What is collision detection?
Collision detection is a shared-inbox safeguard that shows, in real time, when another agent is viewing or typing in the same conversation. When Neha opens a chat that Vikram is already answering, she sees a banner such as "Vikram is typing" — and, in stricter setups, her composer is locked or her draft is flagged as stale when his reply goes out.
The problem it solves is specific to team inboxes. Several agents share one WhatsApp business number, so anyone with access can reply to any thread. Without a warning, two people answer the same question, sometimes with different refund amounts or delivery dates, and the customer gets both. The WhatsApp Cloud API will not stop it: it sends every message your server submits, and has no concept of agents.
Implementations use presence events between the agents' browsers and the inbox server: who has the conversation open, who is typing, who sent last. Common behaviours are a viewing indicator, a typing indicator, a soft lock that warns before sending, a hard lock that only lets the owner reply, and a "new message since you started typing" check that makes the agent reread before sending.
Collision detection and auto-assignment work together. Auto-assignment gives each conversation a single owner, so most collisions never happen; collision detection covers the gaps — supervisors jumping in, unassigned queues, chat transfer moments and agents helping a busy colleague. It is unrelated to the WhatsApp typing indicator the customer sees, which is a Cloud API feature shown on the customer's phone.
Key takeaways
- Collision detection warns agents in real time when a colleague is viewing or typing in the same conversation.
- It prevents two agents from sending duplicate or contradictory replies to one customer on WhatsApp.
- It is an inbox feature: the WhatsApp Cloud API sends whatever your server submits and does not know about agents.
- Auto-assignment prevents most collisions by giving each chat one owner; collision detection catches the rest.
Why the difference matters
Why does collision detection matter?
Two replies to one question look careless, and two different answers erode trust fast — the customer will hold you to the better offer. Duplicate replies also cost money: each service message has been charged per delivered message in India since 1 October 2026. In busy shared queues, collisions are common enough that a simple presence banner pays for itself quickly.
Common mix-ups
Common mistakes with collision detection
Relying on it instead of ownership
Banners help, but an inbox where everyone works one unassigned queue will still collide. Assign an owner to every chat and use detection as the safety net.
Stale presence
An agent who closes a laptop can leave a "viewing" badge for hours and scare everyone off. Expire presence after a short idle period.
Hard locks with no override
If only the owner can reply and the owner is offline, the customer waits. Let a supervisor take over, and transfer the chat when it happens.
Ignoring bots and automations
An auto-reply or workflow sending at the same time as an agent is also a collision. Pause automations once an agent takes the conversation.
How does collision detection work?
- Step 1
Broadcast presence
When an agent opens a conversation, their browser tells the inbox server, which shows a viewing indicator to anyone else on the same thread.
- Step 2
Signal typing
Keystrokes in the composer send a throttled typing event, so colleagues see who is actively drafting a reply rather than just reading.
- Step 3
Check before send
On send, the server compares the agent's last-seen message with the latest one. If a colleague or the customer posted in between, it warns and asks the agent to review.
- Step 4
Apply the lock policy
Soft locks warn but allow sending; hard locks let only the assigned owner reply, while others leave internal notes. Choose per team.
- Step 5
Release and log
Presence clears when the agent leaves or goes idle, and any override of a warning is logged so supervisors can review it.
Example: Two agents, one refund in Kochi
At a Kochi travel agency, a customer asks on WhatsApp for a refund on a cancelled houseboat booking. Anjali opens the chat from the unassigned queue and starts typing. A minute later Thomas opens the same thread from a search result and sees "Anjali is typing". He adds an internal note with the booking reference instead of replying. Before collision detection, the agency had sent customers two different refund amounts in the same minute more than once a week.
Collision detection: key facts and limits
- Signals shown
- Viewing, typing, last sender, new-message-since-draft
- Lock styles
- Soft lock (warn) or hard lock (owner only)
- Where it runs
- Your inbox; the Cloud API does not detect collisions
- Best paired with
- Auto-assignment, so each chat has one owner
- Not the same as
- The customer-facing WhatsApp typing indicator
Rules of thumb
Best practices for collision detection
- Show presence on the conversation list too, not only inside the thread.
- Make the "new message since you started typing" check mandatory for refunds and price quotes.
- Encourage helpers to use internal notes rather than replying in someone else's chat.