Use case: ticket deflection
Answer the repeat questions before they become tickets
Train an AI agent on your help content and the tickets your team already solved, put it where customers ask, and keep a clean path to a person for everything it can’t answer. The tickets that still reach your helpdesk arrive with a summary and the full conversation.
Visitor: Where do I download last month’s invoice?
Agent: Go to Settings, then Billing, and open Invoices. Every invoice has a Download PDF button. Invoices are created on the first day of each month.
Visitor: The invoice shows the wrong company name. Can you change it?
Agent: I can’t change an invoice myself, but I can pass this to our billing team by email. Which address should they reply to?
Visitor: finance@example.com
Agent: Done. The team has the conversation and will reply to finance@example.com by email.
Your helpdesk is full of questions you have answered before
Open your helpdesk on any weekday and many of the new tickets ask something your team has answered many times: how to reset a password, when an order ships, how to change a plan, where to find an invoice. Each one still needs someone to read it, look up the answer and write it out, and while they do, the tickets that need real judgment wait in the same queue. Ticket deflection means answering those repeat questions where the customer asks them, so they never become tickets. It only works if the answers are right and if customers who need a person can still reach one easily. A chatbot that blocks the way to your team doesn’t reduce work; it moves the frustration into your inbox a few minutes later.
What changes for your support team
The agent handles the questions that already have an answer. Your team gets the ones that need a person, with context.
Fewer repeat questions in the queue
Questions answered by your help pages, policies or past tickets are answered in the chat or by email, in the customer’s language, at any hour.
Answers your team already wrote
Solved Zendesk and Freshdesk tickets become questions and answers for the agent, with personal data replaced before anything is saved.
A person when one is needed
Customers who ask for someone get live chat if your team is online, or a hand-off by email and a ticket in your helpdesk if not.
Tickets that arrive with context
A handed-off ticket carries the agent’s summary, three to five AI-written points and the last 30 messages, so nobody asks the customer to start again.
A list of what is still missing
Questions the agent struggled with are collected every day, so you know which answer to write next instead of guessing.
The deflection workflow in intoCHAT, step by step
Find the questions that repeat
Look at your recent solved tickets and note the questions that come back every week. Once the agent is live, Topic analysis can label its conversations; Suggest topics reads up to 50 recent conversations and proposes a topic list you can edit.
Train the agent on the answers
Crawl your help center and website, upload policies and manuals, connect Notion or Google Drive, and write Q&A pairs for answers that must be exact. From Starter, import solved Zendesk or Freshdesk tickets; Intercom Help Center articles can be imported too.
Put it where tickets start
Add the chat widget to your site and help center. Publish the Help Page, a hosted page with the chat and up to six links, and link to it from your contact page. From Starter, forward your support address to the agent’s email address so it answers email too.
Let it resolve, not just reply
Some questions need data: connect Shopify or WooCommerce for verified order lookups, add API actions that read from your own systems, and write procedures (from Starter) for step-by-step requests such as a refund that runs only after the customer confirms.
Escalate cleanly
Switch on hand-off by email and, if your team can watch the chat, live chat. Connect Zendesk, Freshdesk, HubSpot, Gorgias or Help Scout so every hand-off by email also opens a ticket there, and handle the rest in the Inbox.
Close the gaps every week
Work through Questions to improve: fix the source, or click Improve answer under a weak reply to turn it into a Q&A pair. On Pro, the Conflicts card points out sources that contradict each other, such as an old price next to a new one.
Setup checklist and limits
Knowledge sources
Website crawl, files, text, Q&A pairs, Notion, Google Drive and Intercom articles. Everything counts toward the agent’s character limit: 100,000 on Free, 500,000 on Starter and 1,000,000 on Pro.
Ticket import (Starter and up)
Solved or closed Zendesk and Freshdesk tickets only: 200 per agent on Starter, 1,000 on Pro, 5,000 on Enterprise. Internal notes are skipped, and emails, phone numbers, order numbers and names after a greeting are replaced. Names elsewhere in free text aren’t detected, so check the Preview.
Helpdesk tickets on hand-off
One helpdesk per agent and one ticket per conversation. A ticket is opened only when the hand-off email went out; a live chat your team takes over doesn’t open one.
No sync back from the helpdesk
Replies, status changes and notes in Zendesk or Freshdesk don’t come back to intoCHAT. Your team answers the customer from the helpdesk.
Hand-off limits
A conversation can be handed off once every 10 minutes, and an agent sends up to 20 hand-offs an hour. With only live chat on and nobody online, the agent doesn’t offer a person, so keep hand-off by email on as a fallback.
What you can measure
Outcomes on the dashboard shows conversations, hand-offs, the share of rated replies marked helpful and the share of low-confidence replies, each against the previous period. The weekly insights email compares hand-offs and not-helpful ratings week on week.
Questions to improve
Once a day, around 02:30 UTC, replies with confidence below 0.5, with no matching knowledge or rated not helpful are listed on the Knowledge tab, with repeats of the same question counted together.
On every plan, with seats from Starter
Hand-off by email, live chat, helpdesk tickets and the Inbox work on every plan, Free too. Free has the owner only; Starter includes 2 team members and Pro 5.
Frequently asked questions
What is ticket deflection?
Ticket deflection means answering a customer’s question before it becomes a support ticket, usually with self-service content or an AI agent. In intoCHAT the agent answers from your help content and past tickets in the chat or by email, and passes the customer to your team when a person is needed.
Does intoCHAT show a deflection rate?
No. intoCHAT doesn’t calculate a deflection rate, because a conversation that ended without a hand-off isn’t proof the customer’s problem was solved. What you get are counts you can check: conversations and hand-offs on the dashboard, helpful and low-confidence shares, the weekly insights email and the transcripts themselves. Reading transcripts remains the way to judge answer quality.
Can it learn from our Zendesk tickets?
Yes, from Starter. Connect Zendesk or Freshdesk and import tickets solved in the last 30, 90 or 365 days. Each becomes a question and the public answer your team gave; internal notes are never imported. Tickets rank below your website when the two disagree, because they were answers to one customer. HubSpot, Gorgias and Help Scout tickets can’t be imported.
Will customers still be able to reach a person?
Yes, if you switch it on. With hand-off by email on, the agent offers to pass the conversation to your team when a customer asks for a person, or when it can’t help and the customer wants a follow-up. It asks for the address to reply to, sends nothing until the customer agrees and never promises a response time.
What does the ticket in my helpdesk contain?
A subject with the agent’s name and the customer’s first message, the customer as the requester, the agent’s summary, three to five AI-written points about what they want and what was already answered, the last 30 messages and a link to the conversation in intoCHAT. In Zendesk the content is added as an internal note and the ticket is tagged intochat.
What if the agent gives a wrong answer?
Every reply shows Why this answer in the transcript: the knowledge it used and a confidence level. Confidence measures how closely your knowledge matched the question, not whether the reply is correct, so an outdated page can still produce a wrong answer with high confidence. Fix the source, add a Q&A pair, or use Improve answer, and check Questions to improve each week.
Does it work for email tickets too?
Yes, from Starter. Forward your support address from Gmail, Google Workspace, Outlook or Microsoft 365 to the agent’s address. It answers each email in its thread with the same knowledge, and hands threads to your team when asked. Automatic replies, mailing lists and emails that fail DMARC are never answered.
What does ticket deflection cost with intoCHAT?
You can start on Free with 25 messages a month. Starter costs $19 a month for 2,500 messages and adds ticket import, the email agent and procedures. Pro costs $49 a month for 10,000 messages and adds the Conflicts check and Ask your conversations. A reply with the default model counts as one message; Claude Sonnet counts as two and Claude Opus as five.
Start with your most repeated question
Create your agent free, train it on your help center, ask it the questions your team answers every day, and switch on hand-off before it goes live.
Create your agent freeFree plan, no credit card needed.