Progressive Autonomy

Every ticket is handled at one of seven levels, from Ignore to Resolve. Your administrator’s rules choose the level for each kind of ticket. Capabilities describes the levels. This page shows what each one looks like in the Flexivity panel.

When a ticket needs a change in one of your systems, such as a refund, a second ladder applies: each tool that changes something has one of four write modes, set by your administrator per brand.

Write mode	What you see
Manual	The ticket is handed to you with what to do. Nothing is prepared; you make the change in your system.
Assist	The change is ready on a card in Take Action, with every check Flexivity ran. You approve, edit or reject it. This is the default.
Hold	As Assist, while the customer waits in the conversation. Once you approve, the AI tells them it is done and carries on.
Execute	The AI makes the change itself and tells the customer. The card shows what it did, and you can mark it This change was wrong.

Hold and Execute apply only at Resolve; at every other level they work as Assist. See Prepared Changes for the cards, and Capabilities for how the two ladders work together.

The panel

The Flexivity panel is in the apps sidebar of the ticket. It has three parts:

Take Action holds the cards that need you, or tell you what the AI did. It is hidden when there is nothing to show. When more than one card is waiting on you, the heading shows how many.
Context is open by default: sentiment, summary, attributes, knowledge articles and related tickets. See AI Context.
Activity Log is collapsed by default, and lists what the AI and your rules did on the ticket, newest first.

Each card has a short finding, then what to do about it, then a sentence of explanation. Dismiss clears a card you have dealt with. While the AI is working on a ticket, the panel updates by itself.

Ignore

The AI leaves the ticket alone. Take Action is empty, and no summary or draft is written. Context shows the attributes the AI recorded, and the sentiment. The Activity Log says A rule left this ticket alone.

Label

As Ignore, and the AI also fills in the Zendesk ticket fields your administrator mapped, such as a topic. Those values appear in the ticket’s own fields rather than in the panel, so your triggers and views can use them. The Activity Log says A rule only labelled this ticket.

Gated

The ticket is yours from the start. A card reads Routed to a person by policy → This ticket is yours now: one of your rules sends this kind of ticket straight to an agent, so no reply was drafted. Context has the summary and attributes.

Assist

The AI drafts a reply and you decide whether to send it. This is the default level.

The Suggested Reply card opens with what the AI found in your knowledge base:

Finding	Meaning
Match → Recommendation	One article clearly covers this. Review the draft and send it.
Multiple matches → Compare alternatives	Several articles could apply. Check the draft picked the right one.
Request unclear → Clarification	The request is too vague to answer. The draft asks the customer a question.

Show the draft opens it, with the articles and similar tickets it drew on. Then:

Use this reply copies the draft into the reply field, where you edit and submit it. It never sends anything itself.
Not useful records that the draft did not help.
Draft a new reply writes a fresh draft after the customer has written again. Until then it reads Up to date.

Sometimes there is no draft to use:

No match → Escalate: no article covers the question. The gap is recorded for the Knowledge Builder.
Unverified → Escalate: a draft was written but did not match the articles it cited, so it was withheld. You see this only where your administrator has turned on answer verification for tickets.

Reply

The AI answers the customer once, by itself. The card reads Automated reply sent → Waiting for the customer’s reply, and the Activity Log says Flexivity AI replied to the customer.

If the customer writes again, the ticket comes to you: Initial reply only → The customer replied — the ticket is yours now. The AI does not continue, and the summary is brought up to date for you.

Clarify

When a request is too vague to answer, the AI asks the customer one question. The card reads Clarifying question sent → Waiting for the customer’s reply, and nothing is needed from you yet.

When the customer answers, the AI either replies, and the card becomes Automated reply sent, or hands the ticket to you. A clear request is answered straight away, as at Reply.

Resolve

The AI keeps the conversation going until the request is resolved, within a limit your administrator sets. The card follows the conversation:

Flexivity AI is handling this conversation → No action needed. The AI is talking to the customer.
Solved automatically, the customer confirmed it or assumed from silence → No action needed. The ticket is set to solved.
Reopened — the customer came back after this was solved → This ticket is yours now.
Handed to you → This ticket is yours now. The exchange so far is in the ticket. If the AI stopped because it could not back up something it said, the card quotes it.

When the AI hands a ticket to you

At Reply, Clarify and Resolve, the AI stops and hands the ticket over when it should not go on. The card reads Handed to you — reason → This ticket is yours now, and the customer has been told a person will follow up. The reason tells you where to start, for example:

the customer asked for a person
the knowledge base does not cover this
it drafted an answer the knowledge base didn’t actually support
the customer was getting frustrated
the conversation reached its message limit
the customer did not answer our question
one of your rules routed it to a person

Context shows the knowledge articles the AI used. If the summary is behind the conversation, click Refresh on it. See AI Context.

When you reply

Replying to the customer yourself takes the conversation over, and the AI stops. It does not resume on that ticket.