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.