AI development

AI chatbot and AI development
for business automation

AI chatbots, data analysis and AI-powered automation to help your business grow

AI development — MyTarh
Quoted per projectPrice after scoping
Milestone-basedDelivery with your approval
CustomBuilt on your own data and processes
DocumentedDelivered with a user guide

What does an AI chatbot actually do for a business?

An AI chatbot is software that understands a customer's question in their own words — formal, casual or even with typos — and answers it based on the information you have given it. Its real value for a business is answering repeat questions at any hour of the day or night: opening hours, shipping and return terms, the difference between two products, the documents needed to sign up, or the steps of a service. These are the same questions that take up your support team's time today, and their answers are usually already written down somewhere.

But a chatbot cannot do everything, and it is better to know this before you order one. Language models sometimes produce a wrong answer with complete confidence, especially when the answer is not in your information. Decisions with legal, financial or medical weight, handling complaints, negotiating prices and any situation that needs real empathy are not a chatbot's job. A good chatbot knows its limits: where it is not sure, it says it does not know and hands the conversation to a person.

At MyTarh, AI chatbot development starts with a list of your customers' real questions, not with choosing a model. First we define which questions the chatbot should answer, where the answers come from, which systems it connects to, and where it hands the work over to your team. Only then do we choose between a rule-based chatbot, an LLM-based chatbot and an assistant connected to your business data. Data analysis, image processing and predictive models are also part of this service and are estimated in a similar way, but the rest of this page focuses on chatbots.

Six decisions to make before building an AI chatbot

The quality of a chatbot depends more on these decisions than on the name of the language model. In the requirements session, we go through each one with you:

Answer source: your own business documents

What will the chatbot answer from? Price lists, warranty terms, instructions, FAQs and archives of past support replies are the most common sources. With retrieval-augmented generation (RAG), the chatbot finds the relevant part of your documents before each answer and responds based on it, so when a document changes, the answer changes too. If two documents contradict each other, the chatbot will give contradictory answers as well; that is why organizing the documents is part of the project itself.

Answer boundaries

It must be stated clearly what the chatbot does not talk about: prices that are not final yet, medical or legal advice, comparisons with competitors, or topics unrelated to your business. A chatbot without boundaries will sooner or later give an answer you are not willing to stand behind.

Handover to a human

When does the conversation reach your team? When the user asks for it, when the answer is not found in the documents, or when the topic is a complaint or a payment. It must be clear how this handover happens — creating a request, messaging a team member, or taking a phone number for a call back — and what the user is told outside working hours. The team member who takes over should see the earlier conversation so the customer does not have to explain the problem from the start.

Data and privacy

What information from the conversation is sent to the language model service, where and for how long are conversations stored, and who can access them? Information such as ID numbers or card numbers should not be passed around in the chat without a reason. If your data is sensitive, running the model on your own server is also an option, with its own infrastructure and cost.

Conversation channel

A widget on your website, a Telegram bot or WhatsApp. Each channel has its own limits and display format. It is best to start with the channel where your customers already ask questions, and add the next channel once the answers are working well.

Measuring answer quality

Before launch, the chatbot is tested with a list of your customers' real questions and you approve the answers yourself. After launch, regular reviews of conversations show where answers were wrong or incomplete and which document needs more detail. The questions the chatbot could not find answers to are the best list for completing your documents.

Why MyTarh?

Key features

Multilingual AI chatbot

A chatbot with natural language processing that supports your customers 24/7 in their language

Smart data analysis

Extracting valuable insights from your business data with ML

Image processing

Object, face and text recognition, and automated quality control

Forecasting

Models that forecast sales, demand and customer behavior

Smart automation

Automating repetitive processes with machine learning

API integration

Connecting to ChatGPT, Claude, Gemini and custom models

How it works

Our process

1

Needs analysis

Identifying the processes that benefit most from AI

2

Model selection

Choosing the AI model that best fits your budget and needs

3

Development and training

Implementing and fine-tuning the model with your data

4

Integration

Connecting to your existing systems

5

Training and support

Team training and ongoing model improvement

Pricing

Choose a package

Pick the package that fits your business

Basic

A simple chatbot or AI tool

$2,900
  • Basic chatbot
  • Website integration
  • Support for your customers' language
  • 3 months of support
  • Google or OpenAI API
  • Custom fine-tuning
  • Data analysis
  • Multiple platforms
Get started

Enterprise

A complete enterprise AI solution

Custom quote
  • Dedicated custom model
  • Image and voice processing
  • Complete ML pipeline
  • ERP/CRM integration
  • 12 months of support
  • Dedicated ML team
  • Guaranteed SLA
Get a consultation

Rule-based, LLM-based or data-connected assistant: which chatbot fits your business?

All three are called a “chatbot”, but their behavior, running costs and error rates are not the same. This table helps you see which one you are looking for:

CriterionRule-based chatbotLLM-based chatbotAssistant connected to your business data
How it answersPicks from pre-written answers and menusGenerates the answer with a language model, based on its general instructionsSearches your documents or systems before answering and responds based on them
Understanding free-form questionsWeak; only recognizes defined keywords and buttonsGood; understands casual language and different phrasingsGood; like an LLM, plus familiarity with your business's own terms
Risk of wrong answersLow; but has no answer for questions outside its scenariosHighest; may invent details that do not exist in your businessLower, because it relies on documents, but not zero; depends on document quality
Live information such as order statusOnly if a scenario is written for each case and connected to the systemNone; only knows what is in its instructionsPossible through an API connection to your order, booking or CRM system
Running costLow; no language model usage costAPI usage cost in proportion to the number and length of conversationsAPI usage cost, plus maintaining the connections and keeping documents up to date
Maintenance after launchEvery new question or case must be added to the scenarios manuallyReviewing the instructions and a sample of conversationsKeeping documents up to date and checking questions whose answers were not found
Best forSimple menus, appointment booking or a limited set of fixed questionsGeneral answers or a pilot version where exact detail is not criticalBusinesses where the right answer depends on their own documents and data

These three approaches can be combined; for example, rule-based buttons for fixed questions and a language model for free-form ones. API usage costs are separate from the build fee and are paid directly to the model provider.

What do the cost and timeline of AI chatbot development depend on?

For this service we do not quote a final price before reviewing the scope of work, because two chatbots that look the same can involve very different amounts of work. The package starting points, however, are clear: the Basic package at $2,900 for a simple chatbot or AI tool on your website, the Advanced package at $6,900 for a complete solution with custom fine-tuning, data analysis, an analytics dashboard and multiple channels, and the Enterprise package at a custom quote for a custom model and integration with ERP or CRM.

A few specific factors move the price within each package. First, the state of your documents: if your FAQs, terms and price lists are written down, organized and up to date, the work moves fast; if your business knowledge lives in your team's heads or is scattered across messages, part of the project goes into collecting and structuring that knowledge. Second, the number of integrations: every system the chatbot has to read from or write to is separate work, and if that system has no API, the work grows. Third, the number of channels and how complex the handover-to-human logic is. Fourth, where the model runs: using an API is cheaper and faster to set up, while running the model on your own server gives you more control over your data but needs its own hardware and maintenance.

In terms of time, a simple chatbot takes 10–15 days and an advanced solution with custom fine-tuning takes 30–60 days. Delivery is phased: first you test the pilot version with real questions and approve it, then it goes live. Include API usage costs in your budget from the start; they are separate from the build fee, rise and fall with conversation volume, and are paid directly to the provider. In the consultation session, we also discuss them based on the volume of your questions.

What do you receive with a chatbot project?

This list is the minimum delivery for the project; items included only in one package are marked as such. Anything more is written separately into the contract.

Analysis and preparation

  • A list of questions and answer scenarios, approved by you
  • Defining the answer sources and structuring the documents you provide
  • Guidelines for the chatbot's tone and answer boundaries
  • Defining the points where the conversation is handed to a human team member

Build and integration

  • A chatbot in your customers' language, connected to the model chosen to fit your budget and needs
  • Integration with your website
  • Telegram and WhatsApp connection in the Advanced package
  • Conversation analytics dashboard in the Advanced package
  • Testing with real questions and your approval before launch

Handover and support

  • User guide and project documentation
  • Training your team to review conversations and keep documents up to date
  • 3, 6 or 12 months of support, depending on the package
  • An API account in your own name, with usage costs paid directly to the provider
FAQ

Frequently asked questions

The chatbot uses large language models (LLMs) such as GPT-4 or Claude to understand user questions and give accurate answers. It can be trained on your business information so it gives expert answers.

Yes. Recent models such as GPT-4 and Claude understand many languages well, including Arabic. We can also train a custom model on data in your language.

It depends on usage volume. For a typical small-business chatbot, the monthly API cost is usually modest and is paid directly to the provider; we estimate it for your expected usage before you start.

Before building, we decide which data is sent to the language model and make sure sensitive information, such as account numbers or national ID numbers, does not pass through the conversation. If no data at all may leave your server, we can use on-premise models so everything runs on your own servers.

A simple chatbot is ready in 10–15 days. For advanced solutions with custom fine-tuning, it takes 30–60 days.

Yes; this is the most common kind of business chatbot. Documents such as your product list, terms of service and FAQs are given to the chatbot, and before each answer it finds the relevant part of them. Answer quality depends directly on the quality of these documents; outdated or contradictory documents produce outdated or contradictory answers.

For answering informational questions, connecting to documents is usually the better choice, because when a document changes the answers update immediately. Fine-tuning is more useful for keeping tone and answer format consistent or for repetitive specialist tasks, and is offered in the Advanced package. During needs analysis, we determine whether your project needs one, the other or both.

The chatbot is set up so that when it cannot find the answer in your sources, it says it does not know instead of guessing, and offers a way to reach your staff. Even so, language model errors cannot be reduced to zero, so sensitive topics are kept out of its scope and conversations are reviewed regularly.

No, and it should not. The chatbot handles repetitive questions and out-of-hours conversations so your staff have time for complaints, exceptions and customers who need more explanation. That is why a handover path to a human is part of the design from the start.

A list of the questions customers ask most, the documents that contain the answers, and someone from your team to approve the answers. If the chatbot needs to connect to an ordering, booking or CRM system, technical access or API documentation for that system is also needed. Not everything has to be complete on day one; part of it is organized together during needs analysis.

Only if it is connected through an API to the system that holds this information. In that case, the user's identity must also be verified so one customer's data is never shown to another. ERP and CRM integration is included in the Enterprise package, and simpler integrations are quoted separately.

For a chatbot that answers from documents, yes; when you update a document, the answers update too, and your team learns how in the training session and the user guide. Changes to integrations, channels or the human handover logic are technical work and are done during the package's support period.

It depends on how many repetitive questions you answer each day. If questions are few and simple, a good FAQ page may be enough, and we will recommend exactly that. If a large part of your time goes to answering repetitive questions, you can start with the Basic package on your website and decide on expanding it after seeing real conversations.

Ready to get started?

Talk to our team today and get your project moving