AI we're accountable for — in revenue, not slides. 8 weeks to prod.

Before kick-off we fix in the contract: the metric, measurement method, control group, and KPI bonus threshold. The base fixed fee is paid regardless — it covers the team's work. The KPI bonus is paid only when the agreed result is achieved. If the metric is not reached — we don't invoice the KPI bonus and review the reasons together.

Exact price comes from the discovery call — depends on integration complexity and the chosen module. Scope includes a discovery week, development, A/B test and handoff. We send a written quote within 48 hours of the first call.

We pick a process with two conditions: there is a measurable metric (conversion, support load, time on task) and enough data for training/RAG. Best candidates: catalog search with >5% zero-result rate, product pages with clicks on the recommendations block, support with repeating questions.

Yes. Most of our clients are on Bitrix. We know the catalog, orders, and CRM APIs. Integration via webhooks and REST — we don't touch your backend. Data stays in your infrastructure or in our jurisdiction — your choice.

Not necessarily. We choose models together: open-source (Llama 3, Qwen 2.5, Mistral) are deployed on your or our infrastructure — data stays within the perimeter. API models (Claude, GPT) are used when there are no regulatory restrictions and they're faster or cheaper.

We sign a Data Processing Agreement as part of the contract. Personal data is processed in your jurisdiction or on-prem — with an optional FSTEC-compatible infrastructure option. This is not an upsell — it's included in the base pilot price.

Three options: (1) we hand over code and documentation — your team takes it from there, (2) we move to a retainer — support, new modules, (3) scaling — the next business process using the same approach. No forced lock-in.

During the discovery week — 2–3 meetings of 1 hour each plus access to analytics and logs. During development — one technical contact (developer or analyst) for integration questions: 2–4 hours per week. During rollout — one person for A/B setup and monitoring.

We work with what you have. If there's not enough training data — we use RAG (retrieval over existing content) or synthetics for the starter set. The discovery week is partly for assessing the real state of your data — come with that question.

It depends on the task. For semantic search — Elasticsearch with a ready reranker pipeline (fast and reliable). For PDP personalization — an embedding model with custom ranking logic. Admin copilot — LLM with prompt engineering and RAG over your knowledge base. Open code throughout — no licensing traps.