How does AI handle first-line support requests?

How AI handles a request

First-line support is the reps who answer customers first. They pass complex questions to the second line: engineers and specialists. An AI agent goes through a request in the same steps as a first-line rep: it reads it, looks for the answer and prepares a reply for the customer. The rep sends the reply.

Triage. The agent reads the request and adds the topic, priority, customer name and a one-line summary. The person on duty sees what came in without opening every email. The agent can pass urgent requests straight to the shift lead.

Answers from the knowledge base. Changing plans, invoices and closing documents, signing in, settings. The agent finds the article in the knowledge base, writes a draft reply and cites the source. The rep checks the reply against the article and sends it. This is how our agent Kai works at Pachca: a reply to a technical question is ready in 2 minutes instead of 30, and a teammate only has to check and send it.

Similar requests. The agent finds an earlier request on the same topic in the support history and attaches the reply the customer got back then.

Missing details. If the email has no contract number, app version or screenshot, the agent prepares a follow-up question. The customer sends the details before the request reaches an engineer.

A weekly digest. On Friday the agent gathers the week's request topics and the questions it found no answer to in the knowledge base, and sends them as a spreadsheet. It shows which articles to write.

Where a person is needed

The agent can make mistakes, and the customer reads the reply as coming from your company. So for the first few weeks the agent only writes drafts, and reps do the sending. Some requests are worth handing straight to people even after that.

Money. Refunds, double charges, discounts and recalculations. The agent can get an amount wrong or promise something the contract doesn't cover.

An outage affecting many customers. If five requests about the same error arrive within ten minutes, it's most likely an outage. The agent calls in the on-call engineer, and first-line support replies to customers with one text agreed on with them. Setting up on-call alerts is covered in a separate use case.

Complaints. The customer threatens to leave, demands compensation or writes for the third time about the same problem. The agent flags these requests and passes them to the shift lead.

Contracts and personal data. Changing billing details, requests to delete data, questions about the contract. Replies to these are agreed with a lawyer.

No answer in the knowledge base. Let the agent say “I don't know”, or it may make up a plausible answer. Such a request goes to a rep, and the question lands in the weekly digest.

What problems this solves for first-line support

The answer exists, but nobody finds it. The knowledge base has the article, but the rep doesn't remember it and asks an engineer. The engineer drops their own task, and the customer waits for a reply.

Urgent requests sit among simple ones. If all emails land in one queue, a double charge ends up between questions about invoices. The person on duty only finds it after reading every email.

The same question gets different answers. If each rep writes the answer in their own words, customers with the same question get different answers, and one of them may be wrong.

Nobody sees what the knowledge base is missing. If a question without a ready answer gets settled in direct messages, the article about it never appears.

What you need in Pachca

First-line support works in a chat where requests arrive. The AI agent joins it as a bot.

A chat for requests. Create a chat, for example “[Support] Requests”, and add your reps and shift leads. Each request arrives as a separate message and is discussed in its thread. A thread is a discussion branch under a specific message. The chat shows the list of requests, and the conversation about each one stays in its thread.

An engineer in the thread. If a request needs to go to development, mention an engineer in the thread. They get access to the thread even if they aren't in the support chat: this is called a cross-chat thread. The engineer sees the customer's email and everything already discussed about it.

Reactions as status. A rep adds 👀 when they pick up a request and ✅ once they've replied to the customer. The person on duty can tell from the reactions which requests nobody has picked up yet.

Customers in a shared chat. You can invite your customer's employees to Pachca as guests. A guest sees only the chat they were invited to, and can't see your employee list or other chats. Guest accounts are available on the Company and Corporation plans.

A requests bot. Emails, website form submissions and messages from a Telegram bot arrive in the chat from a bot. To create one, open the Integrations menu, choose Bots, click Create bot and add the bot to the support chat. If your ticketing system sends outgoing webhooks, you can connect it to the bot through an incoming webhook without a developer. Requests will then arrive as they are, without a topic or priority.

An AI agent in n8n. Triage, answer lookup and the draft are handled by an n8n workflow. n8n is an automation builder where you put a chain of actions together from ready-made blocks without code. The Pachca node in n8n receives new messages and bot button clicks, searches messages and posts messages to the chat and the thread. The AI Agent node reads the request and looks for the answer in the knowledge base. When a rep clicks “Send” under the draft, n8n sends the reply back where the request came from: to email or Telegram. How to build an agent with a knowledge base is covered in “How do you build an AI agent without code?”, and how n8n works in “How do you automate routine work with n8n and AI agents?”. If n8n doesn't fit, a developer can write the handler with the Pachca API.

Where to start

Pick the five most common questions from the past month and check that each has an article in the knowledge base. Connect the agent to the support chat so that it only triages requests and prepares drafts, while reps do the sending. After two weeks, look at which drafts went out without edits and which questions ended up in the digest without an answer.