Black Friday customer service, like support for Diwali and other festive sales, is planned around the waves of requests a sale sets off: order and payment questions during the sale, "where is my order" and delivery problems once orders ship, then returns and refunds. Prepare by forecasting each wave from your last sale, letting software finish the routine requests in each, telling customers about delays before they ask, and keeping your human agents for the exceptions.
In short
- A sale changes the mix of requests, not just the number. The busiest week for human agents is often the second or third after the sale.
- Forecast contacts per 100 orders for each request type, then how many of each need a human agent.
- Before the sale, decide what software finishes (order status, changes before dispatch, COD confirmation, returns inside your policy), test it, and freeze changes.
- Send order updates and delay notices before customers ask. In the US a delay notice is a legal duty; in India a late delivery gives the customer the right to a return and refund.
What a sale does to your support queue
For ecommerce customer service, a sale plays out in stages, each with its own requests:
| Stage | When | What customers contact you about | What support needs ready |
|---|---|---|---|
| Sale days | While the sale runs | Coupons that don't apply, sizes and stock, payments debited without an order, cash-on-delivery (COD) confirmations, changes to orders just placed | Fast answers while the customer is still buying; payment and COD checks |
| Dispatch | 1–7 days after the order | "Where is my order?", address changes and cancellations before dispatch | Live tracking data; order changes before the warehouse ships |
| Delivery | 3–21 days after the order | Delays, failed delivery attempts, parcels marked delivered but not received, damaged or wrong items, COD orders refused at the door | Courier escalations, photo checks, replacements |
| After delivery | 1–6 weeks after delivery; into January for US holiday gifts | Returns, exchanges, refund status, complaints | Return pickups, stock for exchanges, refund tracking |
The waves stack. In India, the festive season can bring several sales within a few weeks: a brand's own sales around Navratri and Diwali (8 November in 2026) alongside marketplace events such as Flipkart's Big Billion Days and Amazon's Great Indian Festival. Returns from one sale can arrive during the dispatch week of the next. In the US, Black Friday and Cyber Monday (27 and 30 November in 2026) run into December's shipping deadlines, and gift returns arrive in January, later still if you extend your return window for holiday purchases.
The hardest week comes after the sale. Delivery problems and damaged items arrive while order-status questions are still high, and each needs a courier check, a photo or a decision. So the backlog to plan for is the one after the sale, not the sale day, and it is the one that pushes agents into extended shifts when nobody rostered for it.
For the everyday operating model behind this (queues, tiers and a weekly rhythm), see our guide to customer service operations.
Forecast each wave, not the total
"Three times normal volume" doesn't tell you who to roster in week 2. Forecast by request type instead. Take your last comparable sale's contacts per 100 orders for each type and multiply by the orders you expect. Place each type in the weeks it peaks, then apply the share you expect to need a human agent once automation has done its part. Here it is for a sale expected to bring 50,000 orders, with illustrative rates:
| Request type | Per 100 orders (last sale) | Contacts | Peaks in | Needing a human agent (assumption) | Human-agent contacts |
|---|---|---|---|---|---|
| Offers, stock and payment questions | 5 | 2,500 | Sale days | 20% | 500 |
| Where is my order | 15 | 7,500 | Weeks 1–2 | 10% | 750 |
| Changes before dispatch | 3 | 1,500 | Week 1 | 20% | 300 |
| Delivery problems | 8 | 4,000 | Weeks 2–3 | 40% | 1,600 |
| Damaged, wrong or missing items | 2 | 1,000 | Weeks 2–3 | 50% | 500 |
| Returns and exchanges | 6 | 3,000 | Weeks 3–6 | 25% | 750 |
| Refund status | 4 | 2,000 | Weeks 4–7 | 15% | 300 |
| Total | 43 | 21,500 | 4,700 |
In week 2, half the order-status, delivery and damage work lands at once: about 1,425 human-agent contacts, or roughly 200 a day, against about 1,175 in the sale week. At 8 minutes of work per contact and 80% occupancy, one 8-hour shift clears 48 contacts, so week 2 needs four to five agents working every day. Allow 30% for breaks, training and leave, and five-day working weeks, and that is eight or nine people on the roster. Workforce management covers turning daily numbers into hourly schedules.
If this is your first big sale, start from a normal week's rates and raise delivery problems, because couriers slow down in peak. Plan one incident as a separate scenario, such as a courier hub closing for three days in week 2, instead of padding every week.
Decide what software finishes, then freeze
Next, decide which requests software completes without a human agent, meaning those your written rules and your order, courier and payment data can settle:
- Order status, from live tracking; a human agent only when a parcel is stuck past your limit.
- Address changes and cancellations until the warehouse picks the order.
- COD confirmations before the order is packed (see below).
- Payments debited without an order. The answer is the payment status and the reversal timeline. In India, RBI rules give banks until five days after the transaction to reverse a failed UPI or card payment to a merchant, then ₹100 a day in compensation to the customer.
- Coupon questions, from the sale's terms.
- Returns and exchanges inside your policy, with a photo for damage claims.
- Refund status, from the payment or refund system.
Answering isn't finishing: a bot that explains the cancellation policy still leaves a human agent to cancel the order. That gap is what customer service automation should close.
Then set a freeze date about two weeks before the sale, after which no policy, workflow or integration change goes live without the head of support's sign-off. Several brands we speak to freeze system changes from September until the festive season ends, so a workflow that misses the freeze waits until December. Put sale-specific rules, such as a shorter return window for sale items, into the policy, the AI's workflows, canned responses and the website on the same day.
This is the job we built Flowcall's AI Agent for. It starts from ready-made workflows for returns, refunds, exchanges, cancellations, cash on delivery and damaged products. It runs on the SOP Engine: it follows your procedure step by step, your rules make each decision, and order, delivery and payment details come from your systems. So a workflow that is set up correctly gives the same result every time. Before the freeze, you can run each workflow against simulated customer conversations and rerun the tests after every change. On sale days the same AI works as an AI shopping assistant, finding and comparing products from your catalogue with images, prices and buy links.
Staff human agents for the exceptions
Human agents get the work that needs judgement: delivery disputes, damaged items, complaints, suspected abuse and anything your rules don't cover.
- Give permanent agents the exceptions and complaints. They can make the calls the policy doesn't spell out.
- Train temporary or outsourced agents on two or three request types, such as delivery problems and returns, with the same written rules the AI follows, and finish training three to four weeks before the sale.
- Roster for the backlog, not the sale day. Put extra shifts into weeks 2 and 3, so clearing the backlog doesn't depend on overtime.
- Cover the festival days. Diwali and Thanksgiving are days off for your team but not for customers, and WhatsApp's 24-hour reply window keeps running. A rota, or 24/7 customer support that answers and acts overnight, keeps replies inside it.
- Plan languages. Festive buyers in India write in Hindi and regional languages; see multilingual customer support.
Temporary agents are fastest to train when they never leave the ticket. In Flowcall's AI Helpdesk, for the systems connected to Flowcall, the customer's orders, delivery and payments are on the ticket, Agent Assist drafts the reply, and the refund or replacement takes one click once the agent has checked the details the AI filled in. A new agent learns your rules, not your store admin and courier portals. You choose who handles which channels, and routing sends each ticket to the right team by skill, language or workload, with capacity limits for each channel.
Tell customers before they ask
WISMO ("where is my order?") is the largest request type in the forecast above, and an update that arrives first can prevent the question.
Order updates. Message customers when the order is confirmed, shipped, out for delivery, delayed and delivered: on WhatsApp in India, and by email or SMS in the US. Include the tracking link, the expected date and what to do if it hasn't arrived by then.
Delay notices. When a courier or warehouse slips, message affected customers before the promised date passes, with a new date and a way to cancel. In the US this is a legal duty. Under the Federal Trade Commission's Mail, Internet, or Telephone Order Merchandise Rule, a seller that can't ship within the time it stated (or 30 days, if it stated none) must send a delay notice. The notice gives a revised date (or says it can't give one) and the option to cancel for a full, prompt refund. In India, the Consumer Protection (E-Commerce) Rules, 2020 cover brands selling their own goods online. A brand can't refuse to take goods back, or to refund the customer, when they arrive later than the stated delivery schedule, unless the delay was force majeure (Rule 7(4)).
WhatsApp at sale volume. Three rules from Meta's developer documentation (checked 30 September 2026) decide whether these messages go out:
- Outside the 24-hour customer service window, you can send only approved WhatsApp message templates, and review can take up to 24 hours. Get delay, out-for-delivery and COD templates approved two weeks before the sale.
- Keep sale promotions out of order updates. Templates that mix utility and marketing content count as marketing, and Meta reviews approved utility templates for recategorisation.
- Check your messaging limit: how many different customers your business portfolio can message outside the service window in a rolling 24 hours (250, 2,000, 10,000, 100,000 or unlimited). A delay notice to 30,000 customers needs the 100,000 tier. Above 2,000, Meta raises the limit automatically when quality is high and you've used at least half of it in the past seven days, so build up before the sale. Our WhatsApp broadcast guide covers large sends.
COD confirmation (India). With cash on delivery the customer pays at the door, so an impulse order can be refused and shipped back at your cost, a return to origin (RTO). Confirm each COD order on WhatsApp before it's packed, decide what happens to unconfirmed ones, and offer a small incentive to prepay. Inside the confirmation, the incentive makes it a marketing template, charged at the marketing rate; offered after the customer confirms, it leaves the confirmation utility. See our guide to COD orders.
Flowcall's WhatsApp Marketing sends these updates automatically when an order is confirmed, shipped, delayed or delivered, triggered by events from Shopify, ClickPost or your own systems. A customer's reply can start an AI Agent workflow, so "where is it?" gets an answer from the courier data instead of a queue. When a courier delay hits a set of orders, a broadcast goes to those customers as a segment or an uploaded list. For cash-on-delivery orders from a Shopify store, Flowcall confirms each order on WhatsApp and offers a discount to pay online in the same message, and the customer accepts or cancels in the chat. If a template gets no reply, it sends another or runs the action you choose.
Run the sale: service levels, escalations and incidents
Decide before the sale which targets hold. WhatsApp replies must land inside the 24-hour customer service window, or they have to start with an approved template. Complaints need a fast acknowledgement: in India, a brand's grievance officer must acknowledge one within 48 hours and resolve it within a month (Rule 4(5)). Email order-status questions can wait longer if the auto-reply states the real reply time. Write each promise down as a service level agreement per team.
During the sale:
- Check the backlog by age at the start of every shift, not by count. Take the oldest cases first, and anything close to its messaging window.
- Agree escalation contacts with the courier, warehouse and payment gateway: named people and a response time for the sale weeks.
- Switch to incident mode when one failure drives many contacts. If a courier hub closes or a coupon breaks, send one proactive message to the affected customers, give human agents and the AI the same answer, and name one owner who watches the problem until it's fixed.
Our guides to escalation management and how to handle customer complaints, including public ones, go further.
In Flowcall's AI Helpdesk, SLA targets are set for each team, workflow or channel and follow your business hours and holidays, or run around the clock. The people you choose are alerted before and at a breach, and a breached ticket moves up the queue. Every incoming message is checked for frustration, threats to escalate and requests for an agent, so an angry customer isn't stuck behind order-status questions. When a case needs the warehouse or courier, agents open child tickets for them, and the customer's ticket stays open and owned until each is done.
After the peak: returns, refunds and exchanges
The last wave starts about a week after delivery and runs for the length of your return window. Prepare three things before the sale:
- Returns and exchanges. Hold back stock in popular sizes so exchanges don't turn into refunds. Ask for a photo with every damage claim. If sale items have different return rules, state them before purchase; our refund policy template shows how.
- Refund status. "Where is my refund?" becomes its own queue. Give every refund a timeline when you approve it. India's rules require accepted refunds to be paid within a reasonable period (Rule 4(10)). Under the FTC rule, refunds for orders cancelled over a delay are due within seven working days, card payments included.
- Suspected abuse. Repeated "not received" claims and photos that don't match the order go to a senior agent.
Our guide to ecommerce returns covers the reverse logistics behind this wave.
Flowcall's AI Agent handles the routine cases in this wave under your rules, as it does for returns, refunds and exchanges all year. It collects photos and pickup details in a form inside WhatsApp or Live Chat and assesses the damage photos. It checks the return window and product against your policy, then completes the exchange, replacement or refund in your systems once the required checks pass. When a check fails, it hands the case to an agent with the conversation, photos and what's left to do.
How to tell the peak went well
Review within two weeks of the returns peak, while the team still remembers what broke. Measure the plan, not just the averages:
| Measure | How to count it | What it tells you |
|---|---|---|
| Contacts per 100 orders, by request type | Actual against your forecast, week by week | Which wave you misjudged, for next year's forecast |
| Order-status contacts per 100 shipped orders | Days when updates went out against days they didn't | Whether proactive messages worked |
| Backlog age at the start of each shift | The oldest open ticket and the number older than 24 hours | Delays that averages hide |
| Replies after the messaging window | WhatsApp replies sent after the 24-hour window | Where staffing missed the hours customers wrote |
| Delay notices before the promised date | Share of delayed orders told before the date passed | Compliance and complaint risk |
| Refund time | From approval to the money reaching the customer | The source of after-sale complaints |
| Repeat contacts within 7 days | Same customer, same order, any channel | Answers that didn't settle the question |
| Overtime hours | Hours beyond rostered shifts, by week | The cost the service levels hide |
Put them on one customer service analytics dashboard next to the customer service metrics you track all year.
Flowcall's AI Auto QA gives the review its agenda. Conversation insights group the issues that grew during the sale into operations, delivery, product, policy and resolution problems, with the conversations behind each trend. Automation gap detection groups the requests the AI couldn't resolve into the workflows, knowledge or integrations worth adding before the next sale.
A peak-season checklist
| When | What to do |
|---|---|
| 8 weeks before | Pull last sale's contacts per 100 orders by request type. Decide which requests software will finish. List the sale's policy changes. |
| 6 weeks before | Build or extend the workflows. Hire or brief temporary and outsourced agents. Check every limit that grows with volume: your WhatsApp messaging limit, helpdesk and AI usage caps, and your order system's API limits. |
| 3–4 weeks before | Train temporary agents on two or three request types. Test every workflow on simulated conversations. Submit delay, out-for-delivery and COD templates. |
| 2 weeks before | Freeze changes. Agree escalation contacts with the courier, warehouse and payment gateway. Publish sale policies. Hold back exchange stock. |
| Sale days | Check the backlog by age every shift. Confirm COD orders. Watch coupons and payments for incidents. |
| Weeks 1–3 after the sale | Send delay notices before promised dates pass. Run the extra shifts for the delivery wave. |
| Weeks 2–6 after the sale | Staff returns, exchanges and refund status. Give every refund a timeline. |
| 2 weeks after the returns peak | Review against the forecast and fix three things before the next sale. |
How Flowcall helps with peak season
Flowcall combines an AI Agent, an AI Helpdesk, AI Auto QA and WhatsApp Marketing on the same customer and order data. Order updates, delay notices and COD confirmations go out on their own. The AI Agent finishes the routine requests in each wave under your rules, and human agents work the exceptions from one ticket. That way a sale's extra volume can be absorbed without hiring in proportion to it.
To see Flowcall send order updates, confirm a COD order and handle a damaged-item return on WhatsApp, book a demo.
Frequently Asked Questions
When is peak season for customer service?
In the US, from the week before Thanksgiving through Black Friday, Cyber Monday and the December shipping deadlines, followed by January returns. In India, the festive season runs from Navratri and Dussehra to Diwali, with marketplace sales and brands' own sales in the same weeks. Any big sale of your own creates the same waves on a smaller scale.
How do you prepare customer service for Black Friday?
Start about eight weeks out. Forecast contacts per 100 orders by request type, decide which requests software will finish and test them, train temporary agents on a few request types, get WhatsApp templates approved and agree escalation contacts with your courier. Freeze changes two weeks before, and plan extra shifts for the weeks after the sale.
How many extra agents do you need for a sale?
Work from the contacts that will need a human agent, not total orders: last sale's contacts per 100 orders, times the orders you expect, minus the share software will finish, divided by what one agent clears in a shift. Check the second and third weeks after the sale, not just the sale week.
