A chatbot development agency that sets the limits first
We build bots for your website or Telegram that answer routine questions, take down the enquiry and hand anything that needs a human decision to your team. Answers come only from your own rules and data.
Bring in a chatbot development agency when customers ask the same things every day and replies arrive hours later. A bot that invents prices or promises what you do not offer does more harm than no bot, so we agree its limits before we build anything.
When a bot makes sense
when it fits
- Customers keep asking about prices, opening hours, delivery or order status, and the answers are always the same.
- Enquiries arrive in the evening or at weekends and get a reply the next working day.
- You already have prices, rules and standard answers that can serve as the bot's source.
when not now
- Every request is different and needs advice from a specialist.
- Prices and rules change daily and nobody has time to keep them up to date.
- You get a handful of messages a week. A person will answer them faster than a bot pays for itself.
How we build the bot
-
01
Scenarios
We collect the questions customers ask most and decide which ones the bot answers itself and which go to a person.
-
02
Answer source
Prices, rules, opening hours and help texts sit in one place that your team can update without a developer.
-
03
The limit
The bot does not offer discounts, change orders or give legal or medical advice. Those requests go straight to a manager.
-
04
The enquiry
Name, contact details and the request land in your CRM, a spreadsheet or a team chat, and the person responsible is notified.
-
05
Testing
Before launch we test on real customer questions, then read the logs to see where the bot gets things wrong.
THE QUESTIONS
WE HEAR MOST.
Website chat or Telegram?
It depends on where your customers already write to you. We usually start with the channel that gets the most messages and add others once the scenarios have proved themselves.
AI or buttons?
Buttons are reliable for simple flows such as opening hours, prices or bookings. AI helps when people phrase the same question in many ways. Either way, answers rely on your data and risky actions wait for a person to confirm them.
How much does chatbot development cost?
It depends on the number of scenarios, channels and integrations with your CRM or accounting. After a review we put the scope in writing.
What if the bot gives a wrong answer?
That is why we set a limit: the bot answers only from an agreed source and passes anything unclear to a person. The logs show where it went wrong, so the scenario can be corrected.
Do you have a chatbot case study?
Not a public one yet. The same principle runs our Klar mail triage pilot and the content engine we built for BETSA: the system does the repetitive part, people make the risky decisions. In a review we show how this would look for your scenarios.
Other directions
Sound like
your task?
Tell us where customers contact you, which questions come up most and where enquiries should end up. We will reply in writing with the channel and scenarios to start with.
Chatbot for business: what the work involves
A chatbot development agency is worth hiring when the answers are known in advance and the customer wants them now: prices, opening hours, delivery terms, order status, a free slot for an appointment. Everything else the bot should hand to a person, along with what the customer has already written. Below we explain what the work includes, how it runs, what drives the price, who needs a bot and who does not, and which mistakes to avoid.
What our chatbot development services include
The bot itself is only the visible part. To stop it from making things up or losing enquiries, a few more pieces are needed:
- a scenario map: which questions the bot handles itself, which it hands over, and with what information;
- an answer source with your prices, terms, opening hours and rules, which your team can update without a developer;
- the bot’s copy: the greeting, follow-up questions, answers and the line it uses to pass the conversation to a person;
- delivery of each enquiry to your CRM, a spreadsheet or a team chat, with a notification to whoever is responsible;
- a conversation log that shows where the bot misunderstood a question or got the answer wrong.
How the work runs: Telegram or your website
We start with real conversations. We ask for a few dozen past chats with personal details removed and write down what people actually ask, not what the team assumes they ask. From that we build the list of scenarios and agree the limit with you: where the bot answers and where it calls a person straight away.
Then we put together the answer source and write the copy. For a Telegram bot we plan the commands and buttons people see first. For a website chat we decide which pages it appears on and what it asks. The two are built differently, but they must draw on the same answers, or a customer will get two different replies to one question.
Before launch we run the bot through real questions together with your team. After launch we read the log and adjust scenarios where the bot gets things wrong or where people keep asking for a human.
What drives the price
We name a figure after the review. It depends on:
- the number of scenarios and how many branches each one has;
- the channels: Telegram only, website only, or both;
- buttons or AI, since an AI bot needs more testing and a more carefully set limit;
- integrations with a CRM, accounting, booking or payments, each with its own constraints;
- the languages the bot has to answer in;
- whether an answer source already exists or has to be pieced together from chats and from what your managers know;
- who updates the scenarios after launch: your team or us.
An AI bot also has running costs for the language model, which grow with the number of conversations. We put the scope in writing before we start.
Who needs a chatbot for business, and who doesn’t
A bot fits an online store that gets daily questions about order status and delivery. A service business where people look for a free slot in the evening. A B2B company that needs input before it can quote: volume, dimensions, deadlines, location. The bot asks for those details up front, and the manager receives an enquiry they can actually work with.
A bot will not help if the problem is not reply speed but the fact that nobody follows up on enquiries; it will just pile them up faster. It is also out of place where customers expect to talk to a specific person, as in a high-end personal service.
We tested the principle of “the machine does the repetitive part, a person makes the risky calls” on our own product. In the Klar pilot, between 13 June and 8 July 2026, the system sorted work mail from five inboxes at PILLAR, Betsa and GrandMa, created 91 tasks and sent 148 digests, while people wrote the replies. The same approach runs through all our business automation.
Common mistakes
- Launching a bot before the answers are written down anywhere. A bot only knows what it is given. If your terms live in a manager’s head, collect them first; freeing up time from repeated client questions shows how.
- Hiding the way to a human. A customer who gets a menu instead of an answer three times in a row simply closes the chat.
- Collecting contact details with nowhere to send them. The enquiry has to land where someone will see it; a CRM, a spreadsheet or a team chat will do, as long as someone owns it.
- Ignoring the log after launch. The first weeks reveal questions no scenario covered.
- Building a bot where an auto-reply with prices and a link to a form would do. If that is your case, we will tell you before any work starts.
Next step
Write to us through the contact page. Tell us where customers contact you, which questions come up most, where enquiries should end up and who answers them. If you can, add a few conversations with personal details removed. We will reply in writing with the channel and scenarios to start with and what we would need from you.