Customers get a straight answer or a finished task, like an exchange with its return label. Behind each reply, the AI customer service agent reads the thread so far and checks your policies, catalog or systems first.
A reply like “yes,” “the bigger one” or “bought it last week, unopened” is read against what was just said. When one message asks two things, say the return window and a delivery time, the agent looks each one up separately and answers both, or says plainly which part it has no answer for.
Questions like “how many styles do you carry?”, “list them all” or “what colors do they come in?” go to Key facts, the structured records Risolva pulls from your catalog and documents. The agent counts, pages through and groups those records by attributes like color or size instead of estimating from a few search results. Stock answers are as current as your last catalog sync.
When your content has no answer, the agent says so instead of filling the gap. Requests unrelated to your business, like homework or trivia, get a polite no, and anything on your Prohibited topics list stays off the table even when asked directly. An unfamiliar product code still gets searched, because it might be one of yours.
The agent can run several lookups or actions before it writes back. Here is a swap request, from the customer's first message to the return label landing in the chat.
The agent sorts the message as a question, a request to do something, a greeting, small talk or something off-topic, notes the language, and reads it with the turns before it. Long conversations keep a running summary, so a detail from earlier still counts.
Policy questions go to a knowledge search, “how many” and “list all” questions go to Key facts, and anything that lives in your own system, such as an order, goes through a custom action. It can combine these in a single turn.
For a swap, the agent gathers what create_exchange needs, the order number and the new size, and asks “Shall I go ahead?” With Requires approval on, the action runs only after a yes, using exactly the details the customer confirmed. A file your system returns, like the return label PDF, goes back in the chat.
Each turn ends with one outcome: an answer built from what it found, a clarifying question, a plain “I don't have that information,” a polite decline, or a handoff to your team with the reason logged on the conversation.
create_exchangeMost of these live on the bot's Persona, Answers and Actions pages in the app. Test and Publish sit at the top of the bot.
An AI agent for customer service is software that reads a customer's message, works out what it needs, and then answers, carries out a task, or passes the chat to a person. Risolva's agent searches your content, counts and lists catalog items, and calls your own systems through custom actions, on WhatsApp, Instagram, Messenger and web chat.
Book a demo with messages that only make sense in context, like a bare "yes", "the bigger one" or two questions in one line. We'll show what the agent looks up for each, where it would run an action, and where it would hand the chat to a person.
