Summary
Automate the questions a page in your docs already answers.
Approve an answer once when a new question needs a decision, then let it be reused.
Keep refunds, account changes, and complaints with a person.
Count what was answered, and track moved work separately.
Read what your automation missed every week, and fix the docs.
Customer service automation is software doing support work without a person, such as routing tickets or answering questions from your own help content.
Only the answering closes a question. Routing, tagging, and auto-replies move it to someone who still has to reply.
Answering is also where a wrong answer reaches the customer. For a team with real documentation, that answering part can be large.
Jortt, a Dutch accounting platform, resolves about 92% of roughly 30,000 monthly inquiries without a person. By Wonderchat’s arithmetic on those figures, the other 8% is still about 2,400 inquiries a month the agent doesn’t resolve alone.
Start with what your docs already answer, and judge an AI customer service agent by what it resolved.
Does your automation answer the question or move it?
Some automation moves the question, and only answering finishes it. IBM’s definition covers chatbots, ticket routing and tagging, and email auto-responses, and Salesforce’s guide names autoresponders too, so both count the moving and the answering as automation.
Moves the work | Answers the question |
|---|---|
Routing and triage to the right person | An AI agent answering from your help center, with its source |
Tagging tickets by topic and urgency | Help center articles customers find on their own |
Auto-replies that confirm a ticket arrived | Order status lookups from your store system |
Canned replies that a person still picks and sends | Answers a person approved once, reused for the same question |
Take a customer who asks how to export invoices. A routing rule tags it Billing and sends it to a person who still types the reply, while an agent answering from your help center article closes it and shows the page it used. Both columns help. Only the right-hand column lowers the number of questions a person has to answer.
Customer support automation, or automated customer support, is the same idea under another name. Whichever term you use, ask one question of every tool.
Does this answer the customer, or does it pass the customer along?
The next decision is which of your customers’ questions belong in the right column at all.
What should you automate, approve once, or keep human?
Automate what a current page in your docs already answers, approve new answers once, and keep every decision about one customer with a person.
Three questions sort any kind of ticket, in this order.
Only questions a current page already answers go straight to automation, and every decision, complaint, or new question reaches a person first.
Automate the questions your docs already answer
Questions a current page answers are the safe ones to hand over, because the answer already exists and someone already checked it.
How-to and setup questions (“How do I export my data?“)
Plan and feature questions that your pricing or feature pages answer
Published policies, such as shipping times and return windows
Order status, when the answer is a lookup in your store system
One condition applies to all of them. The page has to be current, or the automation repeats an old answer with confidence.
Approve an answer once when it needs a decision
Some questions are new or sit on an edge case, and nobody has written the answer yet.
In that case a person writes or approves the answer once, and the automation reuses it for the same question from then on.
One approval keeps a human in the loop only where judgment is needed, which is the first time. The approved answer also turns every new question into something your docs or your agent can answer next week.
Keep refunds, account changes, and complaints human
Some questions need a decision about this customer, every single time.
Refunds and billing exceptions
Account changes and cancellations
Complaints and upset customers
Anything legal, security, or privacy related
Some tools can take actions in your systems, such as issuing the refund.
For a team on its first AI customer service software, the safer job for automation here is the handover, because a wrong refund costs more than a slow one. Collect the details, pass the whole conversation to a person, and let that person decide.
What are some customer service automation examples?
Answering examples include an AI agent replying to a repeat how-to question and an order status lookup. Routing, tagging, auto-replies, and canned replies move the work to a person.
Eight examples, and what stayed with a person
Example (moved or answered) | What gets automated | What stays with a person |
|---|---|---|
Repeat how-to question in chat (answered) | The AI agent answers from the help center article and shows the source | Questions no article covers |
Order status (answered) | A lookup in the store system | Delivery disputes |
After-hours chat (answered or moved) | The agent answers, or collects an email for follow-up | The follow-up reply |
New ticket (moved) | An auto-reply with the expected reply time | The reply itself |
Routing and tagging (moved) | Tickets sorted by topic and urgency | Every answer |
Canned replies (moved) | The wording | Choosing and sending it |
Outage (moved) | A status update to affected customers | The fix and the apology |
Refund request (moved) | Collecting the order details and handing over | The decision |
Try the same sort on your own last month of tickets. Mark each kind of question as moved or answered, then look at where the replies piled up.
If more than half of them sit in the answered rows, set up an agent that works from your docs first, because those are replies nobody on your team needs to type again.
A named example from Jortt
Jortt, the Dutch accounting platform, is a Wonderchat customer.
The 92% is one team’s figure on its own content, and it says nothing about what happened to the remaining conversations.
The figure shows how large the answered column can get. Your number depends on how much of what customers ask your docs already cover.
How do you automate customer service?
List last month’s questions, fix the docs, connect an agent that answers only from them, set the handover, and test it with a question your docs don’t cover.
You need a help center and no developer for any of the five.
1. List last month’s questions
Go through a month of tickets, chats, and support emails. For each kind of question, write down whether a page in your docs answers it.
The list becomes your automate, approve once, and keep human columns, the first draft of your customer service workflow. The same list shows you which pages are missing.
2. Fix the docs before you automate
An agent that answers from your docs repeats what they say, including the parts that are out of date. Update or archive old pages first, especially policies and pricing.
Plan the time honestly. In Wonderchat’s product data, knowledge bases over 500 pages take a median 15 days to go live.
3. Connect an agent that answers only from your docs
Choose an agent that answers only from the content you connect and shows the source on every reply. A visible source lets the customer check the answer and lets you see which page it came from.
Start with a hosted link to test, then add the chat to your site with one snippet. Of the Wonderchat agents that go live, 80% do so the same day.
Test first on Wonderchat’s free plan, which refills 50 AI replies every month, enough to ask your top 10 questions five times each and read every answer.
Move to Starter, at $39 a month billed monthly with 500 AI replies, once the answers hold up and real customers start asking. The pricing page estimates about 160 resolutions a month on Starter, so compare that with the answered rows on last month’s list.
4. Set the handover before you go live
Decide when a conversation goes to a person, before any customer sees the agent.
When the customer asks for a person
After a set number of messages
After repeated answers that didn’t help
The ticket should arrive with the full conversation attached, so nobody asks the customer to repeat themselves. When nobody is online, collect an email and reply later.
5. Test it with a question your docs don’t cover
Ask something a customer could plausibly ask that no page answers. The right result is “I can’t find that” and an offer of a person.
Asked a question its docs don’t cover, the agent says so and offers a person.
Then make it a weekly habit. Read the questions it couldn’t answer, write the missing answer once, and the list gets shorter.
How should you measure customer service automation?
Count the questions resolved without a person, and write down how you decided each one was resolved. Every other number on your report either feeds that count or hides it.
Metric | What it counts | What it hides |
|---|---|---|
Conversations kept from a person | Conversations that never reached your team | Customers who gave up or were sent elsewhere |
Resolved without a person | Questions answered with no handover | How “resolved” was judged |
Re-contact rate | Customers who came back about the same issue | Little on its own, so pair it with resolution |
Handover quality | Tickets that arrive with the full conversation | Needs a person to read a sample each week |
Watch the first row, because it’s the one that looks best on a dashboard. A conversation that never reaches your team counts as kept, even when the customer gave up and closed the tab.
Fin, formerly Intercom, which Salesforce acquired in September 2026, makes the same point in its own guide. A platform can keep 90% of conversations away from a person and still resolve only 40%.
By Wonderchat’s arithmetic on Fin’s example, that gap is 50 of every 100 conversations kept from a person without being resolved, and nobody on your team hears about them.
Self-service hides the same gap. A Gartner survey of 5,728 customers, run in December 2023, found that 73% use self-service at some point, but only 14% of customer service and support issues are fully resolved there.
So a busy help center tells you how many people looked, and very little about how many issues it closed.
Most customers try self-service, but few issues end there, so count what was resolved.
That’s why the count to keep is how many questions are resolved without a person, with a note on how you counted them. Wonderchat’s study of 5,848,078 conversations shows its AI agent alone resolved 75.3% of scored conversations in 2026.
Where does customer service automation go wrong?
Plan for an old page that answers, a setup that turns into a project, and stale answers with no way out.
An old page answers: Wonderchat’s support team sees one failure more than any other, and it is retrieval pulling an archived policy and answering from it correctly.
The fix is simple. Archive or delete the old page on the day the policy changes.
Setup turns into a project: When Wonderchat read 837 reviews of 18 vendors in September 2026, setup was the second-biggest complaint theme (197 of 1,309 complaints), after pricing (211).
Start with the repeat questions only, and add the rest once those answer well.
Stale answers have no way out: Stale knowledge came up 38 times in the same reviews. When there is also no handover, a wrong answer becomes the last word.
A clear handover and a weekly read of what the agent missed fix all three.
A question no page covers goes to a person and to the review inbox, and the answer written once is there for the next customer.
When is Zendesk or Help Scout the better choice?
Zendesk is the better choice for a team already on Zendesk that wants automation to take actions. Zendesk’s documentation says its AI agents can perform actions in authorized systems and work across messaging, email, and voice.
Wonderchat answers and hands over, and doesn’t process refunds or change accounts.
Help Scout is the simpler buy for an email-first team that wants a shared inbox and a docs site with AI in one product. Help Scout sells its AI agent as the Envoy add-on, at $0.75 per resolution on top of per-user plans, with a 3-month free trial, according to Help Scout’s pricing page on 30 October 2026.
A per-resolution price is cheap at low volume, and the bill rises with every month the agent does well.
By Wonderchat’s calculation from both pricing pages, 160 resolutions a month cost $120 on Help Scout before any user seats. Wonderchat’s Starter costs $39 a month, billed monthly, for about 160 estimated resolutions.
If your help desk already works and the same documented questions fill it every day, Wonderchat answers those with their source. Every refund or account change still goes to a person.
How does Wonderchat automate the questions your docs already answer?
Wonderchat answers from your help center with its source, hands the rest to a person, and sends each miss to the review inbox.
Wonderchat is an AI customer service agent and AI helpdesk in one, for teams adopting their first AI customer service software, whatever their size. A customer asks how to export invoices, and the agent answers from your help center article with its source.
The next customer asks about a refund exception that no page covers. The agent says it can’t find that and offers a person.
The ticket arrives with the full chat, in Wonderchat’s help desk or in Zendesk or Freshdesk, synced from Basic and capped at 50 tickets until Scale.
Start for free with a free plan that refills 50 AI replies every month, or book a demo to see it on your help center.
Frequently asked questions
What are some examples of automated customer service?
Routing and tagging tickets, auto-replies that confirm a ticket arrived, order status lookups, after-hours chat that collects an email, and an AI agent answering how-to questions from your help center. Lookups and AI answers close the question. The rest move it to a person.
Can you automate customer service?
Yes, for the questions your docs already answer. Keep refunds, account changes, and complaints with a person, and make sure every conversation can reach one.
How do I automate customer support with AI?
Fix your docs first. Then connect an AI agent that answers only from them and shows its source, set when it hands over to a person, and test it with a question your docs don’t cover.
What should not be automated in customer service?
Refunds and billing exceptions, account changes, complaints, and anything legal or security related. Automate the handover for these instead, with the full conversation attached for the person who takes it.
Does automating customer service make the answers worse?
It can, when the agent answers from outdated pages or guesses. It shouldn’t, when it answers only from current docs, shows its source, and says when it can’t find something.
