Use case · Repeat customer questions
Customer service automation: your documents answer for you.
Customer service automation starts with the questions that keep coming back: lead times, delivery, warranties, service areas. The tool finds the answer in your terms of sale and product sheets, cites its source, and hands your team whatever your documents don’t cover.
- Every answer cites its document
- Email, website chat, and help desk
- Anything missing goes to your team
30-minute callNo commitmentReply within 24 hours
Question from Ms. LaurentSimulated
The problem, up close
Wednesday, 9:05 a.m. Thirteenth time this month.
Alice opens the customer service inbox. Ms. Laurent writes: “Do you deliver to Belgium? And what would 75 chairs cost?” It’s the thirteenth time this month someone has asked whether the company delivers to Belgium. The answer is already written down, in the terms of sale and the shipping rate sheet.
Alice reopens both documents, checks the lead time, retypes the sentence she’s already written twelve times, then forwards the pricing question to the sales rep. Meanwhile, the website chat has received two questions about the warranty, and the help desk shows six pending requests. Each answer is simple. It’s the sheer number of them that fills the day.
And on the day Alice is out, someone else digs for the answer, and sometimes replies with the old shipping rate sheet.
Wednesday · customer service
- 9:05 AMMs. Laurent: delivery to Belgium, 75 chairs
- 9:09 AMTerms of sale and rate sheet reopened
- 9:13 AMPricing sent to sales, reply out
- 5:30 PM6 support requests still open
Illustrative scene: fictional company and people.
What the tool does, what stays with your team
The tool looks it up and answers. Alice handles the special cases.
What it can say on its own, and what it must pass on, is decided with your team, question by question.
The tool
Catches questions where they land
Email, website chat, or help desk: the question is read in one place, with the customer’s name and message history.
Searches before writing
It first finds the relevant passages in your terms of sale, shipping rate sheet, and FAQ, and writes only from them.
Writes and cites its source
“Yes, we deliver to Belgium, within 5 to 7 business days.” Below the answer, the documents it came from, so you can check at a glance.
Passes on what’s missing
The price of 75 chairs isn’t in any document: the tool doesn’t make it up. It hands that part to the sales rep, along with the request.
Your team
Reviews and sends
Alice reviews the prepared answer, checks the cited source, and sends it. What used to take her eight minutes now takes a few seconds.
Keeps documents up to date
A new delivery area, a lead time that changes: the team updates the document, and the answers follow, with nothing else to set up.
Prices and negotiates
Prices, discounts, and special cases stay with the sales rep, who gets the customer’s request already summed up.
Two modes, set task by task
At first, every answer waits as a draft for Alice to review and send, with its sources right in front of her.
Once it’s proven, the tool sends answers fully covered by your documents on its own, like delivery to Belgium. As soon as something isn’t in them, the question goes back to your team.
Plugged into your software
In your inbox, your website, and your shared folders.
Customers write where they always do. Replies go out from the same place, in your company’s name.
- Outlook
- Gmail
- WordPress
- Shopify
- PrestaShop
- Google Drive
- SharePoint
- Notion
Don’t see your help desk (Zendesk, Freshdesk, Crisp…) or business software here? If it can share its data, I check how to connect it on the first call.
Response time
8 min15 sec
per reply
Time to review a prepared, sourced reply versus searching the documents and writing it by hand, in the demo example.
Simulated exampleQuestions
Customer service automation. Straight answers.
01Can you answer customer questions automatically without annoying them?
Yes, if the answer is correct, specific, and says where it comes from. What annoys people is a vague or off-target answer. The tool only answers when your documents cover the question, and the customer can always ask to talk to a person.
02Which questions can customer service automation handle, and which should your team keep?
Hand it the questions whose answer is written down somewhere: lead times, delivery, warranties, returns, business hours, product specs. Keep negotiated prices, complaints, and unhappy customers for your team. The split is set question by question, and it can change over time.
03Where does the tool find its answers?
In your documents, and only there: terms of sale, shipping rate sheet, FAQ, product sheets. Before writing, it looks for the passages that answer the question, writes from them, and cites its source. That’s called RAG: an AI that checks your documents instead of answering from memory. The service is detailed on the AI chatbot for your documents page.
04What if the tool doesn’t know the answer?
It says so and hands off. In the example, the price of 75 chairs isn’t in any document: the question goes to the sales rep with the customer’s message, instead of a made-up answer. A question that keeps coming back unanswered also tells you a document needs filling in.
05Does customer service automation work across email, website chat, and help desk?
Yes: the same document base serves all three channels, so the same question gets the same answer everywhere. On the website, the tool introduces itself as an AI, as the EU AI Act requires. The article on customer service chatbots and the EU AI Act details these rules, and the website chatbot page shows the service.
The solution
The service that fixes this
- AI chatbot for your documentsAnswers from your specs and PDFs
- Website chatbotAnswers visitors, hands off when needed
Read next
Contact
Which question does your support team get ten times a day?
Write it in one sentence. I’ll show you the answer the tool would give, source included.
On the call, we look at your workflow together and I show you the closest tool, on sample data. If a simple setting in your software does the job, I’ll tell you.
More use cases