Inbox and conversations

Your Inbox holds every conversation your AI has, across every channel you have connected. It is where you check quality, and where you step in when a customer needs a person.

Reading a conversation

Open any conversation to see the full exchange. The details panel next to it shows the context your AI had:

• who the visitor is (or Anonymous if they never gave a name)
• whether they are Online now, or when they were last seen
• Pages visited on your store before and during the chat
• whether a lead was captured, or an order question was involved

Conversations are marked so you can scan them: Order inquiry, Lead captured, Handed Off.

That page history is often the fastest diagnosis of a bad answer. A visitor sitting on a product page while the AI talks about something else tells you exactly what went wrong.

Did this conversation make a sale

When Storebird can tie a purchase to a conversation, you see it in the inbox on the right of the row, under the time — and in full in the details panel when you open the conversation.

There are four things you can see there, and they do not mean the same thing:

• An amount (for example €16.00, always in your store's own currency). The strongest case, and it has to pass three tests: that same browsing session opened your order confirmation page for that order, your store reported the amount at the moment of payment, and the customer actually said something of their own before buying. That last test matters more than it sounds — if someone only tapped one of your suggested questions and never typed a word, we show the timing instead of a number, because a tapped button is not a conversation. Open the conversation to see which order it was, when it was bought, and how long after the chat started.
• "Order 23 min later". An order was linked to this conversation, but we could not prove the chat is what produced it — the billing email matched, or the visitor never typed anything before checking out, or we never recorded when the purchase happened. You get the timing and deliberately no amount: we would rather show you less than claim money that is not ours to claim.
• "Purchase recorded". A sale was booked against this conversation before we started storing how it was matched. Until it has been re-checked, no amount is shown.
• "Checked out". The visitor opened an order confirmation page while the chat was open, but the order value never reached us. This is the most common one of the four, and it is not a fault in the conversation: order values are only reported from WooCommerce plugin 1.6.4 and up.

An empty slot means there is no purchase signal — it does not mean the conversation earned nothing. Someone can read your AI's answer and buy an hour later on their phone; we would rather say nothing than put a zero next to a chat that did its job.

The same rule drives Revenue Generated on your dashboard, so the two screens can never tell you two different stories. See Order tracking for how an order gets linked in the first place.

Photos from customers

Customers can attach a photo in the chat (up to 10 MB per image, JPG/PNG/GIF/WebP). Your AI looks at the image and answers about what it sees — handy for "is this the right part?", a damaged item, or a screenshot of an error. The photo is stored with the conversation, so you see exactly what the customer sent when you open it in the Inbox.

If a customer says they sent a photo and your AI says it did not come through, ask them to send it again — a very large image can fail to upload.

Taking over

Click Take Over to step into a conversation yourself. Your AI stops answering, the customer is told a person has joined, and you type directly.

When you are done, the conversation can be Returned to AI so it keeps handling follow-ups.

See Human handoff and WhatsApp for handling this from your phone.

Statuses

Conversations move through states — Active, Handed Off, Resolved, Abandoned, Reopened — so you can see at a glance what still needs attention.

You can define your own statuses to match how you work at Conversation statuses. Use few and use them consistently; a status nobody sets is worse than none.

Labels

Labels are for themes rather than progress: sizing, delivery complaint, wholesale request, bug.

Labels are how you turn support into product insight. After two weeks of labelling you can see which question dominates, and whether the answer belongs in your knowledge base, on a product page, or in a fix to the shop itself.

Archiving

Archive a conversation to take it out of the active list without deleting it. Unarchive brings it back. Archived conversations stay searchable and still count in your reporting.

Filtering by channel

If you have more than one channel connected — widget, email, WhatsApp — you can filter the inbox per channel, or look at everything together. Same brain, same customer history, one list.

The daily summary email

Each morning you get a short email with the previous day's chats per store, so you can scan them without opening the dashboard. It counts only real visitor conversations — your own test chats, our smoke tests and traffic we classify as automated bots are left out, so the numbers match what your customers actually did.

Two lines deserve a precise reading:

• Handled without your team — conversations where nobody from your side was pulled in. That is a workload number, not a quality score: it says no human was needed, not that every answer was right. Open a few conversations to judge the answers themselves.
• Contact details captured — unique people who left a way to reach them. A visitor who comes back three times counts once.

Conversations that need attention lists the handoffs nobody closed yet, and tells you per conversation whether you can still reply by email or whether no contact details were captured. Times in the email are shown in UTC.

What to look at each week

• Conversations marked Handed Off: did a person have to step in, and why?
• Conversations that ended without resolution: what was missing?
• The answers that made you wince: fix them with Improve this answer — see Train your AI

Ten minutes of reading beats an hour of guessing at settings.