Asking «how much does AI cost» is like asking «how much does a car cost» — it depends on what you actually plan to do. Let us break down what really drives the price and what the ranges look like in Kazakhstan today.
What the cost is made of
The price of implementation is almost never about «the neural network itself». A model subscription is small change in the overall budget. The real money goes into three things.
1. Preparing the data
AI answers from your materials: price lists, policies, catalog, conversation history. If all of that lives in a manager's head and in WhatsApp threads, it first has to be collected into a usable form. This is usually the most underestimated part of a project.
2. Integrations
An assistant with no access to your systems can only chat. Value appears when it writes to the CRM, checks stock, creates a task. Every connection to an external system is separate work, and the older the system, the more expensive it gets.
3. Scenarios and testing
You need to define what the assistant does in each situation, when it must call a human, how it behaves with an angry customer or an off-topic question. Then run it against real requests rather than imagined ones.
Market ranges in Kazakhstan
The spread is wide, and that is normal — the same word «bot» covers very different things.
- $750–2,000 — a simple scripted bot or an assistant over a ready knowledge base with one integration
- $2,000–7,500 — an assistant with several integrations, request triage and handover to a manager
- from $7,500 — complex rollouts in companies with several systems, custom logic and data security requirements
On top of that there is almost always a monthly part: model usage, hosting and support. For smaller rollouts this is typically tens of dollars a month.
What makes a project more expensive
- Data in disarray: no current price list, policies contradicting each other
- Legacy systems without an API — workarounds have to be invented
- A requirement to «understand everything» instead of a defined set of scenarios
- No decision-maker on the client side
How not to overpay
Start with one process. Not «roll out AI across the company» but «the assistant answers questions about stock and price on WhatsApp». A narrow task is easier to quote, faster to launch and immediately shows whether the approach works.
Demand a measurable result. How many requests close without a manager, how many minutes are saved per request, how many enquiries stopped being lost overnight. If a contractor cannot name a metric, there will be nothing to measure the effect with.
Check who owns the result. The knowledge base, the scenarios, account access — all of it should stay with you, otherwise changing contractors turns into a second rollout from scratch.
Where to start
Look at where most working hours go. Usually it is answering the same customer questions and moving data between systems by hand. Take one such process, work out what it costs you monthly in salary — and compare that with the price of automating it.
We do this review for free: we look at your processes and say honestly where AI will pay off and where ordinary tools are enough. Message us on WhatsApp and we will price your task.