How to size your WhatsApp sales chat team for Black Friday

Key takeaways

Team needed = (conversations at the peak × active minutes per conversation) ÷ productive minutes per agent in the same time slot.
A schedule that works on the daily average fails at the busiest hour; size for the peak, not the average.
According to the Harvard Business Review (2011), replying within 1 hour gives about 7 times the chance of qualifying the lead; 23% of the companies audited never replied.
Replying within 24 hours to a click-to-WhatsApp ad opens a 72-hour free entry point window, according to Meta (2025); a slow queue costs money in messages.

Why size by workload, and not by number of conversations?

Because "what's the price?" takes 2 minutes and "I want to change the size on the order I paid for yesterday" takes 15. Counting conversations without measuring the work produces a schedule that is wrong in both directions: idle people at 9 am and a queue at 8 pm.

Three factors change real capacity:

Distribution across the day. According to Opinion Box (2025), 61% of Brazilians open WhatsApp several times a day, and the buying peak rarely matches business hours.
Complexity. Price questions, negotiation, declined payments and complaints take very different amounts of time. Measure the active time of each type in your history.
Returns. A lead who asks, disappears and comes back three times costs three interactions, and on Black Friday that goes up because customers compare offers.

Active time is the time the agent spends reading, typing or resolving the conversation; open duration is the time between the first and last message. A conversation can stay open for 3 hours with 6 active minutes.

How to calculate how many agents you need (hypothetical example)

Calculate per time slot, using the history of your best weeks and the campaign's traffic projection. Hypothetical example: a store projects 360 new conversations per day during Black Week.

Active minutes per conversation: from history, 8 minutes on average (simple question 3, negotiation 12, after-sales 15, weighted by frequency).
Total work for the day: 360 × 8 = 2,880 minutes, or 48 active hours.
Productive hours per agent: 6 in an 8-hour shift (the rest is breaks and admin tasks). 48 ÷ 6 = 8 agent-days, if conversations arrived evenly spread.
Real distribution by time slot: from history, 35% of conversations arrive between 7 pm and 10 pm (126 conversations in 3 hours). Work at the peak: 126 × 8 = 1,008 minutes in 180 minutes → 1,008 ÷ 180 = 5.6 agents in that slot alone, assuming 100% occupancy.
Realistic occupancy: nobody handles chats 100% of the time without the queue blowing up. At 80% occupancy, 5.6 ÷ 0.8 = 7 agents at the 7 pm to 10 pm peak.
Rest of the day: the other 234 conversations over 12 hours require 234 × 8 ÷ 720 = 2.6 → 3 to 4 agents at 80% occupancy.

The right schedule is not "8 agents all day": it is 3 to 4 during the day and 7 from 7 pm to 10 pm.

Decision rule: if you cannot put 7 people on the peak, choose between reducing evening traffic, shifting ad budget to the hours when the team is there, or using an AI Sales Assistant in automatic mode for simple questions, with a human taking over when needed. Reducing the target beats building up a queue.

How to build the schedule and who handles what?

Build the schedule by time slot, with separate roles: intake, closing and exceptions. Mixing everything makes the person negotiating a R$ 900 order drop that customer to answer "what's the price?".

Role	What it does	Who	Reference capacity
Intake (triage)	First reply, simple questions, sending the offer	Junior agents or AI Sales Assistant in copilot mode	60 to 80 conversations/day per person
Closing	Negotiation within authority, sending the link, payment confirmation	Experienced salespeople	25 to 40 conversations/day per person
Exceptions	Declined payment, complaint, exchange, customer asking for a human	Manager or senior agent	15 to 25 cases/day

The capacity numbers are hypothetical references; replace them with your own. Two rules apply to any business: every conversation has an owner and a next action (a CRM with a funnel and an owner per lead, such as RASTRO, shows how many conversations each person has open and for how long), and whoever closes does not do triage. How to separate the leads is in lead qualification without stalling sales.

Which triggers call for reinforcement or a brake on traffic?

Define before the campaign the queue size that triggers reinforcement and the queue size that triggers a brake on ads. Deciding that in the middle of the peak, with 80 conversations waiting, is too late. A template to adapt:

Signal	Threshold (example)	Action	Who decides
Conversations with no first reply	more than 10 for more than 15 minutes	Call in the standby reinforcement	Queue manager
First response time (p90)	above 30 minutes for 1 hour	Cut the daily ad budget by 30%	Media buyer
Queue keeps growing after reinforcement	more than 30 unanswered	Pause click-to-WhatsApp ads until the queue drops	Owner

The case for acting fast comes from James Oldroyd's Lead Response Management Study (MIT, with InsideSales.com; vendor-funded, treat as an order of magnitude): the odds of reaching a lead when replying within 5 minutes are about 100 times higher than at 30 minutes. In the Harvard Business Review (2011), 24% of the 2,241 companies audited took more than 24 hours to reply. Related reading: slow replies on WhatsApp cost sales.

Does an AI Sales Assistant replace the schedule?

It does not replace it; it reduces intake work and leaves humans for closing and exceptions. Treat the AI as one more agent on triage, with clear limits:

Start in copilot mode (the AI suggests, the human sends) two weeks ahead, to check that the answers match the official offer.
Automatic only for simple questions (price, delivery time, installments) and only with the offer and conditions loaded in writing.
A human takes over when the customer asks or when negotiation, a complaint or a declined payment comes up; the AI must pause when the human takes over.
Measure separately conversations resolved by the AI, transferred, and sales that started with the AI. If transfers exceed 50%, go back to copilot.

More in AI selling with control over the offer.

Which common mistakes bring down customer service on Black Friday?

The four most common mistakes:

Scheduling by the average. 8 people all day costs more and still leaves a queue at 8 pm. Size for the peak.
Rewarding fast replies that do not resolve anything. "Be right with you" in 1 minute and the real answer in 40 is worse than 5 minutes with the right answer.
A queue with no owner. A conversation that "belongs to everyone" belongs to no one.
Forgetting after-sales in the schedule. After Friday come "where's my order?" and exchanges; late delivery was 32.19% of Procon-SP complaints (2025). Reserve people for the following week, as shown in a 30-day plan to sell on WhatsApp.

What to measure in customer service during the campaign?

Measure queue, first response time (median and p90) and pending items per agent. The volume of messages answered does not show whether the sales chat is selling.

Metric	How to calculate	Good	Bad
First response time (median)	time between the customer's message and the first useful reply, middle value	up to 5 minutes during service hours	above 30 minutes
First response time (p90)	the value below which 90% of conversations fall	up to 15 minutes	above 1 hour (10% of leads wait more than 1 hour)
Pending items per agent	open conversations with no next action for more than 2 hours, per person	fewer than 5	more than 15 (redistribute)
Conversations unanswered within 24 h	ad conversations unanswered within 24 h ÷ ad conversations	zero	any value (loses the 72 h free window)

The median hides the queue; the p90 reveals it. To measure this, the CRM needs to record the time of every message and the owner of each conversation; the customer service analysis (response time and follow-up per agent) is the report the owner opens every day of Black Week. How to read that dashboard is in dashboard: what the business owner should look at.

Frequently asked questions

Sources

How many agents do I need for Black Friday?

It depends on conversations at the peak hour and the minutes of work per conversation. Multiply peak conversations by active minutes, divide by the minutes in the time slot and then by 0.8 (realistic occupancy). With 126 conversations in 3 hours and 8 minutes per conversation, that is about 7 people at that peak.

Does AI replace the whole customer service schedule?

Do not assume so. AI reduces triage work and simple questions; negotiation, complaints, declined payments and customers asking for a human stay with people. Measure what the AI resolves without a transfer and keep human coverage.

What to do when the queue blows up in the middle of the day?

Fire the trigger agreed beforehand: standby reinforcement, then a cut in the ad budget, then a pause if the queue keeps growing. Reply to everyone within 24 hours, even if only to give a timeframe, because that is what opens the 72-hour free window (Meta, 2025).

Is it worth hiring a temporary agent just for Black Week?

Yes, if they are trained on the official offer and the 10 most frequent questions and placed on intake, never on closing. A temp who promises a discount outside their authority costs more than the queue. Train them at least a week ahead.