AI in the case console: next best action, ranked replies, field proposals, reply review

Everything on this page obeys one switch: the case AI policy (Settings · Cases · AI policy). When it is off, nothing that costs AI runs. Two things work even with the policy off, because they cost nothing: the next best action and the article hints while you type.

Everything on this page obeys one switch: the case AI policy (Settings · Cases · AI policy). When it is off, nothing that costs AI runs. Two things work even with the policy off, because they cost nothing: the next best action and the article hints while you type.

Next best action The panel on the right of a case shows what the case itself asks for right now: send the first reply, reply now (the customer wrote last), escalate because the resolution deadline passed, review why it was reopened, recover a customer who left a low rating, assign an owner, follow up when the customer has been silent for days, write the resolution after closing, or draft a knowledge article when nothing in the base covers the subject. Each action shows the reasons behind it. It is computed from deadlines, messages, reopenings and ratings; no model is involved. Dismissing an action hides it until the reason disappears and comes back. Urgent actions also notify the case owner in the bell, once per occurrence. The Copilot tool for recommended actions lists these too.

Ranked replies with a reason In the reply editor, Suggest replies asks for up to three candidate replies, each with an approach (direct answer, ask for details, empathetic, escalation, proactive update) and one sentence explaining why it fits this conversation. Pick one to load it into the editor; you still edit and press Send. Nothing is sent by the AI. The approaches your team picks most often are counted and influence the next ranking. Draft a proactive update writes a status update for a customer who has not written recently; you approve it before it goes out. Replies to customers only cite public articles.

Articles while you type As you write, the base is searched with your draft and the matching articles appear with the words that made them match. No AI cost.

Review my draft Before sending, ask for a review: an overall score, five dimensions (empathy, clarity, completeness, grounded in the conversation and articles, compliance) and flags such as a promise without basis or a customer question left unanswered. It only suggests; the Send button does not depend on it. A supervisor can also review a reply that was already sent.

Fields the AI can fill Propose values reads the case and proposes values for empty fields that have a fixed vocabulary: product, the customer's asset, and any dropdown you added to the case. A proposal that does not match an existing option is discarded. Nothing is written until you accept; accepting uses the normal edit path with your permissions.

Summaries Summarize now gives a 20-second summary of the conversation. When a case is handed to another agent, a handoff note (request, what was done, what the customer is waiting for, next step) is posted as an internal note once, and the new owner is notified. When a case is resolved without a resolution text, the AI proposes one; you accept or dismiss it.

Priority from sentiment When a new customer message arrives, the case AI re-reads it. Under the same policy that governs category and priority, a change of priority is either applied (auto mode, above the confidence threshold and within the daily cap) or left as a suggestion.

Self-service bot Answers from the help center bot now mark which article each sentence relies on, with links to those articles. The bot still answers only from your published articles.

What is not here Live voice coaching, real-time translation without a translation provider, and training a custom model are not part of the product.

Open this article inside the system →