Peak season is a rerun, not a surprise
Whatever your peak is — Black Friday, enrollment week, tax season, launch day — last year's transcripts already contain this year's questions. Peak planning is mostly the discipline of reading them in October instead of rediscovering them in December.
Four weeks out: pre-write everything
- Pull last peak's transcripts and list the top questions. Shipping cutoffs, availability, returns windows, seasonal pricing — the list is shorter than you fear.
- Write or refresh one KB article per question, dated for THIS season (a "2025 holiday shipping deadlines" article surfacing in 2026 is worse than none).
- Retrain the AI on the refreshed articles so the seasonal tier is absorbed before it arrives — peak volume is disproportionately the repetitive tier, which is exactly what the AI layer is for.
- Update the canned library with the season's specifics, so agents answer the cutoff question in one keystroke instead of forty.
- Narrow proactive invitations to the money pages only. Peak traffic makes every trigger fire more; an invitation that was pleasantly rare in April is wallpaper in December.
- Set honest operating hours for the season — extended if you are extending, unchanged if you are not. The offline layer carries whatever you do not staff.
- Put the cutoff answer on the page itself. The cheapest peak conversation is the one your checkout page prevented.
Borrowed help only works pre-trained
Colleagues drafted in for the rush follow the same onboarding arc, compressed: read transcripts now, shadow this week, answer the routine tier during the peak — with their own scoped logins created in advance, not at 9am on the busiest Monday. A helper answering with wrong information at peak costs more than the queue they shorten.
During: watch two numbers, drop the rest
Missed chats (your honesty gauge — if it climbs, shorten displayed hours or add a body) and first-response median (your saturation gauge — if it doubles, agents are over their honest concurrency cap). Everything else can be reviewed after; these two you steer by daily.
The day after matters as much
Peaks end with a debt: the offline messages and unfinished follow-ups of the last big day. Schedule the morning-after triage shift in advance — oldest first, before the widget goes online — and hold one short retro while it is fresh: which question surprised you, which article was missing, which trigger misfired. Write it down; that document is next year's four-weeks-out list, already done.
The off-season is when the work actually gets done
Every team promises itself that next year will be calmer, and then does nothing about it until six
weeks out. The tasks that genuinely reduce peak load — rewriting the three articles everyone
asks about, fixing the shipping-estimate wording that generates a thousand chats, agreeing who covers
the Saturday before the holiday — all take weeks of ordinary attention and none of them can be
done in a hurry.
Put a recurring block in the calendar for the quietest month of your year and give it a fixed
agenda: read last peak's transcripts, pick the top three question clusters, and fix the cause of one
of them properly. One properly fixed cause is worth more than a dozen new canned replies, because it
removes the conversation rather than speeding it up.
The other off-season task is writing down what happened. Peak memory decays fast, and by the
following year nobody can remember whether the queue broke on the Tuesday or the Thursday, or whether
the extra help was worth it. A one-page note written the week after peak is the most valuable planning
document you will have.
Decide in advance what you will stop doing
Capacity plans usually only add: more hours, more people, more canned replies. The plans that
survive contact with a real peak also subtract. Decide now which activities are suspended when volume
crosses your threshold, and write them down so nobody has to make the call under pressure.
Typical things worth suspending during peak: proactive invitations on low-intent pages, non-urgent
outbound follow-ups, transcript reviews, internal projects, and any chat routing rule that adds a hop
without adding accuracy. Each of these is valuable in a normal month and actively harmful in a week
where the constraint is agent attention.
Pair each suspension with a resume date. Things quietly switched off “for peak” have a
habit of never coming back, and a proactive invitation rule that stayed off for eleven months is a
silent revenue loss nobody attributes to the peak plan.
What to measure
Compare your peak week against the same week last year rather than against last month. Volume,
first response time and missed-chat rate are the three that tell you whether the plan worked; against
a normal month every peak looks like a failure, which teaches you nothing.
Track the ratio of new questions to repeat questions during peak. A high repeat share means your
off-season content work has room to run; a high share of genuinely new questions usually means
something changed in the product, the pricing or the shipping promise, and that is a signal for other
teams rather than a staffing problem.
Record how long it took to return to normal after peak ended. The recovery tail — the backlog,
the follow-ups, the tired team — is a real cost of the season and it is almost never counted when
people judge whether the peak went well.