Breadcrumbs

Real-time AI Suggestions

Agents see suggestions in a Co‑Pilot panel next to the conversation and can insert or adapt them as needed.

Smart agent assistance during conversations

EF Agent Co‑Pilot provides real‑time assistance for agents on both chat and voice interactions:

  • For chat, it reacts to each new customer message.

  • For voice calls, it reacts to the live or recent transcript of what was said on the call.

NLU‑based understanding & suggestions

What it is

The system uses NLU (Natural Language Understanding) to analyze what the customer is saying in:

  • Chat messages, and

  • Voice calls (via call transcription).

It detects:

  • The topic (for example: “password reset”, “invoice question”, “delivery delay”).

  • The intent (what the customer wants to achieve).

Based on this, Co‑Pilot shows:

  • Suggested short replies the agent can send.

  • Suggested articles or snippets from the knowledge base.

  • Suggested next questions to clarify the issue.

What the agent sees

In the Co‑Pilot panel, which updates as the conversation continues (chat or voice transcript):

  • 2–3 possible replies.

  • 2–3 relevant KB articles with a short description.

The agent can:

  • Click a reply suggestion to insert it into the reply box and edit before sending.

  • Click a KB suggestion to open the article or snippet for more details.

This works the same way whether the conversation started as chat or a voice call (through its transcript).

LLM‑based drafted replies

What it is

When agents need a complete, well‑worded answer, they can ask Co‑Pilot to draft a full reply.

The AI uses:

  • The conversation history (chat or voice transcript).

  • Relevant knowledge base content retrieved from the KB Server via Agentic RAG.

  • The organization’s preferred tone and language (for example: professional, friendly).

How it works for the agent

  1. The agent clicks a button such as “Generate with Co‑Pilot” near the reply box.

  2. Co‑Pilot:

    • Reads the most recent customer and agent messages (or the latest part of the call transcript).

    • Fetches relevant information from the KB Server.

    • Proposes 1–3 drafted replies.

  3. The agent can:

    • Use a reply as‑is.

    • Edit it before sending.

    • Discard it and write their own.

This helps agents respond faster, more consistently, and in line with your policies and knowledge.

AI that uses your own knowledge (Agentic RAG)

EF Agent Co‑Pilot uses an Agentic Retrieval‑Augmented Generation (RAG) approach.

In simple terms: the AI does not answer from generic memory alone. It actively searches your own knowledge base first, then composes a response based on what it finds.

What this means for you

When a customer asks something (in chat or via a voice call that is transcribed):

  1. Co‑Pilot reads the latest part of the conversation.

  2. It searches the KB Server for the most relevant:

    • Articles

    • PDF sections

    • Web pages (scraped content)

  3. The AI can make several search calls, refining what it needs.

  4. It then uses those results to:

    • Build suggested answers.

    • Provide step‑by‑step procedures.

    • Show supporting KB links.

What the agent sees

  • Suggested replies that include a “Source: …” reference with:

    • Article name

    • Short snippet

    • A clickable link

  • A small list of “Related Knowledge” on the side, which can be opened in a new tab or side panel.

This ensures your answers are:

  • Consistent – all agents rely on the same sources.

  • Up‑to‑date – responses are based on your current KB content.

  • Traceable – agents can always see where an answer came from.