Custom Chatbot Development
Custom chatbot development for support, sales and internal use. Your chatbot answers from your own content and hands over to a person when it should.
Why most chatbots fail
Most custom chatbot projects fail for two reasons. The bot has nothing solid to answer from, so it guesses. And it has no way out, so a customer who needs a person gets stuck in a loop and ends up angrier than when they started.
A better AI model won't fix either. The work is in the content underneath, the limits around it, and the handover when it runs out of answers. That's where we spend our time.
What we build
Customer support. Trained on your help pages, order data and past tickets. It answers the repetitive questions and passes the rest on with the context attached. The realistic goal is to handle a good share of first-line questions, not all of them.
Sales. It answers product questions, checks whether an enquiry fits what you do, and books the ones that do. Its biggest advantage is that it works at nine on a Sunday evening.
Internal knowledge. Often the one that saves the most time. New staff ask how a process works and get today's answer, instead of a document from two years ago.
How a project runs
We start by reviewing what you have: which documents are current, which contradict each other, and what your support inbox is really full of. A week of that tells us whether a chatbot is the right answer. About a third of the time, we find that a better help page or one fixed process would do more.
If a chatbot is the right answer, we build a focused version first, covering only your most common topics, and put it in front of real users. A bot that answers a few things correctly earns the trust you need to expand it later.
Costs to plan for
AI model usage is a running cost, and it grows with the number of conversations. We'll work it out for your volume before you commit.
Your chatbot also needs upkeep. Prices and policies change, and a chatbot answering from last quarter's information is worse than none. We can train your team to keep it up to date, or do it for you on a monthly retainer. We'll tell you which makes more sense.
How it works
- Answers from your own content
- Your chatbot answers from your documentation, product data, policies and past support tickets, and shows which source it used. That's what makes it a tool your team trusts, and stops it inventing a refund policy one Tuesday. Getting your content ready is part of the job, and it's usually where most of the effort goes.
- It says when it doesn't know
- The worst thing a support bot can do is give a confident, wrong answer. We set a clear limit: if a question falls outside its material, it says so and hands over. Before launch we test it with deliberately awkward questions, including the ones your customers ask when they're annoyed.
- A handover that works
- When the bot can't finish a conversation, it passes it to a person on your team, with the transcript, in the tools your team already uses. Nobody has to repeat themselves. A bot with no way out is a customer service problem.
- You choose where it lives
- On your website, in WhatsApp, in Slack or Teams for internal use, or behind your own login. We've built all of these. The channel changes the interface and some limits, and most of the work stays the same.
- A clear target
- Before launch we agree what it's for: handling a share of repetitive tickets, qualifying enquiries out of hours, or answering product questions before checkout. Then we measure that and show you the conversations it got wrong, because those teach you the most. Without a target, you can't judge a bot or improve it.
Projects like this
What's included
Included
- A review of your existing content and support history, first
- A focused first version, live, covering your most common topics
- Handover to a person, with the conversation history attached
- Testing with awkward and hostile questions before launch
- A written target, agreed before we build
- A monthly review of failed conversations for the first three months
Not included
- Writing the documentation it answers from, unless quoted separately
- AI model usage, which is billed to your own account
- Staff to handle the conversations it hands over
- Voice or phone bots, which are a different build
What we build with
- Python
- TypeScript
- Next.js
- OpenAI
- Anthropic
- PostgreSQL
- pgvector
What does a custom chatbot cost to build?
And what does it cost to run?
How do you stop it inventing answers?
Can it hand over to a human?
Where can it live?
How will we know whether it worked?
Also in AI & automation
Tell us about your project.
Get in touch ↗