Without hype or scare stories
Neural networks are talked about in two genres: “AI will replace everyone” and “it’s a toy”. The truth is more boring: neural networks are a tool that cheaply handles routine tasks with text, images and data. For small businesses they provide what used to require hiring: a copywriter, a designer for small stuff, a first-line operator. Let’s break it down by task.
1. Content: text and images
- Text. Product descriptions, social posts, campaign emails, replies to reviews — neural networks (ChatGPT, Claude, YandexGPT, GigaChat) make a draft in minutes. Important: a draft. Publishing without an expert’s edit means faceless text that is visible a mile away and disliked by search engines. The workable scheme: AI writes the structure and a draft, a human adds facts, numbers and living language.
- Images. Article covers, illustrations for social media, banners — generation (Midjourney, gpt-image, Kandinsky) replaces stock photos and a designer on typical tasks. The cost of an image is cents versus hundreds of rubles for stock.
- What AI cannot do: know your case studies, prices and customers. You provide the substance, AI provides the packaging.
2. Support and sales: new-generation chatbots
Old bots annoyed with “press 1”. Bots on language models genuinely answer questions: you train them on your price lists, FAQ and delivery terms — and the bot handles 50–70% of routine enquiries in site chat, Telegram and WhatsApp, around the clock.
- Answers “how much”, “how to get there”, “is it in stock”;
- collects a request and passes it to a manager with details;
- a mandatory rule: an honest “call a human” button — AI must know how to give up.
3. Processing information
- Documents: extract data from invoices and waybills, check a contract against a template, draft a standard document.
- Calls: transcription and summaries of managers’ conversations — a manager reads digests instead of listening to hours of recordings.
- Analytics for “non-techies”: upload a sales table and ask in plain language “which products dipped in March”.
How much it costs
Subscriptions to ready services — ₽1–5k per month per employee. API access for your own integrations — pay by volume: a typical small-business support bot “eats” ₽3–15k per month. Implementing a bot or integration — a one-off ₽50–300k depending on complexity. This is an order of magnitude cheaper than doing the same with people.
Where to lay down straw
- Hallucinations: a model can confidently make things up. Everything that goes to customers (prices, terms, legal) is checked by a human, and the bot’s topics are restricted.
- Personal data: do not upload customer databases to public services; for sensitive data — Russian clouds or local models.
- Sameness: if all content is made by AI without editing, you blend in with thousands of the same. The “AI + expert” combination wins.
Where to start: a month plan
- Write down the routine with text and data that eats hours.
- Take one task (for example, replying to reviews or product descriptions) and for a month do it via a neural network. Count the time saved.
- The next step — a chatbot on the site: a quick visible effect.
- Integrations with the CRM and documents — once the team has warmed up to it.
Takeaway
Neural networks in 2026 are no longer an experiment but a normal working tool: content faster and cheaper, support around the clock, routine automated. Start with one task, count the effect. Want a bot or an AI integration for your process — let’s talk.
