Загрузка…
AI is used where it is measurably cheaper or faster than a person: parsing documents, searching an internal knowledge base, triaging incoming requests. Where it is not cheaper, it is not used.
from40 000₽
The total depends on the volume of data, how accurate the result has to be, and whether an external model is allowed or everything has to stay inside your own perimeter: a self-hosted model on your servers costs several times more than an external one
A lower bound, not a final price. We work out the exact sum after breaking the problem down and fix it before the work starts.
A list of work, not a description of an approach. The scope is adjusted to the job: anything unnecessary comes off and anything missing goes on at the estimate stage.
Invoices, delivery notes, requests and enquiries turn into structured data with no manual entry.
An employee asks in plain words and gets an answer with a link to the source document.
The model classifies, assigns an owner and answers the routine ones; anything unusual goes to a person.
The AI lives inside the CRM, the portal or the bot, not in a separate window someone has to remember to open.
If the data cannot leave the company, the model runs on your own servers.
Before rollout we measure accuracy on the company's real documents and show you the number.
Tick the work you need and see the order of magnitude. The estimate goes along with your enquiry.
| Include in estimate | Work | Price |
|---|---|---|
| A pilot on your dataAccuracy measured on real documents before rollout. The decision is then made on a number, not on how the demo felt. | 60 000–850 000₽ | |
| Document parsingInvoices, delivery notes and requests into structured data. | 70 000–1 200 000₽per document type | |
| Knowledge base searchAn answer in plain words, with a link to the source document. | 90 000–1 350 000₽ | |
| First-line handling of enquiriesClassification, assignment of an owner, answers to the routine ones. | 70 000–1 050 000₽ | |
| An agent with access to your systemsThe model does not just answer but acts: creates a request, changes a status, prepares a document. Every action is logged. | 120 000–1 900 000₽ | |
| Embedding into your working systemsThe AI lives inside the CRM, the portal or the bot, not in a separate window someone has to remember to open. | 50 000–750 000₽ | |
| Speech recognitionTranscription of calls and voice messages, with analysis afterwards. | 55 000–850 000₽ | |
| Guardrails and answer checkingThe model does not answer what it does not know and does not invent figures. Anything doubtful goes to a person. | 40 000–600 000₽ | |
| Deployment inside your perimeterThe model runs on the company's servers when the data cannot leave. | 180 000–3 100 000₽ | |
| Fine-tuning on your dataFor when a general model does not reach the accuracy your domain needs. | 120 000–2 200 000₽ |
This is an order-of-magnitude estimate, not a costed quote and not an offer. Bundled together, some work gets cheaper and some gets more expensive depending on the specifics: the exact sum is worked out after we break the problem down, and it is fixed before the work starts.
Verifiable facts instead of claims about ourselves. There is not a single invented case in this catalogue.
UFound is our own AI platform. Full access, no NDAs and no approvals: you can open it and see how it works.
We work with Claude, Gemini and DeepSeek, orchestration through LangGraph, vector search on Qdrant.
There is no client AI case in the public catalogue yet, and we do not invent one. This section stands on a platform of our own that you can go and check.
An ordinary algorithm solves it.
Then we write the algorithm: a language model where an if statement would do costs more, runs slower and is less predictable.
Tell us what is happening now. We reply within one business day, and the first conversation commits you to nothing.