Chatbot development services for support, sales and internal help
Our chatbot development services build assistants that answer from your own content and data, not from the open internet: product questions, order and booking status, opening hours, policies, and the internal questions staff ask every day. Every chatbot we build shows where its answers come from, says plainly when it does not know, hands the conversation to a person when it should, and tells users they are talking to an AI, as EU law now requires. It runs on your website, app or chat channels, hosted in the EU, and your company owns it.
- You see the result before it goes live
- Fixed price for the agreed scope, in writing before work starts
- Source code, domain and accounts in your company's name
What a good chatbot does, and what it should not do
A chatbot is worth building when the same questions arrive many times a day, the answers exist in your content or systems, and a fast answer is worth more to the customer than a perfect one later. Typical examples are delivery and returns questions in a store, booking changes for a service business, product and compatibility questions, and internal help for staff on policies, tools and procedures.
It is the wrong tool for complaints that need judgement, negotiations, medical, legal or financial advice to individuals, and anything where a wrong answer is expensive and hard to reverse. Those go to a person, and the chatbot's job is to recognise them and hand them over with the context already collected.
Types of chatbots we build
| Type | How it answers | Fits when |
|---|---|---|
| Guided chatbot | Buttons and fixed flows written in advance | A few predictable tasks, strict wording |
| Knowledge chatbot | A language model that answers only from your approved content, with sources | Many questions, answers in documents and pages |
| Task chatbot | Knowledge answers plus lookups and actions in your systems | Order status, bookings, account questions |
| Internal assistant | Answers from company documents, with access by role | Staff questions on policies, tools and procedures |
Most projects combine two: guided flows for the few tasks that must follow exact steps, such as identity checks or cancellations, and a knowledge chatbot for everything else. Hotels and restaurants use the same pattern for guest messaging; see artificial intelligence in hospitality. For the steps behind any AI feature, see how to make an artificial intelligence program.
Answers from your own content
The reliability of an AI chatbot depends less on the model than on what it is allowed to answer from. We collect and clean the content first: help pages, product data, policies, price lists where you publish them, and the answers your support team already gives. That content is indexed so the chatbot can retrieve the passages relevant to each question, answer from them, and link to the source. When the retrieved content does not cover the question, the chatbot says so and offers a person instead of guessing.
The same retrieval layer can also serve employees as an internal AI knowledge base; see RAG development services.
The content has an owner. When a policy or product changes, the index is updated from the same source your website uses, so the chatbot does not keep repeating last year's rules.
A company answers for its chatbot
Courts have already decided that a chatbot speaks for the business that runs it. In Moffatt v. Air Canada, 2024 BCCRT 149 (decided 14 February 2024), the Civil Resolution Tribunal of British Columbia held the airline liable for negligent misrepresentation after its website chatbot gave a customer wrong information about bereavement fares, and rejected the argument that the chatbot was responsible for its own actions. The practical lesson is the design above: answers grounded in approved content, clear limits, and a handover to people for anything the content does not settle.
EU rules: AI Act and GDPR
Article 50(1) of the EU AI Act requires providers to design AI systems that interact directly with people so that the people concerned are informed they are interacting with an AI system, unless that is obvious from the context. Article 113 sets the general application date of the Act at 2 August 2026, and Article 50(1) has no later date of its own. The separate duty in Article 50(2) to mark AI-generated audio, image, video or text in a machine-readable format has a transition for systems placed on the market before 2 August 2026, which must comply by 2 December 2026 under Article 111(4), added by the Digital Omnibus on AI (AI Act Service Desk, Article 50, Article 111 and Article 113, consolidated text as at 27 July 2026, read 11 November 2026). The information must be clear and given at the latest at the first interaction, so our chatbots say it in their first message.
Conversations often contain personal data, so they are treated as such. We collect only what the conversation needs, keep logs for a period you set, mask payment and identity numbers, and run the system in an EU data region under a data processing agreement. Model providers are configured so that your conversations are not used to train their models, and users can reach a person and ask for their data like anywhere else in your service.
A chatbot is one entry in a wider set of controls. Our AI governance guide covers the inventory, risk classes, AI literacy and deployer duties for every AI system a company uses.
Security of AI chatbots
A chatbot that reads text from users and documents can be manipulated by that text. The OWASP Top 10 for LLM Applications 2025 (read 11 November 2026) lists prompt injection first and excessive agency sixth, and both shape how we build:
- Least privilege. The chatbot can only reach the systems and records its task needs, and only for the identified user.
- Actions confirmed. Anything that changes data, such as a cancellation or an address change, is shown to the user for confirmation and logged.
- Instructions kept apart from content. Text in documents or messages is treated as data, not as instructions, and system prompts hold no secrets.
- Output checked. Links, prices and figures in answers are checked against the source data before they are shown.
- Limits on use. Rate limits and spending caps prevent abuse from running up model costs.
Handover to people
The handover is where many chatbots fail. Ours pass the conversation, the identified customer and the question to your support tool or team inbox, so the person who takes over does not ask everything again. Outside opening hours, the chatbot says when a person will reply and creates the ticket. Users can ask for a person at any point, and the chatbot does not argue.
Measured after launch
Before launch, the chatbot is tested against a set of real questions from your support history, with the expected answers written by your team. After launch we report every month on the share of conversations resolved without a person, handovers and their reasons, questions the content did not cover, and a sample of answers reviewed by a person for accuracy. Gaps in the content are fixed at the source, which also improves your help pages.
Ready-made chatbot or custom build?
If your assistant should also take actions in other systems, not only answer, read our chatbot vs AI agent guide first.
Help desk products such as Intercom, Zendesk and Freshdesk include AI chatbots, and if your support already runs on one of them, its built-in assistant is often the right start. A custom chatbot pays off when the answers depend on your own systems and rules, when it has to work across several channels or brands, when an internal assistant needs access by role to company documents, or when you need control over where data is stored and which model is used. We will tell you which applies. For the wider picture of where AI pays back in a company, see AI consulting; for automating the work behind the conversation, business process automation and API integration. Chatbots are one of several NLP applications. Our guide to chatbot development cost explains build and running costs per conversation.
How the project runs
- Questions first. We read a sample of your real support or staff questions and sort them into what the chatbot should answer, look up, or hand over.
- Scope and fixed price in writing. Topics, channels, systems it connects to, languages, and what the first version leaves out.
- Content and build. Content cleaned and indexed, flows and connections built, notices and handover in place.
- Test on real questions. Your team reviews answers to the test set until accuracy meets the agreed level.
- Launch, then measure monthly. Start on one channel or for one group of users, then widen with the monthly report as the guide.
As a standard term, code, prompts and the content index sit in your company's own accounts. See prices.
Questions we get
What does a chatbot development service include?
Choosing what the chatbot should and should not handle, preparing the content it answers from, building the conversation and the connections to your systems, testing it against real questions, launching it on your website, app or chat channel, and measuring its answers after launch. We also design the handover to a person for everything the chatbot should not answer alone.
Will the chatbot make things up?
A language model can produce confident wrong answers, so we design against it. The chatbot answers from your approved content and data, shows where an answer came from, declines when the sources do not cover a question, and hands the conversation to a person. We test it with real questions before launch and sample its answers every month after.
Do we have to tell users they are talking to a bot?
In the EU, yes in most cases. Article 50(1) of the AI Act, which applies from 2 August 2026, requires AI systems that interact directly with people to be designed so that people know they are dealing with an AI, unless that is obvious. We build that notice into the first message.
Can the chatbot look up orders or bookings?
Yes, through your systems' interfaces. The chatbot can check an order status, a booking or an invoice after the user is identified, and it only sees the records that user is allowed to see. Actions that change something, such as cancelling a booking, are confirmed by the user and logged.
Which channels can it run on?
Your website, your mobile app, and messaging channels such as WhatsApp, Microsoft Teams or Slack, depending on who uses it. An internal help assistant for staff usually lives in Teams or Slack, a customer chatbot on the website and in the app.
Who owns the chatbot and the conversation data?
Your company. The code, prompts, content index and conversation logs sit in accounts in your company's name, in an EU data region, and model providers are configured so that your data is not used to train their models.
Get a written quote
Tell us in a few lines what you need. We reply within one working day with a written quote: scope, timeline and a fixed price or an hour estimate. Asking is free and commits you to nothing.