The Intraday Management Playbook: What to Do When the Forecast Breaks
Every forecast breaks. The marketing team launches a campaign nobody told WFM about, a billing run generates a wave of confused callers, two agents call out, AHT drifts 40 seconds because of a new product flow. Forecasting better helps at the margins, but the difference between centers that hit service level consistently and centers that explain misses in retrospect is almost never forecast accuracy. It is what happens in the four hours after the forecast stops matching the floor.
This is the playbook for those four hours: detect, size, decide, act, and close the loop.
Step 1: Detect the divergence early
The cardinal rule of intraday management is that lead time is everything. A coverage gap detected three hours ahead has cheap fixes: a moved lunch, a shifted break block, offline work pulled forward. The same gap detected twenty minutes ahead has only expensive ones: emergency overtime, abandoned coaching, supervisor escalations. Every improvement in detection speed converts expensive fixes into cheap ones.
Watch three variance streams against plan, at the interval level:
- Volume variance: actual contacts versus forecast, per interval, cumulatively through the day. A morning running 6 percent hot rarely cools off by itself; arrival patterns have momentum.
- AHT variance: handle time drift is the stealth killer because it compounds with volume. Volume 5 percent hot and AHT 8 percent long is a 13 percent workload miss, well past what typical cushions absorb.
- Staffing variance: call-outs, late arrivals, long meetings, adherence drift. You planned shrinkage as an average; today is a sample, and samples vary.
Set explicit trigger thresholds before the bad day, for example: act when projected service level for any future interval falls more than five points below target, or when cumulative volume variance crosses 8 percent before noon. Without pre-agreed triggers, every spike becomes a judgment-call debate, and debates consume exactly the lead time you were trying to protect.
Step 2: Size the rest of the day
A spike that hits an overstaffed interval needs nothing but a note. The question is never "is volume high?", it is "which future intervals break if this continues?" That requires projecting the variance forward: re-run the staffing math for the remaining intervals using observed volume and AHT instead of forecast values.
The math is the same Erlang arithmetic used in planning, just refreshed. If the 1:00 PM interval was forecast at 380 contacts with 360-second AHT and you are now projecting 420 contacts at 390 seconds, offered load moves from 38 to about 45.5 erlangs, roughly eight additional productive agents at the same occupancy target, before shrinkage. Run your own scenario through the Erlang C calculator to see how hard requirements move with seemingly small input drift; the nonlinearity is exactly why sizing matters more than instinct.
Crucially, size in both directions. Intraday management includes catching the overstaffed afternoon early enough to offer voluntary time off, move training forward, or attack the email backlog, capacity recovered from quiet intervals funds the loud ones.
Step 3: Decide, using the cheapest effective lever
Work down the ladder from least disruptive to most. The order matters because the early rungs are nearly free and the late rungs spend real money or real goodwill.
- Move flexible time. Breaks and lunches can usually shift 30 minutes without policy issues; coaching, one-on-ones, and training blocks are movable by definition. Sliding four lunches out of the 12:30 interval is two productive agents in the gap, at zero cost.
- Pull forward offline work. Agents on email, back-office, or projects are a reserve pool. Borrowing two of them for the peak hour costs only backlog timing.
- Targeted overtime. An hour of OT offered to agents already on site, scoped to the broken intervals, beats blanket extensions. Remember the Erlang cliff: near the edge, two or three accepted offers can move service level by several points.
- VTO for the surplus side. In overstaffed stretches, voluntary time off cuts cost and banks goodwill you will spend on the next OT ask.
- Escalate. When the projected miss exceeds what the levers recover, leadership needs the news early, with numbers: "we project 71 percent against an 80 percent target from 2:00 to 4:30, after OT and schedule moves" is a manageable conversation at 11:00 AM and an autopsy at 5:00 PM.
For every move, write down the expected recovery before you make it. "Two lunches moved, one OT hour accepted, projected 2:00 PM coverage improves from −4 to −1" takes thirty seconds and turns the end-of-day review from folklore into data.
Step 4: Act, and tell the people affected
Execution discipline is mundane and decisive: the agents whose breaks moved should hear it from their supervisor with a reason ("the 1:00 interval is short, your lunch shift covers it"), not discover it on a screen. Reasons convert annoyance into participation; teams that explain intraday moves get faster acceptance on the next one.
This is also where the group-chat operating model fails quietly: decisions scatter across threads, nobody records what was tried, and the operation cannot learn from days it cannot reconstruct. Whatever tooling you use, the decision log is the asset.
Step 5: Close the loop weekly
Once a week, review the intraday record: which triggers fired, which moves were made, what they cost, what they recovered, and what the variance turned out to be. Three categories of finding feed back upstream:
- Forecastable surprises (campaigns, billing cycles, product launches) become calendar feeds into the forecast, the fix is communication, not math.
- Systematic drift (AHT creeping for three weeks) becomes a forecast input update and sometimes a coaching or process conversation.
- Genuine noise (the outage, the viral post) validates the playbook itself: the only answer to the unforecastable is response speed.
Teams that run this loop build institutional judgment instead of depending on the one supervisor who has seen everything. They also stop relitigating thresholds mid-crisis, because the thresholds earn their authority from the reviews.
What this looks like with tooling
Nothing above requires software, analysts ran this loop on paper for decades. What software changes is the cycle time and who can run it. QueuePilot's Intraday Copilot watches volume, AHT, and staffing variance against the interval forecast continuously, recalculates risk over 15, 30, and 60-minute windows, and turns the sizing step into standing answers rather than a spreadsheet fire drill. When a trigger trips, it recommends the specific cheapest-lever move, offer OT 2:00 to 4:00, release VTO at 6:00, slide two lunches, with the reasoning written out, and a human approves every action, which produces the decision log for free.
The playbook is the same either way: detect early, size honestly, spend the cheapest lever first, tell people why, and review weekly. Forecasts will keep breaking. The operations that thrive are the ones that stopped being surprised by that.