Українська

AI

AI for business by industry: one five-question frame, a can / with conditions / cannot matrix for eight types of business, and tool prices per person

Search for 'AI for business' and you get the same list every time: write texts, answer customers, analyse data. It is identical for a beauty salon and a law office, although in one of them half the list is prohibited by law.

Updated 2026-10-08 8 min Figures carry a check date
Illustration generated with AI (OpenAI gpt-image-1) and labelled with IPTC metadata.

Why generic advice fails

Take any article titled "10 ways to use AI in your business" and replace the word "business" with "dental practice". Half the advice becomes dangerous: "upload your customer base and ask it to segment" in a clinic means handing medical data to a third-party service. Replace it with "law office" and "let AI answer client questions in chat" runs into attorney-client privilege. Replace it with "coffee shop" and "analyse your contracts" is simply beside the point.

So this series is built differently. There is one frame of five questions, the same for everyone. And there is a separate breakdown for each industry, where the answers differ. This page is the hub: the frame, tool prices and a summary matrix that shows where industries part ways. Each industry gets its own detailed article; links appear here as they are published.

The five-question frame

Every article in the series answers the same five questions, in this order.

  1. What. Five to seven tasks that genuinely eat hours in this industry: not "write texts" but "answers to 40 identical questions a day about booking" or "a 120-line estimate that gets redone three times".
  2. Where. What to do it with: a chat assistant, AI already built into your CRM, till or booking service, or automation that connects services to each other.
  3. How much. Monthly cost for a business of this size, with the date the prices were checked.
  4. What not to do. Industry-specific limits: medical and personal data, attorney-client privilege, fiscal documents. The general filter is in the article on what not to automate.
  5. A first step for one week. One task to start with and a way to measure the result.

The quality test for each article is simple: if you swap the industry name for another and the text stays true, the article has failed.

Three tools, not a hundred services

Behind the signboards of hundreds of "AI for business" services there are three things.

A chat assistant. ChatGPT, Claude or Gemini in a browser or app. A person sets a task and gets a text, a table or an analysis. We compared the three assistants on five Ukrainian business tasks.

AI built into a service you already use. Reply suggestions in a CRM, an automatic answer to a review on the maps, a product description generated in the shop admin panel. You do not pick a model and do not pay separately: the feature is either in your plan or it is not.

Automation. Services such as Make or n8n that link events: a new order creates a CRM card, an email to the customer and a row in a spreadsheet, with an AI model in the middle writing the text or sorting enquiries. We priced 1,000 and 10,000 operations in the article on automation without a programmer.

A customer-facing chatbot stands apart: it is also either an assistant or an automation, but with risks of its own, so it has its own article.

Which of the three tools a given task needs
  1. Is the task done by a person each time, and slightly different each time?A letter to a difficult client, a contract review, next week's menu plan. Yes → chat assistant
  2. Does the task live inside one service you already use?A reply to a review on the maps, a product description in the admin panel, a call summary in the CRM. Yes → first check whether the feature is already in your plan
  3. Is the task identical every day and does it pull data from two or more systems?Website order into the CRM plus a message to the manager; form submission into a spreadsheet plus an email to the client. Yes → automation

What it costs per person

Prices from the providers' pricing pages on 8 October 2026, converted to hryvnias at the National Bank of Ukraine rate for that day: 44.86 hryvnias per dollar and 50.16 per euro. All amounts exclude taxes the provider may add at checkout.

ToolPlanPer monthIn hryvniasCondition
ChatGPTPlus (individual)$20≈ 897 UAHMonthly billing
ChatGPTBusiness (per user)$25 · $20 billed annually≈ 1,122 · 897 UAHFrom 2 users; no training on your business data by default
ClaudePro (individual)$20 · $17 billed annually≈ 897 · 763 UAHMonthly or annual billing
ClaudeTeam, standard seat$25 · $20 billed annually≈ 1,122 · 897 UAH2 to 150 users; no training on your content by default
GeminiGoogle AI Pro (individual)909.99 UAH909.99 UAHUkrainian price in hryvnias; a cheaper AI Plus at 224.99 UAH has lower limits
GeminiGemini Enterprise, Business editionfrom $21 per seat≈ 942 UAHUp to 300 seats
n8n (automation)Starter, cloud€20 billed annually≈ 1,003 UAH2,500 workflow executions a month
Make (automation)Core$9 billed annually≈ 404 UAH10,000 credits a month, 1 credit = 1 module action

The main conclusion from the table: the individual plans of the three assistants cost practically the same, about 900 hryvnias a month, and the business plans per person run from 900 to 1,100. Choosing an assistant on price makes no sense. Choosing on what it does with your data does: the business plans of ChatGPT and Claude do not train the model on your content by default, while on individual plans you have to switch that off by hand. Where exactly is in the article on data for an AI assistant.

For a team of five on a business plan with monthly billing that is 5 × $25 = $125, or about 5,600 hryvnias a month. That figure should be compared not with zero but with the hours those five people spend on the tasks under "What".

The matrix by industry

Three columns are the three most common uses. A cell shows not "can AI do it" but "may you do it in this industry, and on what condition". The conditions and prohibitions come from the provisions we cite below and from the rules for handling data.

What can be handed to AI in eight types of business
Online shop✓ product descriptions with manual review against marketplace rules✓ a knowledge-base bot with handover to a person≈ orders without passport data or full card numbers
Coffee shop, restaurant✓ menus, dish descriptions, replies to reviews on the maps≈ table bookings yes; complaints of feeling unwell after a meal, a person only≈ a name and phone number for a booking, nothing more
Beauty salon, barbershop✓ social posts and replies from the stylist's template✓ booking and visit reminders✗ allergies, skin condition, pregnancy are health data
Medical or dental clinic≈ website texts without real patient cases≈ schedule and booking only; the bot does not answer about symptoms or medicines✗ medical information is a special category of personal data
Law or accounting firm≈ document templates; every provision checked against the law's text≈ booking a consultation; no answers on the merits of a case✗ attorney-client privilege; accounting clients' data only anonymised on a business plan
Construction and renovation✓ commercial proposals and letters to the client✓ initial questions about timing and stages≈ the spreadsheet does the estimate, AI only writes the text; floor plans without a link to the owner
Online school, courses✓ course programme, assignments, notes≈ support from course materials; grades are given by a person≈ student work without names; children's data never goes into a chat
Real estate agency✓ property descriptions from the facts of the listing✓ initial handling of enquiries: area, budget, timing✗ passport data and title documents

✓ allowed; ≈ allowed on the condition in the cell; ✗ do not hand to AI. Anonymised means no names, company names, amounts or details by which a party could be identified.

Three provisions stand behind the crosses. Medical information, together with health data, is a special category of personal data under Article 7 of the Law on Personal Data Protection; processing it needs a separate legal basis, and "it was convenient to paste into a chat" is not one. Attorney-client privilege under Article 22 of the Law on the Bar and Practice of Law covers any information about a client and the content of advice, and the duty to keep it binds everyone who works with the lawyer, not only the lawyer. Fiscal documents for any trade are produced only by a registered fiscal register under Law No. 265/95-VR, so "let AI issue the receipt" is not an AI task at all.

Notice that the "Texts and descriptions" column is green almost everywhere, and the "Customer data" column almost nowhere. That is the main difference between industries: not what AI can do, but what it may be shown.

A first step for one week

Do not "introduce AI in the company". Take one ticked cell in your industry's row of the matrix and turn it into a one-week test.

  1. Pick a ticked task that someone on the team does every day: answers to routine questions, descriptions, letters.
  2. Measure the week before. How many minutes a day the task takes now. Without that measurement you will not be able to tell an impression from a result.
  3. A week with the assistant. Same person, same task, an individual plan at 900 hryvnias or the free tier. Record the minutes and the number of times the output had to be redone.
  4. Do the sums. If the saving is under an hour a week, or more than one result in five has to be redone, the task is not suitable. Take the next cell, not the next service.

An illustrative example for scale: a salon receptionist answers 30 messages a day at 3 minutes each, 90 minutes a day. If template answers with an assistant cut that to 40 minutes, a month of 22 working days gives 22 × 50 = 1,100 minutes, a little over 18 hours. Against a 900-hryvnia subscription that is a concrete conversation. If the saving turns out to be 10 minutes a day, that is 3.7 hours a month, and a different conversation.

Our position

Choose a task, not a tool. The three assistants cost the same and can do roughly the same; the difference between success and disappointment is a well-chosen first task and an honest before-and-after measurement. For industries with crosses in the third column the order is reversed: data rules first, and only then the first task.

The position does not apply if you already have a CRM, till or booking service with built-in AI that nobody uses. Then the first step is not an assistant subscription but one week with the feature you are already paying for.

Prices checked on 8 October 2026 on the pricing pages of OpenAI, Anthropic, Google, n8n and Make; exchange rate from the National Bank of Ukraine for 8 October 2026; legal provisions from the texts of the laws on zakon.rada.gov.ua. The salon example is illustrative; substitute your own minutes.

Still not sure this applies to you?

Describe the situation. The first reply says whether this is our work and how long the diagnostic takes.

Request a diagnostic