Peak Season Support Quality: A Decision Timeline

Why is peak the one quality problem you can schedule?

Every other support quality failure arrives without warning. This one has a date on it. You already know roughly when volume rises, roughly by how much, and roughly which contact reasons arrive with it, because the same thing happened last year. Retailers can be more precise than that, because the trading weeks are fixed and published in advance in the National Retail Federation’s 4-5-4 calendar, which exists specifically so this year’s peak weeks are comparable to last year’s.

That predictability is the entire opportunity, and most teams spend it. Quality gets discussed in November, when every remaining option is a bad one, and reviewed in February, when nobody remembers what happened. The eight weeks in which the outcome was still changeable pass with the subject unopened.

So stop treating peak as a period to survive and start treating it as a sequence of decisions with expiry dates. Each one below is cheap and reversible at its own point on the timeline, and expensive or impossible after it.

Eight weeks out: do you run a quality programme through peak at all?

There are three honest answers, and pretending there are only two is how programmes collapse rather than shrink. Most teams never choose. They intend to keep going, lose reviewers to the queue one at a time, and arrive at the first option anyway, without the protections that would have made it survivable.

1. Suspend deliberately. If your review capacity is one senior agent who is also your fastest closer, the arithmetic can genuinely favour putting them on the queue. Almost nobody will say this out loud, which is why so many teams do it by accident instead of on purpose.
2. Shrink deliberately. Keep a much smaller programme running against a much smaller scorecard, aimed at the parts of the operation you understand least. This is the right answer for most teams.
3. Change the constraint. If reviewing stops drawing on the same headcount that answers tickets, the trade largely disappears. That is a procurement decision with a lead time, which is why it belongs at week eight and nowhere later.

If you suspend, three conditions make it survivable, and all three have to be written down. Name the blackout window with a start and an end date, so it reads as a decision rather than a drift. Keep a minimal safety net on the criteria that would void an interaction, because an unreviewed accuracy or verification failure becomes a second contact at your most expensive moment. And accept in advance that the season will leave a hole in your records exactly where next year’s planning will need data.

Whatever you choose, translate it into names. “We will keep QA running” is not a plan, it is a hope. The plan is “Priya and Marcus do not go back on the queue between 17 November and 5 January, and here is who covers their tickets instead.” Reviewers are almost always your most queue-capable people, which is exactly why they are redeployed first and why the commitment has to be specific enough to defend in a staffing meeting. Getting this wrong costs what understaffing always costs, in ways the staffing spreadsheet never shows.

Two things not to decide yet. Leave the sample alone, because the population you would be sampling does not exist yet. And do not renegotiate the annual target in a panic: whether a 95% target is achievable is a question about your rubric, not about December.

Eight weeks out: what do you tell leadership before the numbers move?

This is the highest-leverage thirty minutes in the season, and almost nobody books it. Telling your VP that a number you own is about to get worse feels like volunteering for a problem. It is the opposite. A dip you predicted in September is evidence you understand your operation. The same dip surfacing in a January board pack is evidence you did not.

Write one page and send it before anything moves. Five things, one screen.

• Which numbers will move, in which direction, and roughly how far. Ranges are fine. Pretending to precision you do not have will cost you the next conversation.
• Why, in one sentence a non-support executive can repeat. Usually: reviewers are queue-capable, so review capacity and answer capacity come out of the same headcount.
• When it comes back, with a date. An open-ended decline reads as loss of control. A dated one reads as a plan.
• The version warning. If the scorecard changes for peak, scores produced under it are not comparable to the annual series. Say so in writing now, because saying it in February sounds like an excuse.
• The one thing you are asking for. A contract reviewer, two weeks of extra ramp, a tooling decision. One thing, not five.

That last point is the one with a real deadline attached. Eight weeks out, budget exists, calendars have gaps and procurement can still complete. By November there is no money and no time, so the honest answer to any request is no. If your ask involves spend, frame it as avoided cost and avoided rework rather than as coverage percentage. The argument in the ROI of automated QA is the version most teams need, and the wider translation problem is covered in reporting support quality to people who do not run support.

One sentence does more work than the rest of the page combined: we expect quality to dip in weeks two to five, we have chosen where it dips, and here is the one thing that would prevent it. That converts a future failure into a present decision, in front of the person who can fund it.

Two weeks out: what do you lock, and what do you freeze?

The two weeks before peak are for stopping things, not starting them. That is counterintuitive, because it is also when everyone is most motivated to improve something before the rush. A change landing in week one of peak has no calm baseline to be measured against and no spare attention to debug it, so you get the disruption without the benefit.

Lock these, in writing, with dates on them

1. The scorecard version and its expiry date. A peak rubric needs a version label and an end date, or it quietly becomes the permanent rubric. Which criteria change is a separate question, and the reasoning behind weighting scorecard criteria is the right input to it.
2. The review commitment, with names. Not a percentage. Names, hours and who backfills their queue.
3. The escalation trigger. One measurable condition that reopens the plan, written down while you are calm.
4. Who owns quality if the QA lead goes back on the queue. Somebody has to, and finding out in week three that nobody does is how six weeks disappear.

Freeze these until the season ends

• New tooling. An implementation starting two weeks out will not finish before peak, and a half-configured system is worse than the spreadsheet it replaced. The sequencing in a QA tool rollout plan will tell you honestly whether it fits, and usually it does not.
• New contact-reason taxonomies. Change the categories mid-season and you lose the ability to compare peak to anything, including next year’s peak.
• New criteria and new definitions. Every change resets reviewer agreement at the moment you have the least time to rebuild it.
• Anything requiring training for the whole floor. The floor is about to be busier than at any other point in the year.

Two exceptions get through the freeze, both because they expire. The seasonal cohort has to be onboarded and their first weeks carry the most variance in the season, so the onboarding quality ramp runs regardless. And reviewers need one alignment session on the rubric they will actually use, not the October one, which puts a calibration session inside this window rather than after it.

During peak: what is the one signal worth watching?

Teams do one of two things during peak, and both are wrong. They try to track the full dashboard, which fails because the people who would do the tracking are answering tickets. Or they track nothing and find out in February. The workable answer is one leading indicator, published weekly, with everything else ignored until the trigger fires.

The quality score is the wrong choice, even though it is the obvious one. It lags by a week or more, it is produced by fewer reviewers than usual, and under a modified rubric it is not comparable to what you were looking at in October. It will move, and you will not be able to tell whether the team got worse or the measurement did.

Pick something cheap, fast and already instrumented. Usually that is reopen rate split by tenure cohort: it needs no reviewer, it moves within days, and it predicts January workload directly. Escalation rate by cohort is the alternative if your reopens are noisy, and the patterns worth reacting to are described in escalation handling quality.

Then define the trigger before you need it. A trigger is a measurable condition plus a pre-agreed response, written in September, so acting on it in December is administration rather than argument. Something like: if reopen rate on agents under 90 days exceeds double the tenured rate for two consecutive weeks, we pull one reviewer off the queue for three days and review only that cohort.

The value is not in the threshold, it is in having decided. Without one, either every bad Monday becomes a debate that burns the attention peak was supposed to protect, or nothing is ever quite bad enough and week six arrives with nobody having said anything. When it fires, what happens next is execution rather than strategy: resizing the sample, choosing what to relax and scoring the seasonal cohort are all set out in keeping quality honest during peak.

One rule for the rest of the season. Do not reopen the relax list mid-peak. You wrote it calmly, in advance, with the whole picture in front of you. Rewriting it in week four on the basis of one bad week is how a standard becomes a suggestion, and the floor reads that change faster than any announcement you make about it.

What does peak actually cost you, and when do you pay it?

Peak does not destroy quality, it borrows against it. Everything you chose not to look at still happened. Habits formed in six unreviewed weeks are now habits, and the coaching that did not happen is still owed. That is quality debt, and the problem is never taking it on, it is taking it on without a repayment plan and then being surprised by the bill.

Four lines make up the ledger, and each produces a recognisable symptom about four weeks later.

Why is January the wrong month to assume you will catch up?

The repayment plan only works if January is actually quiet, and for much of this industry it is not. Peak does not end when the buying stops. It ends when the consequences of the buying are resolved, and those arrive later, with a different shape and a worse tone.

The returns wave is the clearest example. The National Retail Federation’s 2025 Retail Returns Landscape put total industry returns at 849.9 billion dollars, with an estimated 19.3% of online sales coming back. Returns contacts are harder than sales contacts: the customer is already disappointed, the policy edge cases are live, and the agent who sold them the thing has usually left. So the month you scheduled for recovery carries its own quality risk, staffed by a team that has just shrunk.

Which means the repayment plan needs dates in it, booked before peak starts, or it does not happen. Three appointments are enough, and they belong in the calendar in September alongside everything else on this timeline.

1. Week one of the quiet period: coaching restarts. Before reporting, before the retrospective batch, before anything else. It has the shortest shelf life, because turning QA data into coaching works on recent behaviour and stops working on remembered behaviour.
2. Week two: the calibration session and the rubric switch-back, on a named day, so the programme restarts rather than drifts back.
3. Week three: the retrospective batch and the write-up, which is also when you draft next year’s version of this timeline.

Then close the loop with the person you briefed in September: the same one page, with the actual numbers next to the forecast ones. Being roughly right, on the record, twelve weeks early is the most durable credibility a support leader can build, and it is what makes next year’s resourcing conversation easier.

What do you need to have kept for any of this to be checkable?

Every decision on this timeline assumes you can look back and see what happened. Most teams cannot. They have scores produced under a rubric nobody can now reconstruct, no record of which weeks ran under which rules, and no way to tell whether a low December was a real decline or a measurement artefact. The plan was fine. The record was not.

Done by hand, three artefacts cover it and none needs software. A dated change log for the scorecard, one line per change, so you can always say which rules were in force in a given week. A weekly series of the one signal you chose, kept even in the weeks nothing happened, because the flat weeks are what make the bad week legible. And reviewer notes exported somewhere that survives the season. Half an hour a week, and it is the difference between a January review and a January argument.

What gets hard by hand is the evidence behind an individual score, which is exactly what gets challenged. A seasonal agent disputing a deduction in week five deserves the specific exchange it came from, and “the reviewer felt the tone was off” does not survive that conversation. Keeping the reasoning attached to the conversation rather than to a row in a spreadsheet is the practical form of score traceability.

This is where the week-eight constraint comes back. When scoring does not depend on reviewer hours, the record keeps itself, and full coverage is something most platforms now offer, so it is not the interesting part. The interesting part is whether the record is auditable afterwards. Kaizo’s Auto QA runs your own scorecard against conversations natively in Zendesk and Salesforce Service Cloud, keeping the supporting exchange alongside each result, so a December score can still be argued from evidence in February. Kaizo Insights holds the seasonal and tenured cohorts apart as separate series, which is what makes the comparison you promised leadership in September possible to produce.

None of that removes a decision from this timeline. What changes is only whether you can show, afterwards, that you made them well.