Conversations AI

Description: Today, new conversations in Inbox aren’t assigned to anyone. Someone has to watch the inbox and assign each one by hand. For businesses with several staff or departments, conversations get missed, replies are slow, and nobody clearly owns each customer. What we’d like: Assignment rules that automatically give each new conversation to a user or team when it arrives, based on: Receiving address or number. For example, email sent to sales@ goes to the sales team, support@ goes to support, and a second SMS number goes to the front desk. Channel. For example, web chat, SMS, email, WhatsApp, Facebook, or Instagram. AI hand-off. When the AI Receptionist hands a conversation to a human, it goes to the right person or team instead of sitting unassigned. Nice to have: A round robin option within a team, skipping users who are inactive or away (see the related “Round robin assignment in automations” request) A fallback assignee when no rule matches A notification to the assignee when a conversation is assigned to them An “Assigned to me” filter in Inbox Impact: Faster replies, no missed conversations, and clear ownership. This matters most for multi-department and multi-location businesses that send several addresses or numbers into one Inbox. It would also let partners sell Inbox and the AI Receptionist to larger teams. Requested by: Partner ID: N9QP

Phone numbers captured without a country code default to +1. For businesses outside North America, for example in Germany, every such number is stored wrongly. Request: a default country code per account, taken from the business profile, used whenever a phone number has no country code. Impact: correct phone numbers for every contact outside the US and Canada.

For website widget conversations, ListMessages returns no aiThoughts and no function_start events, and the conversation activity does not show tool calls either. Tests in the channel guests actually use cannot check whether the right tools were called, and live conversations cannot be analysed for tool errors. In the in-app chat this information is available. Request: expose tool calls (name, parameters, duration, status) for widget conversations in ListMessages and in the conversation activity. Impact: complete acceptance tests in the guest channel, and faster error analysis of live conversations.

Applying the web chat component of a Blueprint created a second widget in the target account, named like the source account's widget. Both widgets are active as triggers. Every target account already runs its own widget on its website, so after every apply a second active widget with a foreign name has to be found and handled by hand. Request: let the web chat component update the target account's existing widget instead of creating a new one, or at least let us choose the widget in the apply dialog and warn before a new one is created. Impact: Blueprints can roll out web chat settings without manual clean-up in every account.

On WhatsApp a guest writes and a team member answers in the inbox before the AI does. Exactly 60 minutes after the last team message the AI answers the same guest message, as if the team had not replied. We found this 30 times in 26 conversations, always 3,607 to 3,651 seconds after the team message. In at least 3 cases the late AI reply contradicted the team, for example "fully booked" after the team had offered a room. The fixed interval suggests a pause timer after a human reply that ends with an answer to the last unanswered guest message. Request: after a human reply, the AI does not answer guest messages the team has already answered and resumes only on new guest messages. The pause after a human intervention is documented and configurable per account. Impact: no late, contradicting answers to questions the team has already handled.

When we get spam messages in our inbox. Can we have in the dropdown shown a note that this is spam and that will also remove the contact from our CRM or block them? Some clients get a lot of spam messages from social media.

The floating chat widget currently opens automatically when a page loads. There is no setting in the web chat configuration panel to suppress this behaviour. For most use cases, auto-open is disruptive — it fires before the visitor has read any page content and creates a poor first impression. The widget bubble alone is sufficient as a passive entry point. Requested: A toggle or configuration option to keep the widget in its collapsed/bubble state on page load, opening only when the visitor clicks it. This is standard behaviour on most chat widget platforms and would make the Conversations AI widget more deployable across professional and enterprise websites.

When the website Agent is engaged, a notification should be sent to a specific user that activity is happening. Notify by email or text message.

In the future I would like to see a way to change the Automatic Refresh for the AI's knowledge source to be more often than monthly, without having to refresh manually. For example, we have a client that is a live event space, and their chatbot conversations are usually about current shows that could have been posted on that site earlier that day but the AI wouldn't be able to see that new info.

The Appearance panel allows primary colour, accent colour, and text colour to be set. This covers the header bar and chat bubbles. The message area background — the space where the conversation sits — is not controllable from any available setting. It defaults to white and cannot be changed. For operators embedding the widget on a dark-themed site, this creates a mismatch that the current colour controls cannot resolve. The header and bubbles can be themed; the conversation background cannot. The request: add a message area background colour field to the Appearance panel. A secondary request: allow border-radius to be set to 0px across widget elements for brands using a flat design system.