Running one WhatsApp number across a whole team

A shared inbox is one WhatsApp number that several people answer from their own accounts, with each conversation owned by one of them at a time. It is the reason most growing stores move from the free WhatsApp Business app to the API: the app is tied to a device, and a device cannot be in two people's hands.

What changes when you move

	WhatsApp Business app	Shared inbox on the API
People answering	One device	Everyone, from their own login
Who owns a chat	Whoever has the phone	Assigned explicitly
Automation on store events	No	Yes
Approved templates	No	Yes — needed outside the 24h window
Broadcast reach	256 per list, and only to people who saved your number	Governed by your messaging tier

That last row is worth dwelling on, because the 256 figure gets quoted as an API limit constantly. It is not. Per WhatsApp's own FAQ, a broadcast list in the app holds up to 256 contacts and only reaches people who have your number saved. On the API, reach is governed by your messaging tier instead.

The three habits that prevent collisions

1. Assign before you reply. An unassigned chat is everybody's and therefore nobody's. Two people answering the same customer is the failure mode a shared inbox exists to prevent.
2. Use statuses honestly. Open means someone is on it, pending means it is waiting on someone, resolved means done. A queue where everything is "open" tells you nothing.
3. Leave internal notes rather than remembering. The colleague picking this up tomorrow cannot read your mind.

What automation should own, and what it should not

The useful division is by whether the answer is knowable in advance.

• Automation owns: greeting, routing, collecting order numbers, answering documented policy questions, out-of-hours replies, order and shipping updates.
• People own: anything emotional, anything about money outside a standard flow, anything involving a photo, and anything where being wrong is expensive.

Automation should also do the tedious part of handover: assigning to the right team and leaving a note with what it collected. See handing a chat from bot to human.

Routing to the right team

An Assign Chat step can send a conversation to a team rather than a person, which is almost always what you want — an individual is on holiday, a team is not.

A workable structure for a small store is two teams: orders and everything else. Adding more teams than you have people creates queues nobody watches.

The 24-hour window is a team problem too

Your team can reply freely only while the customer service window is open — 24 hours from the customer's last message, resetting each time they write, per Meta's documentation. After that, free text fails and only an approved template will send.

The operational consequence: a chat left over the weekend may not be answerable on Monday. If your team works Monday to Friday, Friday evening enquiries are the ones to clear before you leave. Otherwise Monday's reply needs a template, and most stores do not have one written for "sorry for the delay".

It is worth having one approved for exactly that. A simple utility template that re-opens a conversation is cheap insurance, and it is charged only once, since the customer's reply reopens the window.

Measuring the team, not the bot

Two numbers matter more than message volume: how long customers wait for a first human reply, and how many conversations a person actually resolves. Both live in your analytics; message counts flatter automation and tell you nothing about service.

You can also ask customers directly — a rating question at the end of a resolved conversation feeds customer satisfaction scores.

A sensible starting setup

1. Two teams: orders, and general.
2. An automation that greets, offers three options and assigns.
3. An automation that cancels pending questions when a human replies.
4. An out-of-hours reply that sets the conversation to pending.
5. A rule that Friday's chats are cleared before the weekend.

That is enough structure for a team of two to six people, and it can be built in an afternoon.

The four collisions a shared inbox has to prevent

Collision	What the customer sees	Prevention
Two agents reply at once	Two different answers	Assign before replying
Bot replies over an agent	A robot interrupting a person	The cancel-pending-question flow
Nobody replies	Silence	Statuses that mean something
Agent answers without context	Being asked the same question twice	Internal notes at handover

The second is the one with a specific technical fix rather than a habit: a two-step automation triggered on Agent Sends Message that cancels any pending question. It sends nothing and needs no configuration, and without it your bot will capture a customer's reply to your colleague as an answer to a question it asked ten minutes earlier. Details in handing a chat from bot to human.

Splitting numbers, and when it is worth it

Quality rating is tracked per phone number, not per account. That is a real argument for running marketing on a separate number from order notifications once your campaign volume is significant — a damaged marketing number then does not drag your transactional messages down with it.

It is not a way to escape consequences. Enforcement can reach the account rather than only the number, and opening numbers to outrun a rating reads as evasion. Split for operational clarity, not to dodge — what to do when your account is restricted covers the difference.

For most stores under a few thousand orders a month, one number is simpler and better. Two numbers means two sets of templates, two quality ratings and two messaging tiers to manage.

What automation should own in a team setting

The useful division is by whether the answer is knowable in advance.

• Automation: greeting, routing, collecting order numbers, answering documented policy questions, out-of-hours replies, order and shipping updates, and the tedious part of handover.
• People: anything emotional, anything about money outside a standard flow, anything involving a photo, wholesale and high-value enquiries, and anything where being wrong is expensive.

A useful test for the boundary: if getting it wrong costs more than five minutes of someone's time, do not automate it.

Two numbers worth watching weekly

1. Time to first human reply. The number customers actually experience, and the one automation is meant to protect by absorbing the easy conversations.
2. Share of conversations resolved without a person. If this is not rising after you add a flow, the flow is not doing its job — check its activity view before assuming customers just prefer humans.

Message volume is not on that list. It flatters automation and says nothing about service. Measuring whether your automation works covers what to use instead.

Roles worth separating early

Even a team of three benefits from deciding who does what before the volume arrives, because the alternative is everyone half-watching the inbox.

• Whoever clears the queue first thing. Overnight conversations have a deadline — the 24-hour window may close before anyone replies.
• Whoever owns escalations. Low satisfaction scores and angry messages should reach one named person, not a shared pile.
• Whoever maintains the flows. Keyword lists rot, templates get recategorised, and nobody notices unless it is somebody's job.

A weekly fifteen minutes

1. Any conversation still open from more than two days ago?
2. Any automation with zero runs this week?
3. Quality rating still green?
4. Opt-outs rising faster than opt-ins?
5. Anything the team answered repeatedly that should be a flow or a knowledge-base entry?

The last question is where a shared inbox pays for itself. Every answer typed more than three times in a week is a candidate for automation, and your team already knows which ones they are.

Onboarding someone new

Three things to show a new team member on their first day, in this order:

1. Assign before you reply. The habit that prevents two people answering the same customer.
2. What the statuses mean here. Open, pending and resolved only work if everyone uses them the same way, and "everyone knows" is how that decays.
3. Which conversations are theirs to close and which go to someone else. Wholesale, refunds and complaints usually have an owner.

What they do not need on day one is the automation builder. Answering conversations teaches them what the flows are for, which makes the builder make sense later.

One number, one voice

Customers experience your shared inbox as a single person, so agree a house style early — greeting, sign-off, whether you use the customer's first name, how formal to be. It costs one short conversation and prevents the odd experience of a customer getting three visibly different personalities across one afternoon.

Write the house style down somewhere the team can see it, even if it is three lines in a shared note. Agreements nobody recorded quietly stop being agreements the week someone new joins.