AI agent for customer service
An agent that answers customer questions on chat, email and messaging apps from your real orders, policies and product data, does routine jobs such as returns and redeliveries, and hands the rest to your team with a summary.
- Answered 23 questions about delivery times and stockDone
- Rescheduled 2 appointments at the customers' requestDone
- A customer wants a refund outside the policy windowAsks a person
- Daily summary sent: top 5 questions this morningDone
An example. Your agent is built around your own work.
How it works
-
Read the question
A message arrives from website chat, the support mailbox or a messaging app. The agent works out what the customer wants, such as order status, a return, a delivery change or a product question, and answers in the language they wrote in.
-
Look it up
It finds the customer by email or order number, then reads their orders, tickets and your approved help articles and policies. It answers from those and names the article or record it used.
-
Act within limits
It can check an order, send a return label, rebook a delivery and open a ticket. Refunds and anything involving money go to a person, unless you allow routine cases up to an amount you set.
-
Hand over
Complaints, legal threats, safety issues and anything the agent is unsure about go to a person on your team with a summary of the customer, the order and what was already tried, so nobody asks the customer to repeat themselves.
New to AI agents? How they work, in plain words
Real examples, setup steps and tool connections for this agent
An example conversation
An example of how a conversation runs. The grey lines show what the agent does in your systems meanwhile.
- CustomerHi, my order 20931 was meant to arrive on Tuesday and it's Thursday. Where is it?
- Found order 20931 in the store system. Read the courier's tracking: parcel held at the local depot since Tuesday, address not found.
- AgentYour parcel is at the courier's local depot because they could not find the address. I can ask for redelivery tomorrow. Is the address on the order still right: 14 Palm Street, Unit 3?
- CustomerUnit 3 is wrong, it should be Unit 8.
- Corrected the delivery address on the order and sent a redelivery request through the courier's portal. Logged both in the help desk ticket.
- AgentDone. I have asked for redelivery tomorrow to Unit 8, and the courier will text you before they arrive.
- CustomerThanks. Also, the kettle I bought five weeks ago has stopped working and I want a refund.
- Read the returns policy: refunds within 30 days, repair or replacement under the 12-month warranty. This purchase is 35 days old. Refund decisions outside the window are marked for a person.
- AgentI'm sorry about the kettle. It is outside the 30-day refund window but covered by the warranty, and a refund or replacement is a decision for my colleague Mariam. I have sent her your order details and the problem you described, and she will reply today.
- Created a ticket in the support queue for Mariam with order, purchase date, the policy lines read and a summary of the chat.
How it is built
- Every channel lands in one place
- Website chat, the shared support mailbox and WhatsApp feed into the same queue. Mail is read through the Gmail API or Microsoft Graph, WhatsApp through Meta's official Business Platform, and if you already use a help desk such as Zendesk or Freshdesk, tickets are read and written through its API and webhooks. Each message is saved with the customer's record before anything else happens.
- A language model reads, your rules decide
- A large language model, the same kind of AI behind ChatGPT and Claude, understands what the customer is asking. Your rules decide what happens next: which questions it answers, which it must pass on, and how sure it has to be. Refunds, legal complaints and safety issues go straight to a person's queue.
- Answers come from approved content and live data
- The agent answers only from the help articles, return policy and product data you approve, and from your order system at the moment of the question, not from what the model remembers. It says which article it used. If nothing in your content covers the question, it says so and hands over instead of inventing a policy.
- A fixed list of actions
- The agent gets a short list of things it may do, such as look up an order, send a return label, rebook a delivery or open a ticket, each with limits you set. Anything outside the list, it cannot do.
- Your team keeps the controls
- Staff see every conversation in the help desk or a shared inbox, can step into any chat and hand it back. Every message and action is logged, so you can read what was said and done.
- Review that catches the quiet failures
- Each week we look at how many questions the agent resolved, how many tickets were reopened and how many customers came back within 7 days, plus a random sample of replies read by a person. Reopened tickets matter because a customer who gives up looks like a resolved one.
- Voice, outbound messages and where it runs
- The agent can answer phone calls through a telephony provider such as Twilio, passing the speech to the agent and transferring to a person with the caller's details. Tell callers it is an AI and that calls may be recorded, and check the rules for that in each country you serve. WhatsApp only allows free replies for 24 hours after the customer's last message; after that only approved templates can be sent. The system runs in the cloud or on your own servers.
What you can connect to it
- Zendesk, Freshdesk, Salesforce Service Cloud
- Reads and updates tickets, cases and contacts, adds the summary when it hands over, and closes tickets only under the rules you set.
- Online store (Shopify, WooCommerce and others)
- Reads orders, fulfilment status and return rules, and starts a return within your limits.
- WhatsApp Business Platform
- Receives customer messages through the official API and replies in the chat, using approved templates outside the 24-hour window.
- Gmail and Outlook shared mailboxes
- Reads incoming support mail, drafts or sends the reply, and files the thread with the ticket.
- Website chat
- Answers visitors from your own pages and FAQ, and collects a name and phone number for a person when it cannot answer.
- ERP or order management system
- Reads stock, order history and delivery dates for trade customers.
- Courier and delivery services
- Reads tracking and requests redelivery or a change of address through the courier's portal or API.
- CRM (HubSpot, Salesforce, Zoho and others)
- Matches the customer, reads their history and writes the conversation back.
- Telephony provider (Twilio and others)
- Receives calls and sends speech to the agent, then transfers to a person with the caller's details.
- Google Sheets or Excel
- If you have no help desk, the agent keeps a log of questions and tickets in a sheet your team already uses.
Connects to your tools. You stay in control.
- Works with what you have
- Website chat, email, WhatsApp Business Platform, phone, Zendesk, Freshdesk, Salesforce Service Cloud, Shopify, your order system or ERP, CRM, courier portals, spreadsheets.
- A person approves what matters
- Payments, orders, discounts or anything you choose wait for approval. Every action is logged.
- Can run on your own servers
- The agent, its data, its logs and even the language model (open models such as Llama, Qwen or Mistral) can run on your servers, so documents and customer data stay inside your network. How that works
What it does not do
- It does not decide refunds, compensation or exceptions to your policy unless you allow routine cases up to a set amount.
- It does not invent policies, prices or delivery dates. When your content or systems have no answer, a person gets the conversation.
- It does not give legal, medical or safety advice. Those conversations go to a person at once.
- It does not pretend to be a person. If a customer asks, it says it is an AI agent and offers a human.
- It does not close a complaint by itself, and it does not message customers who never contacted you or agreed to hear from you.
Questions
Will customers know they are talking to an agent?
Yes, we recommend telling them. It sets expectations, and customers can always ask for a person.
Is this a chatbot for our website?
It can sit in your website chat, but it does more than a chatbot. A chatbot answers from a script. This agent looks up orders, starts returns, makes bookings and hands over to your team with a summary, on the website, email and messaging apps alike.
What if it gives a wrong answer?
It answers only from content you approved and data read live from your systems, and it can only take the actions on its list, so the harm a mistake can do is small by design. Every conversation is logged, and when a wrong answer shows up we fix the article or rule behind it.
How does it know our policies?
You give it the documents: help articles, return and warranty policy, delivery rules, product data. It retrieves from those and cites the one it used. Before launch we test it with the real refund and warranty questions your customers ask.
Can it answer phone calls?
Yes, through a telephony provider such as Twilio. It handles simple questions and transfers to a person with the caller's details. Callers should be told it is an AI and that the call may be recorded, and the rules differ by country, so we check them with you.
Do we have to change our help desk or software?
No. It works with the tools you have, such as Zendesk, Freshdesk, a shared Gmail or Outlook mailbox, and your store or ERP. If all you have is a mailbox and a spreadsheet, it can work from those.
Where does our customer data go?
It stays in your systems and in the logs of the server we set up for you, in the cloud or on your own premises. We agree with you which language model provider is used before anything goes live.
How will we know it is helping?
You will see it in your own numbers. Before launch we note your current first-response time and tickets per week, and afterwards the logs show what the agent resolved, what was reopened and what went to your team. Vendor figures from other companies say little about yours.
How much does an agent like this cost?
It depends on the task and the systems it connects to. After a first call we give you a clear estimate of cost and time.
Want this agent built around your work?
Tell us in a few lines what the job looks like today and we will set up an online intro call.
Request sent
We will contact you to set up an intro call.