How to Start AI Automation: A 90-Day Plan for SMBs

Most small business owners in Tashkent don’t need a five-year AI strategy. They need to stop losing customers who message on Telegram at 11pm and get an answer twelve hours later. If you’re wondering how to start AI automation without hiring a team or burning a season’s profit on software you’ll never fully use, the answer is a staged 90-day plan, not a single big-bang project.

This guide breaks that plan into three 30-day phases: fix the obvious leak, connect the systems you already use, and let AI handle the repetitive judgment calls. Each phase ends with something live and measurable — not a slide deck.

---

Why 90 days, and not a “one big AI project”

Businesses that try to automate everything at once — bot, CRM, inventory sync, voice agent — in one sprint tend to stall around week six. Requirements shift, the team resists a tool nobody explained to them, and the budget runs out before anything ships. A phased plan avoids this by shipping something usable every 30 days, so you get a return before you spend the next tranche.

It also matches how AI agents for business in Uzbekistan actually get adopted successfully: incrementally, starting with the highest-friction customer touchpoint and expanding outward only once that piece is proven.

Days 1–30: Fix the channel your customers already use

Telegram is the channel. With roughly 27 million users in Uzbekistan — about 76% of internet users and 95% reach — it is where your customers already are, whether you’ve built anything there or not. WhatsApp and Viber are minor by comparison.

Phase one priorities:

• Stand up a Telegram bot that answers FAQs, order status, and working hours instantly
• Route anything the bot can’t answer to a human within a defined SLA (e.g., 15 minutes during business hours)
• Log every conversation so you have real data on what customers actually ask
• Add a simple order or booking flow if you’re a retailer, café, or service business

If you’re unsure whether you need a fully custom AI agent or a simpler rules-based bot at this stage, it’s worth reading AI agent vs chatbot: what does an Uzbek business actually need before committing budget. Many businesses start with a scripted bot and only add AI once volume justifies it.

A simple Telegram bot at this stage typically runs 500,000–1,500,000 so’m; an ordering-plus-payment bot with Payme or Click integration runs 1,500,000–5,000,000 so’m. Give yourself the full 30 days to also collect a month of real conversation data — you’ll need it for phase two.

Days 31–60: Connect payments and your CRM

Once the bot is live, the second bottleneck is almost always payments and record-keeping. Payme (10M+ users) is the most-used option in Uzbekistan, followed by Click, with Uzum Bank growing fast; together with UzCard/Humo card rails they cover the large majority of local transactions. Libraries like PayTechUZ exist specifically to unify integration across these providers, which shortens this phase considerably.

At the same time, connect the bot to wherever your sales data actually lives:

Tool	What it’s for	Typical fit
amoCRM	Lead and pipeline tracking	Most common CRM choice locally
Bitrix24	CRM + internal ops	Larger teams, multi-department
1C / MoySklad	Inventory & accounting	Retail, wholesale
Sipuni / OnlinePBX	Call tracking & telephony	Businesses still taking phone orders

Businesses that connect a CRM to their customer channels commonly report sales lifts in the range of 30% — not because the CRM sells anything itself, but because leads stop falling through the cracks between a Telegram message and a follow-up call. Budget for this phase in the 1,500,000–5,000,000 so’m band if it’s still bot-plus-payment work, or from 5,000,000 so’m upward if you’re layering in a CRM-connected AI agent that can check stock, log leads, and answer in both Uzbek and Russian.

Days 61–90: Let AI handle judgment calls, not just scripts

This is where a rules-based bot graduates into something closer to an AI agent — able to hold context across a conversation, answer open-ended questions in natural Uzbek or Russian, and make basic decisions (apply a discount rule, escalate a complaint, suggest an upsell) without a human writing every branch of logic.

Typical patterns worth automating in this phase:

• Natural-language customer support that understands intent, not just keywords
• A sales assistant that qualifies leads before a human ever gets involved — market patterns suggest AI-assisted salespeople can be roughly 40% more efficient at this stage
• A voice AI agent for inbound calls, which in some deployments has cut call-centre costs by up to 5x compared to staffing an equivalent human team
• Automated reporting that pulls from your CRM and messages you a daily summary on Telegram

Custom AI agent work at this depth typically starts from 5,000,000 so’m, with ongoing support and monitoring from roughly $50/month. If you want the full cost breakdown by feature before committing, see how much an AI agent costs in Uzbekistan, broken down in so’m.

What “done” looks like at day 90

By the end of the plan you should have: a Telegram bot handling the bulk of routine questions, payments flowing through Payme/Click without manual reconciliation, a CRM that’s the single source of truth for leads, and at least one AI-driven decision point running without daily human input. That’s a real automation stack — not a pilot.

Should you build this in-house or bring in a partner?

If you have an in-house developer with bot and API integration experience, phase one is often doable internally. Phases two and three usually benefit from a partner who has already connected Payme/Click and amoCRM/Bitrix24 dozens of times, since the local payment and CRM quirks aren’t always documented in English. If you’re new to the concept entirely, what an AI agent actually is, explained simply for business owners is a good starting point before you scope phase one.

Fera Tech’s own services cover exactly this staged build — Telegram bots, CRM integration, and custom AI agents — and our past work includes projects shaped like each of the three phases above.

Frequently Asked Questions

Do I need all three phases, or can I stop after phase one? Many small businesses stop after a solid Telegram bot and never move further, and that’s a legitimate outcome if volume doesn’t justify more. The 90-day plan is a ceiling, not a mandate.

What does the government’s AI push mean for my SMB? Since the October 2024 presidential decree and the AI Strategy 2030, Uzbekistan has committed over $50M to AI infrastructure, expanded IT Park’s tax-free zones for AI companies, and set a goal of training 1 million AI specialists. For SMBs, this mostly translates into more local expertise and tooling becoming available — worth watching if you’re planning phase three.

Can this work in Uzbek and Russian at the same time? Yes — modern AI-based bots handle both languages well, and serving customers in whichever language they write in is now standard practice, not a premium feature.

What if I only have budget for one phase this quarter? Start with phase one regardless of your final ambitions. A working Telegram bot pays for itself the fastest and gives you the conversation data that makes phases two and three cheaper to scope correctly.

If you’d like help mapping this plan to your specific business, get in touch and we’ll help you scope a realistic 90-day version.