The alert reaches whoever can fix it.
When a conversation needs a person, the lead owner knows right away: what happened, the verbatim phrase that triggered it and a button to open that chat.
8 built-in reasons · The quote is checked against the conversation · One alert per reason and per lead
Complaining about a delivery that never arrived, and has written three times already.
«this is the third time I ask about the order and nobody answers me»
Conversations are lost by minutes.
Not because the team does not want to answer, but because nobody was looking at that chat at the exact moment it was being decided.
Eight readings of the chat and three hard facts.
Some are detected by the agent reading the conversation. The others go through no AI at all: they are conditions the system measures by itself.
The AI agent reviews the conversation and can only raise an alert for the reasons that are active in your catalog. Each reason comes with its severity and its repeat window, and both are editable.
- The customer has been waiting too long for a replySeverityWarningDoes not repeat before2 hours
- The customer is annoyed or angrySeverityUrgentDoes not repeat before1 hour
- The customer has a real emergencySeverityUrgentDoes not repeat before1 hour
- The customer asks to speak to a managerSeverityUrgentDoes not repeat before30 minutes
- ComplaintSeverityUrgentDoes not repeat before2 hours
- The sale is about to be lostSeverityWarningDoes not repeat before3 hours
- Someone on the team treated the customer badlySeverityUrgentDoes not repeat before30 minutes
- A question the AI cannot resolveSeverityInfoDoes not repeat before4 hours
These are the eight reasons Victoria ships with. You can pause them, change their severity and their window, or write your own.
- 8built-in reasons
- 3severity levels
- 60messages of window to anchor the quote
Your CRM already notifies. This is something else.
Notifications tell you that something happened. The agent alert tells you a conversation needs a person, and gives you what you need to decide before opening the chat.
A CRM notification
- Tells you something happened — a message came in, a lead was assigned to you, an opportunity deadline is approaching.
- Fired by a system fact — nobody has read what is being said in that chat.
- Shows you the message — whether that is urgent, a complaint or a sale about to fall through is for you to decide by opening the conversation.
- One per message — in an active chat that is dozens, and they all weigh the same.
An agent alert
- Tells you a person is needed — the customer has been waiting for hours, is complaining, is asking for a manager or is walking over to the competition.
- Fired by a declared reason — it comes from a closed catalog you define: if the reason is not active, the agent cannot raise an alert for it.
- Carries the phrase that triggered it — verbatim, checked against the conversation and against who said it. It reads without opening the chat.
- One per reason and per lead — with its severity, so you know what to look at first, and a configurable quiet window.
The two coexist: the notifications you already had keep working exactly the same, and each user chooses which ones to receive.
An alert that interrupts has to earn the right.
The difference between a tool the team looks at and one it learns to ignore is who it reaches and how often.
The alert does not dismiss itself. It exists precisely because someone is waiting on the other side.
The model proposes. The code decides.
An alert that cannot be verified is the word of the AI against the time of the rep. So none goes out without passing the same check.
- The phrase has to exist — the agent is required to quote verbatim. If that phrase does not appear in the last 60 messages of the lead, the alert does not go out.
- And the right person has to have said it — Every reason defines who it expects the phrase from. "The customer is annoyed" requires a message from the customer; "Someone on the team treated the customer badly" requires one from the rep.
- The rejection explains the mistake to the agent — If the phrase is there but someone else said it, the agent is told exactly that, not "quote not found". It corrects against the real error.
- And the quote travels with the alert — The rep reads the phrase on the card, before opening the chat. They decide with the fact in view, not with a label.
That the phrase exists proves the model did not make it up. Who said it is the other half of the fact, and that half is not delegated to the model.
«this is a disaster, I do not know what else to tell you»
- The phrase exists in the conversationFound in the last 60 messages
- The customer wrote itA rep wrote it
Without this check, a message written by the company itself was enough to raise an alert saying the customer was angry.
What gets checked in the first demo.
It ships off and you turn it on
While the switch is at zero the agent does not even have the function, so it cannot interrupt anyone. When you turn it on, it gets it on the next run.
The card does not dismiss itself
It stays there with two actions at hand: open the conversation, or clear it by marking it as seen without going into the chat.
It survives a page reload
The alert is stored as well as shown. If the rep closes the CRM and comes back, the alert is still waiting for them.
The reason decides whose phrase it is
The model does not pick it per alert: it is defined in the reason. A reason about how the team behaved only accepts a phrase written by the team.
With no clear recipient it is not spread around
If the owner of the lead cannot be resolved, the alert is recorded and visible in the panel, but the card is not sent to anyone.
The built-in reasons can be restored
You can edit them with confidence: there is an action to go back to the original catalog without rewriting all eight by hand.
The module that comes next.
Bring a chat that slipped through.
We read it together in the demo and see which reason would have caught it, with what phrase, and who on the team the alert would have reached.
It is switched on with one toggle and the reasons are tuned to how your team works.