AI Automation in Kazakhstan and Uzbekistan: A 2026 Playbook

Kazakhstan and Uzbekistan are among the most interesting markets for AI automation in 2026. Customers already live in messengers, mobile payments are everywhere, and both governments have made AI a national priority. At the same time, most companies are still at the very beginning: in Uzbekistan only 7.2% used generative AI in the first quarter of 2026 (Gorgona, Fera Tech). That combination, high digital readiness and low adoption, is exactly where early movers gain the most.

This playbook explains what works locally: which channels to build for, how to connect payments and CRM, what Kazakhstan’s new AI law requires, what projects cost and how to run a first pilot.

Why now

• Policy tailwind. Kazakhstan has declared 2026 the year of digitalisation and artificial intelligence (Kazinform). Uzbekistan’s AI strategy targets $1.5 billion by 2030 (Gorgona).
• Full access to leading models. OpenAI and Anthropic officially support Kazakhstan, Uzbekistan, Armenia, Georgia, Kyrgyzstan and Azerbaijan (OpenAI, Anthropic). Companies in the region can build on the same stack as European and US firms, with official contracts and invoices.
• A gap between demand and supply. Local agencies already sell bots and AI agents, but most offers are simple scripted bots. Integrated assistants that take payments, update the CRM and hand over to a person are still rare.

Channels: build where your customers already are

The most common mistake is to copy a playbook from another market. Channel habits differ sharply.

Sources: Fera Tech, Gorgona, Kursiv.

WhatsApp rules matter. Since 15 January 2026, WhatsApp’s Business API no longer allows general-purpose AI assistants. Company bots for their own customers, such as sales, support or booking, remain allowed (TechCrunch). Build on the official API through an approved provider. Unofficial gateways that imitate a personal WhatsApp account risk blocks of the very number your customers know.

Instagram still works for marketing in Kazakhstan and the rest of Central Asia, so direct-message automation remains a valid channel alongside WhatsApp and Telegram. Our guide to AI chat funnels in WhatsApp, Telegram and Instagram DMs covers funnel design in detail.

Five use cases with the fastest payback

1. Sales assistant in WhatsApp or Telegram. Answers product questions from your catalogue, qualifies the lead, sends a Kaspi, Payme or Click payment link, and creates a deal in the CRM. Most value comes from speed: a lead that waits an hour often buys elsewhere. See Speed-to-lead with AI.
2. Order status and support. Order lookups, delivery questions and returns handled in chat with a clean handover to a person for complaints.
3. Booking for clinics, beauty and services. Appointment booking, reminders and rescheduling in the messenger the client already uses.
4. Voice receptionist. Answers calls after hours in Russian and Kazakh or Uzbek, books appointments and sends a confirmation message. Some commercial speech engines already list Kazakh (kk-KZ) and Uzbek (uz-UZ) models, but published quality metrics are scarce, so test on real calls with real background noise before launch. The general approach is in our AI receptionist guide.
5. Internal knowledge assistant. Answers staff questions from regulations, price lists and 1C instructions, which is valuable in growing companies with high staff turnover.

Anatomy of a messenger sales assistant

What does a well-built assistant actually do? Here is the flow for a typical retailer or service business in Almaty or Tashkent:

1. Greeting with disclosure. “Hello, this is the AI assistant of [brand]. I can help you choose, check availability and pay. A manager is available if you need one.”
2. Language detection. The assistant answers in the language the customer writes in: Russian, Kazakh or Uzbek, including mixed messages.
3. Need discovery. Two or three short questions, not a form: what the customer needs, for when, budget if relevant.
4. Answer from the catalogue. Prices, availability and delivery terms come from your live catalogue or 1C, never from the model’s memory.
5. Offer and payment. The assistant sends a Kaspi, Payme or Click payment link or invoice for the chosen items.
6. CRM update. A deal is created or updated with contact details, the conversation summary and the payment status.
7. Handover. Complaints, unusual requests, large orders and low-confidence answers go to a manager with the full context, so the customer does not repeat themselves.
8. Follow-up. Reminders for unpaid orders and a short satisfaction question after delivery, within the messaging rules of the platform.

Language strategy

Language is a commercial decision, not a translation task. In Kazakhstan most business communication runs in Russian, but a brand that answers properly in Kazakh earns trust, especially outside the largest cities. In Uzbekistan, Uzbek is used in Latin and Cyrillic script and many customers switch between Uzbek and Russian within one conversation.

Practical rules:

• Let the customer choose the language implicitly by writing; do not force a menu.
• Keep product names, prices and legal texts identical across languages by storing them once and translating templates, not improvising.
• Uzbek and Kazakh are lower-resource languages for AI models. Test answers with native speakers on your real conversations and keep a list of terms the model must use.
• For voice, test speech recognition separately per language and per typical background noise, such as a car, a shop or a street.

The technical stack

• Models: official OpenAI or Anthropic APIs, with a cheaper model for routine classification and a stronger one for complex answers.
• Channels: the official WhatsApp Business API through an approved provider, the Telegram Bot API, and an Instagram integration where relevant.
• Business systems: the CRM you already use, such as amoCRM, Bitrix24 or HubSpot, plus 1C for stock and prices where needed.
• Payments: Kaspi in Kazakhstan; Payme, Click or Uzum in Uzbekistan, with payment status written back to the CRM.
• Logging and quality: every conversation stored with the model’s answer, sources and the human handover reason, reviewed weekly against a test set of real questions.

Metrics to track from day one: first response time, share of conversations handled without a manager, conversion from chat to payment, average order value in chat, and the reasons for handover.

The law: what Kazakhstan’s AI act requires

Kazakhstan’s law “On Artificial Intelligence” No. 230-VIII has applied since 18 January 2026 (Kursiv, EY). For business automation, four points matter:

• Inform users that a service is provided using AI. In practice: a short line at the start of every chat and call.
• Label synthetic content, such as AI-generated images, videos and voices in marketing.
• No emotion recognition without consent. Do not analyse customer or employee emotions in calls unless people have explicitly agreed.
• Fines range from 15 to 100 MRP, that is 64,875–432,500 tenge.

In Uzbekistan we did not find an equivalent disclosure statute in our research. We still recommend the same practice: it builds trust and prepares you for regulation that is clearly coming. For content-labelling details, see AI-generated content: copyright, likeness and labelling rules.

Personal data. Both countries regulate personal data, and localisation requirements in the region differ and change. Before choosing where to store chat logs and call recordings, get a short opinion from local counsel. It costs far less than migrating a live system.

Choosing a partner: sanctions and due diligence

Many businesses in the region work with partners from the EU. EU sanctions (Article 5n of Regulation 833/2014) prohibit EU companies and residents from providing IT consulting, software development, advertising and market research services to legal entities established in Russia, including indirectly through subsidiaries in third countries when the benefit goes to a Russian parent (European Commission FAQ). Companies in Kazakhstan and Uzbekistan are not affected as such, but an EU vendor will run a short ownership check before signing. That is normal and protects both sides.

For the same reason, design your stack on officially available services: official OpenAI or Anthropic APIs, the official WhatsApp Business API, and hosting you can contract for directly.

What it costs locally

Sources: FlowAI, OSN.kz, ИИ.kz, ai-bot.kz, Fera Tech.

Note the structure: almost every offer is a one-off setup plus monthly support plus usage such as messages, minutes or tokens. Ask for usage costs as a separate, transparent line. Across CIS markets, first-year operating costs typically reach 30–60% of the development cost, so compare offers over twelve months, not by the setup price.

A 30-day pilot plan

1. Week 1: baseline and scope. Measure response time, share of unanswered messages, conversion from chat to payment. Pick one channel and one scenario.
2. Week 1: content. Collect the catalogue, price list, FAQ and 100–200 real customer conversations in all languages you serve.
3. Week 2: build. Assistant on the official API, disclosure message, payment link, CRM deal creation, handover to a person when confidence is low or the customer asks.
4. Week 3: soft launch. Run on part of the traffic or during off-hours; review every escalated conversation daily.
5. Week 4: measure and decide. Compare with the baseline. Expand, adjust or stop, with numbers rather than impressions.

Typical mistakes

• Copying another market’s channel mix. Kazakhstan is WhatsApp-first, Uzbekistan Telegram-first. Build for one first, then add the other.
• Russian only. Customers notice when a brand ignores Kazakh or Uzbek. At minimum, the assistant should understand both and answer in the customer’s language.
• Bot without payment. If the customer must leave the chat to pay, much of the benefit disappears. Integrate Kaspi, Payme, Click or Uzum from day one.
• No disclosure. In Kazakhstan it is a legal requirement; elsewhere it is a trust issue.
• Unofficial WhatsApp gateways. Cheap at first, expensive when the number is blocked.
• Pilots without an end date or metric. Agree on the KPI, duration and decision rule before you start.

Conclusion

Kazakhstan and Uzbekistan combine messenger-first customers, widespread mobile payments and official access to the world’s leading AI models. The companies that win in 2026 build assistants that live in WhatsApp or Telegram, speak the customer’s language, take payment in the chat and update the CRM, while respecting Kazakhstan’s disclosure and labelling rules.

We design messenger funnels through our Growth team and build integrated AI agents with payments and CRM through Agents. Book a free 15-minute audit and we will map your first scenario, channel and metric for a 30-day pilot.