AI chatbot development for business AI chatbot and assistant

An assistant that answers customers from your price list, policies and documents — in Telegram, Instagram and on your website, in Uzbek and Russian. It takes the phone number, creates a task for a manager in your CRM and hands a person anything the knowledge base cannot answer. Built on OpenAI models, with quality measured on your real questions before launch.

Price
from $3,500
Timeline
from 3 weeks

Price and deadline are fixed in the contract before work starts

Oculus. Concrete, light

What is included

  • 01Discovery: where the assistant pays off and what counts as a correct answer
  • 02A knowledge base built from your price list, policies and documents
  • 03Answers with a link to the source: no source, the question goes to a manager
  • 04One assistant across Telegram, Instagram and your website
  • 05Uzbek and Russian: it replies in the customer’s language
  • 06Hand-off of hard conversations to a manager, contact and task in the CRM
  • 07Quality control on a set of your real questions
  • 08Leak protection: the assistant can access only what it needs to answer

Who it is for

  • Businesses whose managers answer the same questions about price, timing and stock all day
  • Companies that lose leads at night and at weekends when nobody is in the inbox
  • Teams with a lot of documents: policies, contracts, an internal knowledge base
  • Owners who want to see where AI pays off before investing in a full agent

Three levels and prices

Pick the level that fits the task and budget. The exact scope is assembled in the configurator and fixed in the contract.

Audit and prototype

Where AI pays off, a prototype on your data

$3,500–5,000

≈ UZS 42M · from 3 weeks

Assistant

Answers in Telegram, Instagram and on your site, writes to CRM

$6,000–15,000

≈ UZS 72M · from 5 weeks

Agent

Acts in CRM and 1C on its own, speaks by voice

$15,000–40,000

≈ UZS 180M · from 10 weeks

UZS amounts are indicative, at the Central Bank of Uzbekistan rate on 14.11.2026. Contracts are in USD, or in UZS at the rate on the invoice date.

In detail

What drives the price

The price is not about “plugging in a chatbot”. It comes from three things: the volume and state of your data, the number of channels, and what the assistant is allowed to do on its own. The Audit and prototype tier shows, on your own documents, where AI pays off and how accurately it answers. The Assistant tier goes live in Telegram, Instagram and on your website, writes to your CRM and passes quality control on real questions. The Agent tier books appointments and creates tasks in your CRM and 1C by itself, handles voice and adds conversation analytics. Document recognition, translation, style tuning and keeping the knowledge base and chat history on your own server are separate lines in the configurator.

How the work runs, week by week

Week one is discovery: we collect real customer questions from chats and calls, agree on what a correct answer looks like and assemble the knowledge base. Next comes a prototype on OpenAI models: we run your real questions through it and show, topic by topic, where it is right and where it fails. At the Assistant tier we connect the channels and the CRM, define when a conversation goes to a person and launch on part of the traffic first. You see a working version every two weeks. After launch we review what people ask most and where the assistant missed, and extend the knowledge base — cheaper and more reliable than switching models.

Uzbek, Russian and the channels that matter

Customers in Uzbekistan write in a mix of languages: Uzbek in Latin and Cyrillic script, Russian, transliteration and shorthand. The assistant replies in the language it was addressed in, and we test quality separately on Uzbek and Russian questions taken from your own chats, because a combined score easily hides the weaker language. The main channels here are Telegram and Instagram direct messages, so the assistant works in both from one knowledge base. Leads go to amoCRM, Bitrix24 or your own CRM. At the Agent tier the assistant checks data in 1C and books appointments itself, understands voice messages and calls in Uzbek and Russian, and sends the manager a daily summary in Telegram.

Risks and common mistakes

The worst mistake is launching a bot that confidently invents prices and deadlines. We build answers only from your documents and require a source: if the base has no answer, the assistant says so and calls a manager. The second is a stale knowledge base: prices change and the bot keeps selling at the old ones. So we agree upfront who updates the base and how, and rerun the control questions after every update. The third is ignoring the cost of model requests: we calculate the cost of one conversation at the prototype stage and show a forecast for your volume as a separate line. The fourth is excess access: the assistant sees only what it needs to answer.

What the contract says

2%

for every day we are late

If we are late, your invoice drops by 2% a day, up to 10% of the contract. So we only name deadlines we can keep.

1

company per niche per city

For a year we do not take on your competitor in your city. What we build for you does not end up next door.

10%

refund after an outside review

Bring your own engineer. They check the code against the contract. If a contracted item is missing, we refund 10%.

Questions

How much does an AI chatbot for business cost?

From $3,500, with a timeline from 3 weeks. That is the Audit and prototype tier: discovery and a working prototype on your data with measured quality. An assistant across channels with CRM logging, and an agent that acts in your CRM and 1C, cost more; their prices are in the tier table. The exact amount is fixed in the contract before work starts.

How long does implementation take?

A prototype takes from 3 weeks. The Assistant and Agent tiers take longer; the timeline of each is shown in the table. You see a working version every two weeks, and for every day of delay the invoice drops by 2%, up to 10% of the contract.

What if the AI gets it wrong and misleads a customer?

The assistant answers only from your knowledge base and cites the source. If there is no answer, it says so and passes the conversation to a manager. We measure quality on your real questions before launch and after every change to the base, and where a mistake is costly — discounts, refunds, medical questions — a person confirms the answer.

Where is our data stored, and could it end up training someone else’s model?

The assistant runs on OpenAI models through the business API, and those requests are not used to train models. Only the fragment of the knowledge base needed for an answer is sent, and internal data the answer does not need is out of the assistant’s reach. The knowledge base and conversation history can live on your own server — a separate line in the configurator.

Can the assistant work in Instagram and with our CRM?

Yes. One assistant answers in Telegram, Instagram direct messages and your website chat, and the customer’s number, the conversation summary and a task for the manager land in your CRM: amoCRM, Bitrix24 or your own. If you have no CRM yet, leads arrive in a Telegram work chat.

What does the assistant cost to run after launch?

Requests to OpenAI models are paid by actual usage. We calculate the cost of one conversation at the prototype stage, show a forecast for your volume and bring it down by matching the model to the task and caching. Bugs in our code are fixed free of charge with no time limit; new scenarios and channels are priced in the configurator.

Other solutions

Practice: AI & Data