Answer routine questions instantly, day or night

The problem

Routine questions take your team's time in small pieces, all day. Every business has its list: opening hours, prices, what's included, delivery times, how to reschedule, whether there's parking. The answers rarely change. But each question still arrives as a call, an email, a chat or a DM that someone has to stop and answer.

The interruptions add up. Your team switches between the customer in front of them and the question on the screen. Replies get shorter and less consistent: one person says delivery takes three days, another says five. After hours, the questions wait until morning, and a customer who wanted a quick answer before buying has bought somewhere else.

Older chat tools didn't fix this. Decision-tree bots make people click through menus that never quite match their question. A generic AI chatbot can answer confidently and still get your prices wrong, a problem known as AI hallucination. Either way, customers learn not to trust the chat.

What we build

We build an FAQ chat agent on Concierge, our AI sales and service agents. It answers only from what is true about your business.

1. We build the knowledge base with you. Your services, prices, policies, hours, service areas, delivery and returns terms, and the questions your team actually hears, gathered from your website, documents, inbox and call notes. You approve it.
2. The agent answers from that source only. It uses retrieval-augmented generation: it looks up the relevant approved content first, then writes the answer from it. It doesn't answer from the open internet.
3. It says when it doesn't know. If the answer isn't in the knowledge base, the agent says so and offers a person, rather than guessing.
4. It works where your customers ask. Website chat, WhatsApp, SMS, social DMs and email triage all draw on the same knowledge base, so the answer is the same on every channel. The agent says it is an AI at the start of each conversation.
5. It hands off with context. Complaints, complex cases and anything your rules flag go to a person, with the full conversation and a summary attached. During staffed hours that can be a live handoff; after hours, it's a ticket or a callback task.
6. Questions that are really leads become leads. When a question signals buying intent, such as availability, pricing for a specific job or a booking request, the agent captures the details and creates a lead in Cadenza, our lead follow-up system.
7. We review the logs. Unanswered questions and flagged answers show what's missing from the knowledge base. Updating it means editing the source, not retraining anything.

How it connects

• Cadenza. Buying questions become leads with the conversation attached, and follow-up starts if they don't book. Instant lead response covers what happens next.
• Maestro (workflow automation). Some questions need a lookup, not an FAQ: order status, appointment times, account balances. Maestro connects the agent to the systems that hold those answers, with read-only access where that's all it needs.
• Questions that are really bookings go straight to appointment booking. For more on setup, scripts and handoffs, read website chatbots that actually capture leads.

What we measure

Before launch, we sample your inbox and chat history to see which questions arrive most often and how long they wait for an answer. After launch, we report against the agreed targets.

• Questions resolved without a person. The share of chat conversations that end with the customer's question answered and no handoff requested. Measured from conversation logs, and checked by reviewing a sample of transcripts with you.
• Time to first response in chat. Seconds from a customer's first message to a meaningful reply, by channel and hour of day. Measured from conversation timestamps, median and 90th percentile.
• Handoffs resolved without the customer repeating themselves. The share of handed-off conversations your team resolves without asking for information the customer already gave. Measured from a quick tag your team adds when closing each one, and by sampling. See the glossary entry on human handoff.
• Answers flagged as wrong or incomplete. Answers flagged by customers, by your team or in our transcript reviews. Each one is logged with its fix, so the knowledge base gets better over time.