Preparing customer support for Black Friday and peak season

Black Friday does not create new support problems. It multiplies the ones you already have, in a week when every hour of agent time is worth more than usual. Teams that come through it well have generally done the same unglamorous things in September and October: worked out how large the spike will actually be, automated the one category that dominates it, written the articles that carry the exceptions, and decided in advance what a human still has to see.

Peak support volume is a lagging indicator of orders

Support volume does not spike once. It spikes twice: a modest bump during the sale itself — stock, sizing, delivery cut-offs, discount codes — and a much larger one seven to twenty-one days later, when everything that was ordered turns out to be late, wrong, or on its way back. Teams that staff for the Friday and stand down in December get the second wave badly wrong.

The model is simple: tickets scale with orders, with a lag and a multiplier. The arithmetic below uses round numbers so you can substitute your own.

Two corrections, both in the uncomfortable direction. Contact rates usually rise in peak rather than staying flat, because a larger share of orders are gifts, first-time customers, and promotions with conditions attached. Handling time rises too, because more conversations involve an exception. Plan with a contact rate above your quiet-period one, never equal to it.

What to automate first

If your time is limited — and in October it is — spend it on the largest, most repetitive category rather than on breadth. In almost every e-commerce queue that is order status.

1. Order status and tracking. Normally the single biggest category and almost entirely mechanical: look up the order, say where it is, explain what the carrier status means. An AI agent with read-only, email-matched access to the store handles this without a person.
2. Delivery cut-offs and shipping times. In peak these change weekly. Put the dates in an article rather than in a banner nobody reads, so the answer is retrievable.
3. Returns and the extended holiday window. Most stores extend the window for the holidays and then answer the same question about it several hundred times.
4. Discount code conditions: which codes stack, what is excluded, and what happens when a customer forgot to apply one.
5. Stock and restock questions. Even a plain "we cannot promise a restock date" is a better automated answer than a place in a queue.

Order status deserves to be pulled out on its own, because it is the category where automation both removes the most work and improves the customer's experience — the honest answer is a lookup, and the customer wants it at 22:00 on a Sunday rather than on Monday morning. The safe pattern is read-only access, matched against the email address the customer is chatting from, with a neutral not-found on a mismatch. What that looks like for a store is covered on the e-commerce page, and step by step in automating Shopify support with AI.

The knowledge-base checklist

Run this in October, not in the last week of November. Every line is an article, and every article you skip is a conversation that reaches a person during your busiest week.

• Last order dates for pre-Christmas delivery, per country and per shipping method, written out as dates.
• The extended returns window: exact start and end dates, what it covers, and what it does not.
• What happens when a parcel is marked delivered and was not received, including how long you wait before opening a carrier investigation.
• Gift orders: gift receipts, returns by someone who is not the buyer, and whether the price is visible in the parcel.
• Discount and bundle conditions, including exclusions and whether codes stack.
• Out-of-stock and backorder policy, with what the customer can expect and by when.
• Carrier-delay language: what you say when an entire network is running late, which is a normal December event rather than a per-order problem.
• Your peak opening hours, the days you are closed, and what happens to a message sent at 23:00 on a Saturday.

Test each of them the way you would test a release: ask the question the way a customer would phrase it, and read the answer. The interactive demo is a quick way to try phrasings, and the structural rules that decide whether an article is answerable at all are in how to write help-center articles an AI can answer from.

Staff the escalation path, not the queue

The instinct is to hire temporary agents to absorb the volume. The better use of the same budget is usually to automate the repetitive layer and staff the layer above it properly, because temporary agents are least effective exactly where they are most needed: the unusual, judgement-heavy conversations that require product knowledge you cannot transfer in a two-hour briefing.

• Agree the handoff rules before peak: complaints, anything mentioning a chargeback or legal action, anything where the customer has already asked twice, and anything that moves money.
• Route by topic rather than round-robin. One returns specialist clearing returns is worth several generalists doing everything.
• Give temporary staff a narrow, well-documented lane — order-status exceptions, for example — rather than the full queue.
• Decide the out-of-hours plan explicitly. In peak, a closed widget is a lost order: either the AI covers those hours, or the widget collects the question and promises a response time you will actually meet.
• Protect one person per day from the queue entirely, to fix content as gaps appear. It is the highest-leverage hour of the day and it is always the first thing cut.

What to watch during the week

Peak dashboards are usually full of numbers that are interesting afterwards and useless during. Four are worth looking at every morning.

That last one is why our own counting applies a three-day settling window instead of counting the moment the AI stops typing — the rules are written out on how we measure AI resolutions. In a week where volume is four times normal, the difference between a generous and a strict definition is a number you will act on wrongly.

A four-week countdown

1. Four weeks out: pull last year's peak tickets, group them by topic and rank them. Estimate this year's volume from your commercial forecast and your own contact rate.
2. Three weeks out: write the checklist articles above. This is the week that decides your peak.
3. Two weeks out: connect the store if it is not connected, verify that order lookups are read-only and email-matched, and test fifty real questions end to end.
4. One week out: set handoff rules, opening hours and out-of-hours behaviour. Brief temporary staff on their lane. Agree who fixes content daily.
5. During: watch the four signals, write one article a day from the top handoff reason, and resist changing anything structural mid-week.
6. After: the delivery and returns wave is still coming. Keep the content loop running through December and January — the articles you write then are the ones that answer next year's peak.