Chatbot Development Guide for Businesses
A business chatbot is developed by choosing the right type after its purpose is defined, training it with real customer questions, and connecting it to existing systems. This guide covers the types, steps, integration, and the personal data side.
Nurullah Okatan · Content Writer
· Updated: · 12 min read

For most businesses, the most balanced choice is a hybrid chatbot: it answers frequent, routine questions with rules and handles open-ended questions with artificial intelligence. The development process consists of scenario collection, training, testing, integration, and post-launch monitoring. The topics that require the most attention during setup are the protection of personal data and system connections.
What does a chatbot bring to a business?
A chatbot allows customers to get answers without waiting. It also frees representatives' time from repetitive questions. Customers want this route: according to Statista data, in 2024 82 percent of consumers said they would use a chatbot instead of waiting for a representative to answer the phone. In the same Statista survey, 96 percent of shoppers think more companies should choose a chatbot over traditional support.

The tasks that can be handed over to a chatbot are those with known answers and frequently asked questions. Order status, cargo tracking, opening hours, appointment booking, price lists, and return conditions fall into this group. What these questions have in common is that their answers are ready in a database or document.
Tasks that should stay with a representative require judgment and authority. Complaints, compensation claims, special price negotiations, and legal questions are of this type. A customer who is emotionally upset should also speak with a human. A well-built chatbot recognizes these situations and transfers the conversation to a representative along with its history.
The basic working logic and use areas of a chatbot are explained in more detail in the article what a chatbot is and what it is used for.
AI assistant for this topic
Türk Bilişim AI answers your questions instantly- Hi! Ask whatever you're curious about.
The difference between rule-based, AI-supported, and hybrid chatbots
The difference between the three types is how the chatbot finds its answer. A rule-based chatbot follows a pre-drawn flow. An AI-supported chatbot understands the question and generates an answer. A hybrid chatbot uses both within the same conversation.
A rule-based chatbot is sufficient when the number of questions is small and routine. A hairdresser's appointment bot or a logistics company's cargo inquiry bot are examples. The user progresses through buttons, and the bot does not encounter unexpected sentences. Setup is fast and the answers are fully under the business's control.
According to Microsoft's chatbot guide, hybrid chatbots combine rule-based scenarios with artificial intelligence, offering both structure and flexibility. In practice, sensitive steps such as payment and returns run on a fixed flow. Open-ended questions such as product comparisons are left to the artificial intelligence. In this way, the bot stays bound to rules on topics where the risk of giving wrong information is high.
AI-supported chatbots use natural language processing and large language models. The same Microsoft guide notes that Python is widely used in these chatbots because of its language processing and learning-from-data tools. The bot often connects to a knowledge base to find answers from the business's documents.
| Criterion | Rule-based | AI-supported | Hybrid |
|---|---|---|---|
| Answer source | Pre-written flow | Model and knowledge base | Flow and model together |
| Answering open-ended questions | Cannot | Can | Can |
| Answer control | Full | Limited, requires monitoring | Full at sensitive steps |
| Suitable business | Few and routine questions | Wide range of products or content | Businesses with both needs |
Chatbot development process step by step
The process starts with determining which questions the bot will answer. Technical development is shaped according to this list.
Usage scenarios are collected from existing channels. Call center records, the email inbox, WhatsApp conversations, and the contact form on the website are the richest sources. Questions are grouped by topic and sorted by frequency. The most frequently asked topics at first form the scope of the bot's first version. A short meeting with customer representatives also reveals questions that do not appear in the records.
The chatbot is trained with the business's own data. The product catalog, price list, return policy, FAQ page, and approved response templates make up this data. Before launch, the bot is tested with a test set prepared from real customer questions. Wrong, incomplete, or off-brand answers are corrected. Then a closed trial is conducted with a small group of users.
| Step | Work done | Output |
|---|---|---|
| Discovery | Collecting and grouping question records | Scenario list |
| Design | Drawing flows and handover points | Conversation map |
| Training | Transferring documents to the knowledge base | First version that answers |
| Testing | Test set and closed user trial | Corrected answers |
| Launch and monitoring | Regular review of records | Updated knowledge base |
After launch, conversation records are reviewed weekly. Questions the bot could not answer, conversations transferred to a representative, and negative feedback are listed separately. When a price or campaign changes, the knowledge base is updated the same day. The technical details of the steps are covered in the article how to make a chatbot.
Türk Bilişim, which is considered one of the best Android app development companies in Turkey, has been carrying out chatbot and automation projects since 2022. The company also takes on post-launch maintenance and updates for its AI chatbot and customer assistant service.
Connecting the chatbot to the website, WhatsApp, and CRM
The chatbot reads order, stock, and appointment information from the business's own systems via API. The bot does not store this information itself. When a customer asks "where is my order?", the bot obtains the order number. It then sends a query to the e-commerce infrastructure or ERP and passes the returned answer to the customer. For an appointment question, the same process is carried out with the calendar system.

Each of these connections requires a separate authority definition. The bot can read stock quantities but cannot change them. It can create an appointment but cannot see another customer's appointment. If permissions are kept narrow from the start, the harm of an incorrect answer stays limited.
The same bot can run on multiple channels. It is published as a chat window on the website and, on WhatsApp, through Meta's WhatsApp Business Platform infrastructure. Since the knowledge base is single, the answer given on both channels is the same.
Conversation records are transferred to the CRM in two ways. In the first, when the conversation ends, the summary and contact information are recorded as a note on the customer card. In the second, the bot automatically opens a new opportunity record from conversations showing sales interest. When a representative takes over a conversation, they see the previous messages on the CRM screen and do not ask the customer the same question again.
How these connections are set up is explained on the API development and integration page.
Points to watch for in chatbots regarding personal data and KVKK
A chatbot should ask for the minimum data needed to do its job. For a shipping inquiry, the order number is sufficient; name, address and phone number are not requested separately.
Some data should never be collected in the chat window:
- Turkish ID number
- Credit card number and security code
- Special category personal data such as health information, religion, ethnic origin
- Passwords and one-time verification codes
If a user voluntarily types one of this data, the bot should issue a warning and mask the record. If payment is required, the bot directs the user to a secure payment page.
The privacy notice should appear at the beginning of the conversation. A short note and a link to the full text are placed beneath the first message. The note clearly states the purpose for which the data is processed and how long records will be kept. If conversation records are sent to an AI provider abroad, this transfer is also stated in the text.
General compliance steps are listed step by step in the how to ensure KVKK compliance guide.
Looking for Corporate AI Training?Frequently asked questions
How long does it take to develop a business chatbot?
The duration varies depending on the number of scenarios and the systems to be connected. A rule-based bot that only answers FAQs goes live much faster than a hybrid bot connected to ERP and CRM. The exact timeline is determined when the scope is clarified in the discovery meeting.
Should a ready-made chatbot platform or custom development be chosen?
A ready-made platform is sufficient for a small number of template questions. For a bot that must connect to your own stock, order or appointment system, custom development is more suitable.
Can a chatbot respond in more than one language?
Yes, an AI-supported bot can recognize the language the user writes in and respond in that language. Türk Bilişim projects can be run in Turkish, English, Arabic and Russian; the knowledge base for each language should be checked separately.
Does a chatbot completely replace the customer representative?
No. The bot takes on repetitive questions, while complaints and matters requiring authority are passed to a representative. The representative then devotes their time to these conversations.
Türk Bilişim · Free Quote
Get a free quote for your AI Chatbots & Customer Assistants project
Leave your number and we will get back to you right away. The call is free.
Your request is saved on this device
We can't reach our server right now. Your request will be sent automatically once the connection is back. To talk right away:

Share this article
Comments
Comments
No comments yet.
















