We forge AI-powered solutions that free up time to create

Starting prices are on the pricing page — if your task is typical, use them as a reference. We name the exact sum after a short conversation, because tasks that look identical cost differently: “a bot for clients” is either three reply scenarios or thirty with stock access and online payments, and the gap is several-fold. The sum is fixed in a contract before the start.

Simple solutions (a card-bot, a landing page) — a few days. More complex systems (CRM, client portal, several connected automations) — a few weeks. Exact timelines are agreed in the contract before the start, not “sometime within a month”.

Key decisions are agreed before the start and at intermediate stages, so the final result shouldn't be a surprise. If something is off at a stage, we fix it within the agreed scope. Domain, hosting and passwords are registered in your name from day one — you don't depend on us, even if you decide to stop working together.

Yes — automation needs access to your CRM, website or messengers, otherwise there is nothing to automate. We work strictly within the agreed scope and never pass credentials to third parties.

The work is led by the founder of Kuznya — from the first conversation about the task to delivery, without handing you over to account managers. For larger projects (marked as “team” in the price list) we bring in vetted contractors for specific tasks, each with a clear area of responsibility.

Most often — what a person does by hand every day following the same rule: moving requests between systems, assembling a report from several sources, reminding clients, sending documents, handling incoming requests. The rule is simple: if the task can be described as “take it from here, do this, put it there” — it can be automated. If every step needs a human decision — it can't, and automation will only add confusion. Start with one routine, not everything at once.

A chatbot follows a script: a question from the list gets an answer from the list. One step aside and it doesn't understand. An AI assistant answers in its own words based on your materials — price lists, instructions, correspondence history. It handles a question nobody anticipated. The flip side: if something in your materials is inaccurate or outdated, the assistant will repeat it confidently. That's why before launch we work out exactly what it reads, and show the source of every answer.

It depends on where that data goes. Cloud services are convenient and cheap, but data leaves your company — for some businesses that's unacceptable. The alternative is a model on your own server: nothing goes outside, but it needs hardware and costs more. We build both, and we say upfront which layer of data ends up where. If you hold personal client data or trade secrets, the mode is decided before the first line of code, not after.

Yes — most of our clients have no technical team. But one condition is mandatory: we need a person on your side who knows how the process actually works and can answer questions during the build. Not a developer — just the one who does this daily. Without that person we build on assumptions, and that is the most expensive way to build the wrong thing.

Mostly the amount of work before the markup, not the markup itself. A cheap site is a ready template with your texts and photos: fast, predictable, and often that's enough. A more expensive one is when we first work out who your client is and why they should contact you, then build the structure around that, write the copy and prepare the pages for search and AI systems. The difference isn't beauty — it's whether enquiries come in. We offer both, and we say plainly when the first one is enough.

If you handle it yourself, domain and hosting come to roughly 20–60 US dollars a year, and that is not our money: you pay the registrar directly and everything is registered in your name. If you would rather not deal with it, our “site under care” plan starts at 59 US dollars a month and already includes hosting, the domain, backups, updates and up to three small fixes a month. But honestly: if the site is simple and nothing on it changes, you can go the first year with no maintenance at all. The downside is that when something does break, an emergency fix costs more than a month of care would have.

With one task — the one done most often and least dependent on human judgement. Not the most painful one: the most painful is usually complex because it is full of exceptions, and it is easy to get stuck on it for months and give up. One working automation in two weeks gives you two things a big plan does not: you see how much time is actually freed up, and the team gets used to a system doing something for them. After that the next ones go faster, because the access is already set up.

A spreadsheet lasts longer than people assume. While you are one or two people with dozens of clients, a CRM adds work rather than removing it. It is worth moving when one of three things appears: several people edit the same record at once and overwrite each other; you cannot say what stage a particular deal is at without asking a colleague; the history with a client is scattered across personal messengers and leaves with the employee. If none of those apply, we will tell you that you do not need a CRM yet.

Yes, and it is worth knowing before you start. The assistant calls a language model, and every call costs money — from a few dollars a month for a light flow of questions to a few dozen for an active one. That sum goes to the model provider, not to us, and is paid from your account. The exact figure cannot be named in advance, because it depends on how many people ask and how often. What we can do is measure the cost of a single call at the prototype stage, so you multiply it by your own volume and see the ceiling.

Most often because the AI crawler sees a different site from the one you see. It takes only the plain text, without images, animation and anything a script draws. If your services are shown as pictures or loaded after the page opens, they do not exist as far as AI is concerned. Then it discards everything a model could have written itself: “individual approach” and “team of professionals” carry no signal at all. What survives is the concrete part: prices, terms, industry, what you do not do. We covered this in detail in our journal and did it on our own site first.