Most support teams plan for “Q4 peak” as if it were a single event. For brands selling into China, it isn’t. National Day Golden Week, the Singles’ Day presale window and the 11.11 sales day itself each produce a different mix of questions, at a different pace, from customers in a different state of mind. A plan built for one of them quietly fails the other two.

With National Day starting on October 1, there is roughly one week left to get the first peak right, and about six weeks for the rest. This is the readiness plan we walk teams through.

Why Q4 in China Is Three Peaks, Not One

Window Timing What drives contacts Typical contact mix
National Day Golden Week October 1 onward, one week Travel, gifting, delivery during holidays Delivery-time questions, store hours, travel bookings and changes
Singles’ Day presales From October, several weeks Deposits, bundles, price-protection rules Pre-purchase questions about deposits, coupons, stacking rules
11.11 and the week after November 11, plus a long tail Payment, fulfilment, returns Order status, failed payments, returns and refunds

The shift that matters is from pre-purchase questions in October to post-purchase questions in November. Pre-purchase questions convert: answer them fast and accurately and you win the sale. Post-purchase questions protect: answer them badly and you lose the customer and invite a platform complaint.

Step 1: Forecast by Intent, Not by Volume

A single contact-volume forecast hides the thing you need to staff for. Break last year’s Q4 transcripts down by intent and by week, then apply this year’s growth assumption per intent rather than across the board.

Three intents usually deserve their own line in the forecast:

  • Deposit and coupon-rule questions during presales. They are highly automatable, but only if your AI agent has this year’s rules, not last year’s.
  • “Where is my order” in the ten days after 11.11. This is high volume, low complexity, and ideal for proactive notifications that stop the question being asked at all.
  • Returns and refunds from mid-November. Volume is lower, but the cost of a wrong answer is high, so this is where most of your human capacity should go.

Step 2: Stress-Test Your AI Agent Before the Traffic Does

The most common peak-season failure is not an agent that crashes. It is an agent that answers confidently with an outdated policy. Before October 1, replay a sample of last year’s peak conversations against your current agent in shadow mode, and review every answer that touches pricing, deposits, delivery cut-offs or return windows.

Then update the knowledge the agent relies on for this year’s campaign: presale dates, deposit rules, courier holiday schedules and any changed return terms. Treat these articles as campaign assets with an owner and an expiry date, not as permanent help-center content.

Step 3: Write the Escalation Rules for Peak, Not for Normal

Escalation thresholds that work in a quiet month can flood your human queue at peak, or starve it. Revisit your escalation matrix with peak-specific rules:

  • Raise the automation ceiling for low-risk, high-volume intents such as order status and delivery estimates.
  • Lower it for anything involving money leaving the customer’s account: failed payments, double charges and refund disputes.
  • Add a time-based rule. A customer who has contacted you twice about the same order within 48 hours goes to a human, whatever the intent.

Step 4: Staff the Human Layer for Exceptions

At peak, your human agents should not be answering the questions your AI agent handles well. They should be resolving the exceptions it cannot. That changes who you schedule:

  • Fewer generalists, more specialists in payments, logistics and returns during the November window.
  • A dedicated supervisor per shift who watches escalation trends in real time and can pause an automated flow when something breaks upstream, for example a courier outage.
  • Clear authority limits. Decide in advance what a front-line agent can refund or compensate without approval, so peak-day decisions don’t queue behind a manager.

Step 5: Run a Post-Peak Review Before You Forget

Book the review for the last week of November, while the details are fresh. Four questions cover most of the value:

  1. Which intents exceeded the forecast, and by how much?
  2. Which automated answers were wrong, and why was the underlying knowledge out of date?
  3. Where did customers repeat themselves after a handoff?
  4. Which escalation rules fired too often or too rarely?

A Four-Week Countdown

When Focus Done when
Now to October 1 Golden Week readiness Holiday delivery and store-hour answers verified; courier schedules loaded
First half of October Presale readiness Deposit and coupon rules reviewed in shadow mode; campaign articles carry end dates
Late October 11.11 readiness Peak escalation rules live; specialist rota published; proactive order notifications tested
Week of 11.11 Live operations Supervisor on every shift; daily review of escalation and error trends

Peak season rewards the teams that did their thinking in September. Forecast by intent, test your agent against last year’s reality, and put your people where the exceptions are, and Q4 becomes a set of known problems rather than a surprise.