PPactDocs
Support

Agent assist

AI-assisted support tooling — tone-matched draft replies, empathy rewrites, KB deflection recommendations, macros, and an escalation-risk badge — all reviewed and sent by a human agent.

Agent assist speeds up the work of resolving a case without taking the agent out of the loop. Every feature here produces a suggestion, a draft, or a signal — the agent decides what to send.

These are suggestions, not auto-solve

Agent assist never resolves a ticket or sends a reply on its own. It drafts and recommends; the agent reviews, edits, and sends. There is no autonomous ticket resolution in Pact.

Draft reply suggestions

On a case with enough context, Pact generates three tone-matched draft replies in a single pass, so the agent can pick a starting point that fits the situation. The suggestions are:

  • Grounded — they won't fire on a thin case with too little to go on.
  • Filtered — output is checked before it's shown.
  • Fallback-safe — with no BYOK AI provider, Pact falls back to deterministic templates rather than nothing.

Empathy rewrite

Take a draft an agent wrote and rewrite it in a warmer, more empathetic tone — useful when a customer is upset and the wording needs to land gently.

KB deflection recommendation

Pact checks the case against your knowledge base and, when a strong article match exists, recommends "send this article?" This is a deterministic, KB-first check — no AI required — so an agent can deflect a common question to self-serve content in one click.

Macros

Macros are agent-run canned replies for the responses your team sends over and over. The agent triggers a macro; it doesn't fire automatically.

Escalation-risk badge

Each case carries a green / amber / red escalation-risk badge from Pact's escalation predictor — a rules layer with an AI fallback for ambiguous cases. It's a heads-up on which cases are most likely to blow up, so an agent can get ahead of them.

Works with or without AI

Draft replies and the escalation predictor use your BYOK AI provider when it's configured and fall back to deterministic behavior when it isn't. Deflection recommendations and macros need no AI at all.

What's next