Section · 35 articles
AI for business
What AI actually does in a small business and what it costs

An AI model on your own computer: how much memory it needs, how many months the hardware takes to pay off against a ten-seat subscription, and why pay-per-request adds up differently
The question usually goes like this: ten people pay for a chat assistant every month, so what if we buy one powerful computer and run a model on it? The answer depends on what you compare it with. Against a subscription the hardware pays off in a little over a year. Against pay-per-request on a cheap model it does not. And there is a third reason to run a model in-house that has nothing to do with money.

AI for a corner shop: how to calculate a supplier order, what to do with repricing and what an assistant will not write about beer and cigarettes
A corner shop owner does the same thing every week: looks at the shelves, recalls how last week went and dictates quantities to the supplier. An assistant helps here, but not where it is usually put. A formula calculates the quantity, while the model writes the message, checks price tags after repricing and replies to reviews. It writes nothing about cigarettes, and that is how it should be.

AI in hiring: the job ad, CV screening and the limits of the law, with a table of prohibited requirements in a vacancy and a flow where a person makes the rejection decision
An assistant writes a job ad in a minute and sorts a hundred CVs in an hour. Both actions touch rules that advice on "setting up AI for HR" usually skips: the list of grounds on which you may not discriminate, and the rules for candidates' personal data.

AI for analysing a sales spreadsheet: five questions that have several correct answers in one table, reference figures on a sample order set, and a check rule built on control totals
You export a quarter of orders, upload the file into a chat and ask for September revenue. The answer arrives in seconds, with a table and an explanation. The trouble is that the same table holds three different September revenues, and each one is correct for its own definition.

Products in AI assistant answers: what the platforms' own documentation says a shop controls, what the platform decides, and why "we'll get you into the chatbot" is not yet something a Ukrainian shop can buy
An agency offers to "get your products into ChatGPT" for a fixed monthly fee. The platforms' documentation, read on 6 November 2026, describes a different world: shopping in ChatGPT is live only for users in the US, a direct feed is granted on application with a waitlist, and what a shop in Kyiv can actually do today comes down to Merchant Center, Product markup and a robot in robots.txt.

AI for source documents: what it costs to read 1,000 pages of delivery notes and acts, which mandatory details under Article 9 of Ukraine's Accounting Law it can only check, and where a person must stay in the process
Every month, the bookkeeper at a small company types hundreds of delivery notes and service acts into the accounting system. Document recognition services promise to do it for them. But a source document is not just a picture with numbers: the law requires certain details on it, and the service is not the one responsible when a faulty document is accepted into the books.

Who owns content created with AI: what Article 33 of Ukraine's Copyright Law says, what four services' terms pass to the user, and what to do with a logo, a product photo, site text and ads
A logo in five minutes, product photos without a photographer, descriptions for the whole catalogue. Before putting them on the site, into ads or into a trademark application, it is worth knowing two things: what Ukrainian law says about such work and what the terms of the service you used say about it.

AI transcription of meetings, voice notes and calls: what 10 and 100 hours of audio a month cost across five services, and which of them handle Ukrainian
Lists of the "best transcription services" rarely give the price of an audio hour and almost never say whether a service understands Ukrainian. Both are visible on the providers' own pages, if you open them.

A free AI course for business: the five-module programme mapped onto the tasks of a company of ten, and a plan for the first month after the certificate
There are more announcements of free AI courses than stories about what happened to a business a month after the certificate. This article is about the second part: what in the course programme actually concerns a company of up to ten people, and what to do in the first four weeks so the course does not end up as a line on a CV.

How to check whether AI search sees your website: a ranking of free and paid tools
Tools that “check AI visibility” measure different things. Some assess whether a site is ready for AI search, others look for brand mentions in answers, and others show your own data from Google and Bing. We sorted them by job and ranked each group separately.

AI for a small manufacturer or workshop: how a model turns tender documents into a checklist, where it gets deadlines wrong and why a person signs the bid and the safety instruction
A small workshop making furniture, metal structures or packaging sees a tender for its product on Prozorro and opens forty pages of documents. Half the requirements concern not the product but the bidder: equipment, staff, experience, finances. A model quickly turns these requirements into a checklist. But a bid is a legally significant document, and a mistake in it gets you rejected, not warned.

AI for a farm: what a model does with an application in the State Agrarian Register, how to keep a field log by voice and why the application rate of a product comes only from the label
A farmer spends hours on three kinds of paperwork: reading the terms of a support programme and collecting documents, writing down what was done where during the day, and making sense of a soil test report or a supplier's letter. A model takes over part of this time. But there is one task it must not be given even partly: how much product to apply per hectare.

AI for a hotel or a guesthouse: which language to reply to reviews in, who confirms a booking, how much occupancy must grow after a discount and what not to give a model
A small hotel or a country guesthouse gets reviews in Polish, English, German and Ukrainian, guest emails at one in the morning and questions like 'will it be cheaper in November?'. A model takes over part of this work: it writes replies in the right language, translates the house rules and drafts emails. But the booking, the price and the guest's passport stay with a person, and each of these points has a rule behind it.

AI for a car service: how a customer's voice message becomes a draft work order, what 300 requests cost and what a model should not do
A car service customer rarely writes 'replace the front brake pads'. They send a voice note: 'something knocks at the front when I brake, it's a Skoda, about a hundred and sixty thousand on the clock'. That has to become an entry in the request log, a draft work order and the questions the mechanic will ask at intake. That is a job for a model. A diagnosis and a part number are not.

AI for a real estate agency: what 40 listing descriptions and 600 lead replies cost, why the median beats the average for prices and what you must not change in photos
An agency writes similar descriptions and answers similar questions every day: 'is it still available?', 'what is the price?', 'can I view it?'. That is a good job for a model. But a model cannot set an object's price or decide what is true in a photo.

AI video for ads: the price of a generated second, a budget for six ad variants, when you still need a shoot and what ad platforms require for labelling
The price of generating video has fallen to cents per second, so the question is no longer whether you can afford it. The question is what may be shown generated, what has to be filmed, and how ad platforms will label such a clip.

AI for an online school: the cost of checking homework, how many checks to read by hand, why an AI-text detector is not evidence and where a support bot must stop
For an online school the most tempting task is homework checking. It is also the easiest to damage trust with: one unfair grade or a false accusation of 'copied from AI' costs more than a month of savings.

AI for construction and renovation: an estimate under Article 844 of the Civil Code, a warning letter about extra work, photo reports without other people's faces and the cost for a crew
In a renovation or construction firm the costliest mistakes lie not in the text but in the estimate and in the correspondence about changing it. This is exactly where an assistant most tempts you to hand it work that it does worse than a spreadsheet.

What AI costs for a team of ten: four sets of plans for 10 seats in hryvnias and the point where pay-per-request beats a subscription
The price of one seat is known to everyone. For a team of ten the question is different: which set to buy, what it comes to in hryvnias, and whether paying per request is cheaper than paying per seat. We took the plans from the official pages and did the arithmetic.

AI for a law or accounting firm: attorney-client privilege, client data, checking legal norms and document templates
An assistant quickly writes a letter, a contract or an explanation. The trouble starts where the conditions of a case, a client's bank statement or a payroll sheet are pasted into it: for a lawyer that is attorney-client privilege, for an accountant it is other people's personal data and other people's commercial information.

AI in sales: what it costs to transcribe all managers' calls, which metrics to measure before and after, and what the personal data law requires
A head of sales listens to a few calls a week out of hundreds. AI can transcribe all of them, but the decision to launch depends not on the price but on three things: whether the client was told about the recording, where the voice is sent, and what happens next.

AI for a medical or dental clinic: three classes of data, what may go into a chat assistant, and what the advertising law forbids in website texts
In a clinic the same assistant function can be safe or unlawful depending on what was pasted into it. A text about the price of teeth cleaning is marketing. The same text with a patient's surname and an examination result is information the law protects separately.

AI product photos for an online shop: which images you may generate, which label a shopping feed requires, and where that label disappears
Generating a product shot on a white background or in a room now costs a few hryvnias. But Google Merchant Center accepts such an image on two conditions only: it shows the very item being sold, and the file itself says AI made it. The second condition breaks quietly — at the moment your site compresses the image.

AI for a beauty salon or barbershop: what to hand to the booking system, what to the assistant, and where Meta's rules and Ukrainian law draw the line
A beauty salon is usually offered AI for Instagram posts. More money sits elsewhere: in empty slots nobody filled and in clients who came once and never returned. That is a job for the booking system rather than a chat assistant, and you can work out whether it pays off in advance.

AI for a café or restaurant: what to hand to an assistant, and where the allergen law and the Labour Code draw the line
In a café, AI mostly writes posts and dish descriptions. It is more useful where someone is already stuck with a spreadsheet: a recipe card, a shift rota, a sales report from the till. But that is exactly where its mistakes cost money or break the law.

A prompt for a business task: the five parts of an AI instruction, and why a ready-made template gives you only two
Collections like "100 prompts for business" promise results without effort: copy, paste, done. But in their own guides OpenAI, Anthropic and Google describe the same five-part instruction, and three of those parts can only be written by the business itself.

AI for an online shop: what to hand to an assistant, what it costs per month and where Merchant Center's rules draw the line
Product descriptions are the first thing a shop hands to AI, and the least interesting in terms of savings. More time goes on supplier price lists, the same questions in messengers, and refused parcels whose cause nobody counts.

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.

A chatbot for business: how to choose between a button bot, an AI assistant and custom development — and test it before launch
Most conversations about chatbots start with picking a service. It is better to start with the questions your customers actually ask: that decides whether you need AI at all.

AI product descriptions for an online shop: what's fine on the site, what's mandatory in the product feed, and where a mistake costs you impressions
Generating descriptions for a thousand products takes one evening. But Google applies two different sets of rules to those descriptions: one for site pages in Search and one for the product feed in Merchant Center. The first is mostly recommendations; the second has hard requirements and disapproves products.

What data you can give an AI assistant and what you can't: a traffic light for business owners, and where to switch off training on your chats
Someone on the team pastes a supplier contract or a spreadsheet of customer phone numbers into a chat so the assistant can 'quickly go through it'. What happens to that data next depends on the plan, one toggle in the settings and the rating button under the answer.

An AI assistant for a small business: we tested three models on five Ukrainian tasks
Comparisons of AI assistants usually go by feature lists. We gave three models five tasks a small-business owner sets every week and checked the answers against a known correct result.

Automation without a developer: which service to choose and what the same workflow costs in three of them
Automation services get compared by their cheapest plan. That comparison is wrong: each one counts a different unit, and the same workflow costs between $11 and $134 a month.

AI content and Google from October 2026: what actually changed and what site owners should do
Since 1 October, social feeds say Google has "cracked down on AI text". We compared the new documentation pages with their archived versions. Less changed than people claim, but one requirement is now explicit.

AI in business: what not to automate, and how to work it out
Everyone publishes the list of what AI can do. Almost no one publishes the list of where it takes more than it gives — and that is the list savings start from.