AI chatbot development that answers from your content and knows when to stop

Start with the scope table

Before any code, we fill in one table with you. It decides what the bot answers, what it passes to a person and what it refuses. Here is a filled-in example for an online store.

This one page prevents most of the chatbot failures we hear about from buyers: confident wrong answers, endless loops and no way to reach a person.

Rule-based, AI or a mix

An AI chatbot reads free text and answers from documents. A rule-based bot follows buttons and fixed paths. Many good bots combine both: buttons for the five most common requests, AI for everything else. The trade-offs are in AI chatbot vs rule-based chatbot.

If all you need is opening hours and a booking link, you do not need us. A form and a clear page will do. We say that on the first call.

What happens when the bot is wrong

Every chatbot is wrong sometimes. The design question is what happens next. We build four safeguards.

• Answers come only from retrieved passages of your content, and the bot shows which page it used. See RAG chatbot development.
• If nothing relevant is found, the bot says so and offers a person.
• A thumbs-down or a repeated question triggers a hand-over with the transcript attached.
• A weekly review list shows low-confidence answers so you can fix the source content.

Here is the kind of exchange we test for. A customer asks whether a product works with a model that is not in the documents. The right reply is not a guess. It is: I could not find that in our product information, so I have passed your question to the team, who reply by email within one working day. The time shown is whatever your team really commits to.

Our process

1. Collect real questionsWe read three to six months of emails, chats or tickets and group them. This gives the scope table real content.
2. Prepare the knowledgeClean up and structure the pages and documents the bot will rely on. Gaps found here are often the most useful output of the project.
3. Build the botRetrieval, answer rules, handover, and the channel connection: website widget, WhatsApp or helpdesk.
4. Test against real questionsA set of past questions with the right answers, re-run after every change.
5. Soft launchLive to a share of visitors, with a person watching the transcripts daily.
6. Tune and hand overAdjust, then widen. You receive the content pipeline, test set and admin guide.

What you get, how long it takes and what it costs in the market

• A chatbot on the channels you choose, with your branding and tone.
• A scope table, test set and admin guide.
• Conversation logs and a topic report so you see what customers ask.
• A privacy note covering what is stored, for how long, and the disclosure that the user is talking to an AI, as GDPR, UK GDPR and consumer rules in several countries expect.

A focused bot on one channel and one content set can reach soft launch within a few weeks. Several channels, order lookups or a large messy knowledge base add time. Your quote gives dates for your scope.

For context, market data from Clutch puts typical AI development projects between $10,000 and $49,999, with an average of $120,595. That covers everything from chatbots to custom models. Our price for your bot is a fixed quote after a short call; start from the contact page. The AI chatbot cost calculator gives a first sense of what drives cost.

How we report and related work

After launch you get a short monthly note: conversations, share resolved without a person, hand-overs, top unanswered questions and content fixes we suggest. The numbers come from the bot's own logs.

We have no published chatbot case study yet. The nearest real work is Interakt, where we built landing pages for a digital communications company on a two-week deadline. It shows how we handle fast, focused delivery, not chatbot results.