Does your business even need an AI chatbot?
That’s the question worth asking before any call with a vendor. An AI chatbot makes sense when customers ask repetitive but varied questions - about your offer, availability, terms of service - and when “we’ll get back to you shortly” is costing you sales. If your business handles 20 inquiries a week, a plain contact form will do the job cheaper and faster than an AI deployment.
An AI chatbot starts paying off where traffic is steady, questions repeat, and customers expect answers outside office hours. With our clients, that’s usually the point where the front desk or customer service team starts “falling behind” - not because people are working badly, but because volume is growing faster than the team.
What types of chatbots are there - and how do they actually differ?
Rule-based chatbots
A simple decision tree: the customer clicks an option, the bot replies from a fixed pool of answers. Cheap, predictable, but rigid - any question outside the script ends in “I don’t understand” or a handoff to a human. A good fit for very narrow use cases (e.g. order status lookup by number).
AI chatbots built on language models (LLMs)
They understand context, can answer questions phrased in many different ways, and learn from your documentation - product catalog, price list, terms, FAQ. This is the technology behind assistants like Claude from Anthropic or ChatGPT from OpenAI - the same models also power business deployments, not just consumer chat apps.
Hybrid chatbots
A combination of both approaches: a fixed flow for simple, frequent paths (like booking an appointment), and a language model for everything that doesn’t fit the script. In practice, most of our deployments are hybrids - they give you predictability where it’s needed and flexibility where a customer asks something nobody anticipated.
What does it actually cost?
Pricing depends on how many conversations or bookings the agent needs to handle each month. Our two most commonly deployed solutions in this space:
Customer Care Specialist - answers questions on your website and social media, knows your catalog, pricing and company policies, collects contacts and hands hot leads to your team:
- Starter tier: up to 500 conversations/month for 790 PLN
- Pro tier: up to 2,500 conversations/month for 1,490 PLN
Booking Assistant - books appointments via phone, SMS and social media, sends reminders, reschedules on its own and fills calendar gaps:
- Starter tier: up to 200 bookings/month for 690 PLN
- Pro tier: up to 1,000 bookings/month for 1,290 PLN
These are prices for a finished, deployed solution - not “API access.” The price includes configuration for your business, integration, and ongoing maintenance.
How long does deployment take?
A typical AI chatbot deployment takes us 2-4 weeks, depending on how many systems need to be connected and how much material (documentation, pricing, conversation history) needs to be prepared as a knowledge base. The process looks like this:
- Week 1 - we gather materials (offer, FAQ, terms, sample conversations) and decide where the chatbot needs to run
- Week 2 - we build the first version and connect integrations - most often your CRM and knowledge base
- Weeks 3-4 - testing against real customer questions, fixes, tuning tone and scope of answers
Deployment is iterative - we don’t hand over a finished bot on day one; we refine it based on real conversations before it goes live to full traffic. With our clients, the first two weeks after launch usually see the most fixes - after that, the bot stabilizes and only needs minor updates as your offer changes.
Which integrations are worth connecting from day one?
A chatbot without access to your data is, at best, a nice-looking FAQ search box. Two integrations we do almost every time:
- CRM - so a hot lead reaches the right salesperson right away, instead of getting lost in chat history
- Company knowledge base - product catalog, pricing, terms, documentation - so the bot answers from current data, not from something someone typed by hand six months ago
Depending on the channel, we also connect WhatsApp, Messenger, or a website widget. The more systems that need to be connected, the closer you get to 4 weeks rather than 2.
What questions should you ask before deployment?
- Where will the chatbot run? Website, WhatsApp, Messenger, or maybe an internal company channel? Each channel is a separate integration.
- What data will it process? If the chatbot has access to customer data, that needs to be planned for GDPR compliance from day one, not tacked on at the end of the project.
- How will you measure success? Fewer calls to the front desk? More leads passed to sales? Faster response time? Without a clear metric, it’s hard to tell whether the deployment paid off.
- Who keeps the knowledge base updated? A chatbot is only as good as the data it runs on - someone on your side needs to keep pricing and offer information current.
What does this look like in practice - a production example
The best proof that an AI agent is actually built to work, not just to look good in a demo, is what happens after the project is handed over. In our case study on Parlour Dev we cover an agent deployment for a software house - a client that builds software for a living and won’t accept anything below its own standards. The agent went live in production on the day of handover, with no stabilization period and zero fix requests, along with full technical documentation delivered to the client’s team.
How does Basalt AI deploy chatbots?
We build chatbots on Claude and GPT-4 models, matching the model to the use case - not every task needs the same engine. Every deployment is integrated with your CRM and knowledge base, and we run the process iteratively: from a first version, through testing on real questions, to stable operation at full traffic. If you’re wondering whether your business is ready for an AI chatbot or actually needs something simpler, that’s the first question we answer together, before we write a single line of integration.