Answer repeat questions from your own articles. Pass the rest to a person.
Customers ask the same things all day. A chatbot on your website or app answers them from the articles your own team approved, in the language the customer wrote. When it cannot answer, the chat goes to a person with the whole conversation attached, and the customer is never asked to start again from the top.
Open last month’s tickets. A clear pattern shows up. Where is my parcel? Can I change the address? How do I reset the login? Your agents know each answer by heart, yet they type every one by hand while the harder cases wait behind them in the queue.
A chatbot takes the first layer. It answers from articles your own team wrote, and where your order or helpdesk software allows it, it also looks up a status for the customer. Where a question falls outside what it was given, it stops at once and offers a person, rather than inventing something that sounds plausible. Staff read the chats afterwards. They correct every answer that missed.
The problems this is meant to ease.
What AI chatbots does.
What this looks like in a working week.
The people who open this every day.
From first conversation to daily use.
Discovery with the support desk
We sit with the people who answer messages and list the questions that come up most. Past tickets, the inbox and the WhatsApp history show which topics are worth handing to a bot.
Getting the articles ready
The approved answers are collected, checked for age and conflicts, and written into the knowledge base. Order, booking and helpdesk access is arranged only where your software allows it.
A trial on real past chats
Everything is run against a few hundred earlier conversations, with nothing sent to a customer. The answers it would have given are laid beside what your agents actually wrote.
Review with agents and the lead
The people who use the answers read the misses and the doubtful cases. They edit articles, switch off topics that need a person, and set the handover rules in plain words.
Going live on one channel
Go-live starts on a single channel, usually the website, with a visible route to a person. Staff watch the first chats live. They can pause it at will.
Reviewing what happened afterwards
After a few weeks the team reads the declined questions, the handovers and the ratings together. New articles are written, topics are widened or narrowed, and a second channel is added only then.
How you would know it is working.
Record where you stand before it goes live, so the change is measured against your own figures.
- Resolution rate counts chats that end without a person. The month of chats your agents answered by hand before go-live gives the starting figure.
- Escalation rate is how many chats are handed to an agent, measured from the helpdesk ticket history before the bot starts.
- First response time measures the wait from a customer’s opening message to the first reply. Inbox timestamps from before go-live give the starting figure.
- Declined questions are counted weekly from the review list, with the first month of the list serving as the baseline.
- Customer ratings on finished chats are compared with ratings or survey replies your support desk already collects.
Where a person stays in charge.
A person approves anything that changes an order
A refund or address change is collected and attaches the facts to a ticket. A member of staff decides. Money and orders stay untouched.
Customers are told it is a bot
The first line says so and shows how to reach a person. Countries set their own duties on telling people they are talking to a bot. Your lawyer reviews the wording, and the software only supplies the text.
Personal data stays limited
Only the fields you allow are visible to it, such as an order status after a phone number match. Chat logs are kept under the retention period your adviser sets, and access is limited to named staff.
Answer quality follows article quality
Old or missing articles produce weak answers. Every answer names its source page, and the review list shows where the gaps are. A named owner keeps each article current.
It is one part of a system, not an island.
A module earns its place by what it passes to the next one. These are the connections we set up most often.
- Helpdesk and ticket management
- Knowledge base
- Order management
- WhatsApp Business
- Your website chat widget
Products that include it.
How we would put it in.
Questions owners ask about AI chatbots.
What happens with a question it cannot answer?
It says plainly that it cannot help and offers a person or a callback. It does not guess. The question lands on the review list, so somebody can write the missing article before the next customer asks.
What do we need in place before this works?
You need written answers of some kind. A plain document of fifty questions and replies is enough to start. With nothing written down, the bot has nothing to quote, and writing that document is the first job. It is yours.
Can it issue refunds or change an order without a person?
No. It collects the request and attaches the order details to a ticket, which is as far as it goes. A member of staff approves and carries out anything that moves money or alters an order.
How do we know its answers are correct?
Each answer names the article it came from, and staff read sampled chats every week. Mistakes usually trace to an old or missing article. Fix the article and the answer changes for every customer.
What does it connect to, and what if we have no system?
It can read order, booking and helpdesk records where your software opens them up, and we check that before work starts. Without such a link, it still answers from articles and takes messages for staff.
Who sees the chats, and is customer data safe?
Named staff see the logs, and the bot reads only the fields you allow. Your own adviser decides which privacy and retention rules apply in each place you serve, because those duties are set locally and change between regions. The software prepares the data. It offers no legal opinion.
When is a chatbot not worth setting up?
Skip it when you receive only a handful of messages a day, or when most of them are long and personal. A shared inbox serves you better there. A bot earns its place on short, repeated questions.
What is the effort and cost in plain terms?
Most of the effort is yours: writing and checking the answers. Our part is setup, links to your systems and a trial. Cost depends on channels, languages and integrations, and it is quoted after discovery.
Businesses that ask for this.
Modules that work with it.
Tell us how you handle AI chatbots today.
A spreadsheet, a register or another system: say which, and we will tell you plainly what is worth changing.
- No obligation
- A reply within one business day
- Your details stay private
Start with your requirements
Seven short steps that tell us what you need. An engineer reads them before getting in touch.
Share your requirements