever Deployment; GP-106 (attributions feed the stairstep); the Forecast Collision Calendar; the post-mortem process where a variance was an incident
Resources	a forecast store that keeps vintages; a reporting view of actuals by driver; the event log; the action log; a standing slot with the three leads present. Platforms are named at L3, never here
Decision points	signal or noise; which function owns it; reforecast; schedule change; lever-policy change; incident rather than planning miss
Authority tier	chaired at tier 3 (the planning lead or ROC duty lead); a reforecast inside the lock's threshold at tier 1 (the forecaster); a change to a locked month through the lock's exception path at tier 4; a lever-policy change at tier 3, and at tier 4 where it changes committed cost
Risk and compliance flags	attribution read as blame (the review attributes to a function, never a person); acting on common-cause variation, which the quality tradition calls tampering; a decision that changes a partner's billable hours is a change-control item; a schedule change inside a statutory notice window carries predictability pay
Variant-prone steps	the driver set where a channel has no observable in-contact state; the attendance driver where a works council limits individual-level review; the schedule-change step inside a fair-workweek notice window; the lever-policy step at a partner supply node (a site, centre or partner)
Automation candidates	six L2 sub-steps (listed below)

L1 flow

Owner seats in parentheses, one per step. Decisions are questions. The walk is exception-based: the pack carries every row, but only signal rows are discussed in the thirty minutes [A].

1. Assemble the pack: forecast vintage against actual by driver, per queue, each number with its grade (reporting analyst).
2. Do both sides of every comparison cite the same register entry and version? (reporting analyst)
3. Open the review and walk the four drivers in order, volume, handle time, shrinkage, attendance, discussing signal rows only (planning lead, as chair).
4. Is each variance outside the band the function has set for noise? If not, it is recorded and not discussed (planning lead).
5. For each signal, read the event log and the post-lock variance log for that queue and day (real-time lead).
6. Was the miss in the forecast, the schedule, the in-day steering, or was it an incident? The chair decides after hearing the three leads (planning lead).
7. Is the signal a recurring signature, and if so, does it become a forecast input, a response-library entry, or a collision-calendar entry? (forecasting lead)
8. Does the variance trigger a reforecast? Inside the lock's threshold the forecaster acts at tier 1; above it the item goes to the lock's exception path (forecaster).
9. Does it require a change to the coming weeks' schedules? (scheduling lead)
10. Does it require a lever-policy change: the thresholds and rules for overtime, voluntary time off, skill rebalancing or break resequencing? (real-time lead)
11. Write the action log: item, owner seat, date, and the observation that will close it (planning lead).
12. Close last week's actions, carry open ones with their age, and publish the log and the attributed table (planning lead).

L2 step table

The documentation join key is the key an L3 job aid, a ticket and a variant row cite. Thresholds and system names are L3 values.

Step	Sub-step	Owner (seat)	Inputs	Outputs	Duration	Automation candidate	Documentation join key
1	Pull the forecast vintage in force when the week's schedule was built and the actuals by driver per queue on the register's definitions; carry each grade	reporting analyst	the forecast store; the reporting view	the pack	minutes [A]	Y	GP-602/1.1
1	Attach the week's stairstep and last week's action log	reporting analyst	the stairstep; the log	the full pack	minutes [A]	N	GP-602/1.2
2	Confirm entry ID and version on both sides; withhold a mismatched comparison	reporting analyst	the pack; the register	a checked pack	minutes [A]	Y	GP-602/2.1
3	Walk volume, handle time, shrinkage, attendance in that order; signal rows only; keep to the slot	planning lead	the pack	the walked pack	thirty minutes [A]	N	GP-602/3.1
4	Compare each variance with the noise band for that driver and queue (SET LOCALLY)	planning lead	the pack; the band	signal or noise per row	before the slot	Y	GP-602/4.1
5	Read the event log and the post-lock variance log for the signal's queue and days	real-time lead	the two logs	the context	inside the slot	N	GP-602/5.1
6	Propose an attribution from the context; hear the three leads; decide	planning lead	the context	an attributed row	inside the slot	Y	GP-602/6.1
6	Route an incident to the post-mortem process and drop it from attribution	planning lead	an attributed row	a routed item	minutes [A]	N	GP-602/6.2
7	Test for recurrence against prior weeks' attributed tables	forecasting lead	the tables	recurring or not	inside the slot	Y	GP-602/7.1
7	Choose the destination: forecast input, response library or collision calendar; record it	forecasting lead	a recurring signature	a routed item	inside the slot	N	GP-602/7.2
8	Compare the variance with the lock's threshold (SET LOCALLY); act, or route to the lock's exception path	forecaster	the variance; the threshold	a reforecast trigger or an exception item	minutes [A]	N	GP-602/8.1
9	Raise a schedule change request for the coming weeks with the reason and, where a notice window applies, its pay estimate	scheduling lead	the attributed row	a change request	minutes [A]	N	GP-602/9.1
10	Raise a lever-policy change with the reason and the cost effect	real-time lead	the attributed row	a policy change	minutes [A]	N	GP-602/10.1
11	Write each action with its closing observation	planning lead	the decisions	the action log	inside the slot	N	GP-602/11.1
12	Age open actions; close those whose observation has arrived; publish	planning lead	last week's log	the published log	minutes [A]	Y	GP-602/12.1

Variants

Recorded per Standard and Variant in Process Documentation:

• Statutory. Where a works council restricts review of individual attendance in a cross-functional forum, the attendance driver is reviewed at team aggregate and individual follow-up moves to the supervisor: the step's inputs change, and it takes a variant row owned by the works-council liaison. Where a fair-workweek statute requires notice and predictability pay for a schedule change inside its window (Predictive Scheduling Laws), step 9's change request carries the window check and the pay estimate as parameters: SET LOCALLY, no variant row.
• Contractual. At a partner supply node a lever-policy change that alters billable hours is parameterized by the contract's change-control clause and routed through Vendor Change Control for Routing: SET LOCALLY, no variant row.
• Structural. An asynchronous channel with no observable handle time replaces that driver with work-item throughput: a variant row reviewed each quarter until an inferred measure exists.
• Transitional. A client portfolio migrating from a predecessor organization that still forecasts on its own calendar runs its review on that calendar under a variant row that expires at the cut-over the standards committee dates.

L3 register

Job aid	Anchored to	Form
The pack template with its grade column and vintage stamp	step 1	work instruction
The thirty-minute agenda, the driver order and the signal-only rule	step 3	work instruction
The noise band per driver and queue, and how it is set	step 4	work instruction
The attribution decision tree (forecast · schedule · steering · incident)	step 6	playbook
The recurring-signature test and its three destinations	step 7	playbook
The reforecast-trigger rule and the routing to the lock's exception path	step 8	work instruction
The schedule change request and the lever-policy change forms	steps 9 to 10	work instruction
The action log layout and the closing-observation rule	steps 11 to 12	work instruction
The missed-review runbook: what happens when the slot cannot be held	steps 1 to 12	runbook

Automation candidates

Flagged at L2 and released per class of action on a published catch rate per Human Gates and Number Grades. The list is ranked by how easy a catch rate is to publish for the sub-step; each item after the first says why it ranks after the one before.

1. Pack assembly (1.1): deterministic given the vintage store and the register, and a wrong pack is caught in the room the same week.
2. The definitions check (2.1): a comparison of entry IDs and versions; ranks after the pack because a missed mismatch is found only when two numbers disagree later.
3. Noise-band flagging (4.1): flags rows and decides nothing; ranks after the definitions check because its error, a signal flagged as noise, is seen only if someone reads the undiscussed rows.
4. Action-log ageing (12.1): clerical; ranks lower because a mis-aged action is noticed only when its owner is asked.
5. The recurrence test (7.1): a match against prior weeks' tables; ranks lower because the pattern it must find is defined by the function and its misses are invisible until the signature recurs again.
6. The attribution proposal (6.1): an agent may propose from the event log; last because the chair's decision is the only measure of its error, and the attribution stays with the chair until a catch rate is published.

Steps 8, 9 and 10 are never candidates. Each changes what the floor does or what the business commits to, which is where the planning gates sit; and the evidence on judgmental adjustment is that small, frequent changes without a recorded reason reduce accuracy more often than they improve it, so the review's value lies in deciding fewer things, with reasons, not in deciding faster. Whether a week's difference is real at all, common cause or special, is the question the noise band exists to ask before anyone acts.

Maturity Model Position

Four scales on this wiki use the word level; the launch page states which is which. On the WFM Labs Maturity Model™ a function that holds the review weekly with the three leads present and an action log is at Level 2; one whose pack, definitions check and noise flagging run on a published catch rate and whose recurring signatures feed a response library is at Level 3, where the event log the review reads is the harvest record of Variance Harvesting; one whose attribution is agent-proposed and whose stairstep is scored forward each week is at Level 4.

See Also

• Process Standardization — the hub and the inventory this card belongs to
• WFM Processes — the three-pillar architecture and the feedback loop this review closes
• Forecast Value Added in Workforce Management — the stairstep the review reads
• Variance Harvesting — the in-day practice whose event log is the review's input
• Forecast Lock Process — the exception path a reforecast above threshold must take
• MAPE WAPE and Forecast Bias — the measures on the pack
• Forecast Bias Detection and Correction — when attribution finds a persistent sign
• Decision Authority Tiers in Workforce Operations — the tiers the profile uses
• Standard and Variant in Process Documentation — how the variants above are recorded
• Wiki:Packs/Process Decomposition — the pack that localizes this card (CP-OPS-001)