On this page8 sections
- The formula: agents busy = arrivals × duration ÷ concurrency
- Add headroom for your first-response target
- A worked example for a D2C brand
- Split the table by language and skill
- Plan around the 24-hour window, not round-the-clock cover
- What staffing does to your message bill
- Review the plan every week
- Running the numbers in InfiQ
To work out how many agents WhatsApp support needs, size it hour by hour: chats reaching agents in an hour × minutes each chat stays open ÷ 60 ÷ chats each agent runs at once gives the agents busy in that hour. Add headroom for your first-response target, build shifts around the result, then add leave.
The formula: agents busy = arrivals × duration ÷ concurrency
This is Little's law from queueing theory applied to a chat queue: the average number of conversations in progress equals the arrival rate multiplied by how long each one lasts. Divide that by the number of chats one agent works at once and you have the agents who are busy, on average, in that hour. It assumes WhatsApp already carries your support; if you are still deciding, start with whether WhatsApp suits customer support.
Three inputs, all from your own inbox data:
- Arrivals per hour (A): new conversations that reach a human in that hour, after the bot has closed what it can. Count them by hour of day in IST and by weekday.
- Chat duration (D): minutes from assignment to resolution, measured at the concurrency your team runs today. This is wall-clock time, not typing time.
- Concurrency (C): the average number of chats an agent has open during that hour, not the cap you configured. The agent concurrency entry explains why the two differ.
Agents busy = A × D ÷ 60 ÷ C.
Pull four ordinary weeks and take the median for each hour block. Build a separate profile for sale periods such as an end-of-season sale or Diwali week, because an average across both hides the peak you are staffing for.
Add headroom for your first-response target
The formula sizes for average load. Staff exactly to it and every burst of arrivals waits for a free slot, and that wait is your first response time. The headroom you leave is the lever that controls it.
Express headroom as a planning utilisation: the share of agent capacity you expect to use. Seats = agents busy ÷ planning utilisation, rounded up. A lower figure buys faster first replies at the cost of more idle capacity.
Pick a starting figure, then let the data correct it. For the first two weeks, compare the 90th-percentile first response time in each hour block with your target. Where a block misses, add a seat to it; where it beats the target comfortably, take one away. Calculators built for voice queues assume one call per agent, so treat their output for chat as a starting point, not an answer.
A worked example for a D2C brand
The numbers below are hypothetical, chosen to show the arithmetic. They are not benchmarks.
A skincare brand in Pune has about 1,150 conversations a day starting on WhatsApp. Its bot fully resolves 40% of them (order status, delivery dates, store timings), so about 690 reach agents. Chats stay open 18 minutes from assignment to resolution, agents average 3 open chats at a time, and the team plans at 85% utilisation. Agents work 09:00 to 21:00 IST, and the first block includes the overnight queue.
| Hour block (IST) | Chats reaching agents per hour | Agents busy (× 18 ÷ 60 ÷ 3) | Seats at 85% |
|---|---|---|---|
| 09:00–12:00 | 45 | 4.5 | 6 |
| 12:00–15:00 | 90 | 9.0 | 11 |
| 15:00–18:00 | 55 | 5.5 | 7 |
| 18:00–21:00 | 40 | 4.0 | 5 |
Two overlapping nine-hour shifts cover this table: 6 agents on 09:00–18:00 and 5 on 12:00–21:00. That puts 6, 11, 11 and 5 agents on the floor across the four blocks.
The 15:00–18:00 block carries 4 more agents than it needs, and that is where the one-hour breaks go. Eleven breaks spread over those three hours take about 4 people off the floor at a time, leaving the 7 the block needs, so no other block drops below its seat count.
To turn 11 agents a day into headcount, cover seven days with six-day weeks and add an allowance for leave and absence (10% here): 11 × 7 ÷ 6 = 12.8, then 12.8 ÷ 0.9 = 14.3, so the roster needs 15 agents.
Bot containment moves the answer more than most inputs. If the same bot resolved 50% instead of 40%, the peak block would fall from 90 to 75 chats an hour and its seats from 11 to 9. Measure your containment rate honestly before you hire, counting only chats the customer did not reopen.
Split the table by language and skill
One seat table works only if any agent can take any chat. Most Indian teams cannot: a Tamil-speaking customer in Chennai needs a Tamil-speaking agent, and a refund dispute needs someone with refund permissions. If routing sends chats by language or skill, each queue needs its own table, because an idle Hindi agent cannot absorb a Tamil backlog.
Run the same formula per queue, with that queue's arrivals, duration and concurrency, and round up per queue. Small queues lose the most to rounding: a queue that needs 1.2 agents in a block still needs 2 people on shift. Two habits keep this affordable. Cross-train a few agents in two languages so a small queue can borrow them, and route only the chats that truly need a specialist, leaving order-status and delivery questions in the general queue.
Plan around the 24-hour window, not round-the-clock cover
Meta's pricing documentation defines the customer service window: when a user messages you, a 24-hour timer starts, and it resets if they message again before it expires. You can send free-form, non-template replies only while it is open. Once it closes, only an approved template reaches the customer.
That rule sets your real coverage requirement. A chat that arrives at 22:30 IST can still be answered without a template until 22:30 the next day, so a 09:00 start leaves a wide margin. What breaks the model is a gap longer than 24 hours, such as a Sunday with nobody on shift: chats from Saturday evening reach Monday's team with the window shut.
Overnight automation is allowed, with a condition. The WhatsApp Business Messaging Policy permits automation when responding in the 24-hour window, but requires prompt, clear and direct escalation paths, such as an in-chat transfer to a human agent, a phone number, email or an in-store visit. When no agent is on shift, the away message should name the path that is open and the time the team starts.
What staffing does to your message bill
Since 1 October 2026 Meta has charged for service messages, the free-form replies sent inside the window, which had been free since 1 November 2024. Meta's notes on the change treat any non-template message not powered by Meta Business Agent as a service message, so your own bot's replies count too. Each business phone number gets 1,000 free delivered service messages a month, and charging starts at the 1,001st. On InfiQ's India rate card a service message costs ₹0.145 per delivered message, ex-GST.
That makes messages per resolved chat a cost metric, not just a style point. In the example, agents resolve about 20,700 chats a month (690 a day × 30). If coaching and better saved replies remove one message from the average chat, the saving is 20,700 × ₹0.145 a month at the rate charged since 1 October 2026, and shorter chats lower D, which lowers the seat count too.
Review the plan every week
A staffing plan is a forecast, so check it against what happened. Each week, compare these for every hour block:
- arrivals after the bot, forecast against actual;
- 90th-percentile first response time against target;
- chat duration and average open chats per agent;
- chats that ran past the 24-hour window, which should be close to zero.
If arrivals drift, update the profile. If arrivals hold but response times slip, look at duration and concurrency before adding people. New joiners need a lower concurrency cap for their first weeks, so count them as part of a seat until their duration matches the team's. For a sale week, build the seat table from the previous sale's hourly profile and add temporary agents to the peak blocks only.
Running the numbers in InfiQ
InfiQ's shared team inbox reports response times, resolution times, conversations per agent, CSAT and SLA breaches, and those reports export to supply the inputs above. Auto-assignment runs round-robin, skill-based or team-based, out-of-office hours can route to the bot or a queue, and first-response and resolution timers raise breach alerts, as the guide to setting up SLA rules shows. Lite includes 10 users, Growth 25 and Enterprise a custom number, so match the plan to your roster.

