AI customer service set-up.
AI that answers your customers, and AI that helps your own team answer. Built from information you've checked, tested on your real questions, and always with a way through to a person.
AI in your customer operation
Two places we set it up. In front of your customers, and beside your team. Separate pieces of work, often in the helpdesk you already pay for.
In front of your customers
An AI agent that answers the questions customers ask most by chat, email and messaging, at any hour, from help articles you have checked and in your tone of voice. When it can’t answer, or shouldn’t, it hands over to a person.
Beside your team
AI that helps the people who answer the rest, with suggested replies they can check and change, summaries of long conversations, the right article as they type, and messages sorted and labelled as they arrive.
In front of your customers
One question, three endings.
An AI agent doesn’t just reply. Pick a question and watch what it does with it.
The conversation
Can I change the delivery address on my order?
AI agent
Yes, you can, right up until it’s packed. Open your order, choose “Change address” and you’re done.
What the AI does with it
- Understands the questionA delivery change
- Checks it may answerYes, it’s on the list
- Finds the checked articleFound “Changing your address”
- Writes the reply in your toneShort and friendly
- Checks the reply against the articleMatches the article
Answered, from an article you’ve checked.
What it takes
Treat it like a new starter.
You'd give them the handbook, tell them what not to touch, sit beside them at first and keep correcting them. Setting up AI is the same work.
What it knows
The help articles and policies you have checked, and nothing else. It is only as good as what it answers from.
What it leaves alone
The topics that always go to a person, agreed with you and written down before it meets a customer.
How it sounds
Your tone of voice, and how long an answer should be, taken from the replies your team already gives.
What it can look up
An order, a booking or an account, where your systems allow it, so it can answer about that customer.
Beside your team
At the hand-over, nobody starts again.
The customer carries on the same conversation. The person picking it up sees everything, with a head start.
My order arrived damaged. Can I get a replacement?
AI agent
I’m sorry about that. Which item was it?
The blue lamp. And I’d rather talk to someone.
AI agent
Of course. I’m passing you to the team with all of this, so you won’t need to explain again.
Your team
Hi, I can see the lamp arrived damaged. I’ll get a replacement sent out today.
- Asked for
- A replacement for a damaged item
- Already told us
- The blue lamp, from their last order
- The AI checked
- Still within the returns period
- Passed over because
- They asked for a person
Suggested reply, from the AI beside your team
“Hi, I can see the lamp arrived damaged. I’ll get a replacement sent out today.”
Theirs to check, change and send.
Without a proper hand-over, the first thing your customer hears from a person is “Can you tell me what the problem is?”
Where you’re starting from
Which of these is you?
Pick the one closest to where you are now.
Starting from nothing
The right place to start is the questions your customers ask most, not the tool. We work out what it should handle, and get what it answers from right before any customer meets it.
- We decide what it should handleThe simple, common questions first, and a written list of what always stays with people.
- We get the knowledge rightHelp articles checked, gaps filled, anything out of date taken away.
- We test it on your real questionsYour customers’ actual conversations, the awkward ones included. Not a demo.
- We switch it on carefullyOne channel or one kind of question first, then more as it proves itself.
Fixing one that isn’t working
Usually the tool is fine. It was pointed at a help centre nobody had checked, and told to answer as much as it could.
- Confident answers from out-of-date information
- Customers asking for a person straight away
- Nobody reading what it actually says
- We read what it got wrongThe conversations themselves, and where each wrong answer came from.
- We fix what it answers fromThe articles that are out of date, missing, or contradict each other.
- We set what it must leave aloneThe topics that always go to a person, and a hand-over that carries the conversation.
- We test it again before it goes backOn the questions it failed, and on new ones.
Getting more from it
Once it handles the simple questions well, there is more it can safely do, and your team can have AI of their own.
- More kinds of questionAdded one at a time, each tested before it goes live.
- Letting it look things upAn order, a booking or an account, so it can answer about that customer.
- AI beside your teamSuggested replies, summaries and the right article to hand, set up as its own piece of work.
- A routine for checking itWhat to review, how often, and who owns it.
Honest expectations.
AI is good at the questions customers ask most, from information you've checked. It isn't a replacement for your team, and it's only as good as what it answers from.
- Resolved isn't the same as rightA confident wrong answer can look like a success. Someone has to read what it actually said.
- Your team's corrections are a signalWhere people change what it suggested shows where the knowledge needs work.
- It needs an ownerSomeone keeping its knowledge up to date as your products and policies change.
AI answer check
Returns question
- Matched the returns policy
- Used last year's opening hours
- Handed over when asked
Article updated, answer re-checked

A person within reach
Good AI still leads to people.
The best experience is a quick, right answer when the question is simple, and a person when it isn't.
- Customers can always ask for a person
- Anything sensitive, or anyone upset, goes straight to your team
- Your team spends its time on the conversations that need judgement
Questions, answered.
The things people ask us first.
What can AI realistically handle for a business like ours?
The simple, common questions, answered from information you've checked, such as where an order is, how returns work or opening hours. Anything that needs judgement stays with people.
Should we use the AI built into our helpdesk, or a separate tool?
Often the one you already pay for is the right start. We'll tell you if something else would serve you better, and why.Helpdesk set-up
How does the cost of AI work?
AI that answers your customers is usually charged by the conversation it handles, so it is worth knowing what your helpdesk counts as “handled” before you switch it on. AI that helps your team is usually an extra per person. Both vary between makers, so we’ll help you work out what yours would come to.
Is our help centre good enough?
We'll find out early. If it isn't, we'll help you fix it first, because the AI can only be as good as what it reads.Processes and knowledge base
Will it sound like us?
We set its tone from your saved replies and help articles, and test it on your real conversations until it does.
What does it mean for our team?
Fewer repetitive questions, and more time for the conversations that need a person. It helps them and does not replace them.
Can we set up AI for our team without putting it in front of customers?
Yes. They are separate pieces of work. AI beside the team is often the easier place to start, because a person checks every reply before it is sent.
Can you fix an AI we have already switched on?
Yes. We start with what it is getting wrong, and why, which usually leads back to the knowledge it answers from.
Who keeps it up to date?
Someone on your team, with a routine we set up together. Or us, if you'd like regular checks of what it tells your customers.
Alongside it
What AI depends on.
An AI agent is only as good as what it answers from and who it hands over to. We can look after those as well.
- Processes and knowledge baseThe checked articles it answers from.We set it upWe set it up
- Helpdesk set-upWhere it lives, and where the hand-over lands.We set it upWe set it up
- ReportingWhat it answered, what it passed on, and why.We set it upWe set it up
- AI chatbot quality checksPeople reading what it tells your customers.We run itWe run it
- Customer serviceThe people it hands over to.We run itWe run it
Let's talk.
Tell us what's going on. If we're not the right fit, we'll point you to someone we trust.
Book a call







