Agent assist

How it works

How it works

The assistant gets chat and email text right away; for calls, it works from live speech recognition as the conversation goes. A language model figures out what the customer is asking, finds the answer in the knowledge base, and shows the agent one or two suggestions on screen. The agent decides whether to use a suggestion and how to say it in their own words.

• Canned responses and knowledge base articles for the customer's question
• Reminders of required phrases and script steps
• Agents' best answers become canned responses for the whole team

Metrics

Results to track

• Average handle time (AHT): the conversation and the work after it
• First contact resolution (FCR)
• How many weeks it takes a new agent to work without a mentor
• Share of suggestions agents actually use

Practice

Best practices

• One or two precise suggestions instead of a stream of text: agents have no time to read during a call
• A "helpful" or "not helpful" rating on every suggestion
• An up-to-date knowledge base: a suggestion cannot be more accurate than the article it comes from
• Start with one queue, compare the metrics before and after, then expand

In CONTACT.WORK

How CONTACT.WORK uses it

The assistant opens right in the agent workspace – in a call, chat, or email – and draws on the same knowledge base as the AI agents. Agents' best answers are saved to the knowledge base as canned responses. You can train the model on your own conversations or connect an off-the-shelf model via API. When a conversation ends, AI writes a short summary that goes into reports with no manual entry.

The platform itself is free. The speech recognition and suggestion models are paid separately, or you can run them on your own hardware.