AI automation
Recurring tasks without manual work: typical enquiries, product descriptions, reports from data, routing of requests. Start with a pilot on one process.
AI · Services
AI rollouts that work for your success
There are so many promises around AI that it is hard to tell the tool apart from the advertising. I take a practical approach: I pick one process with you that takes up your time every week, check on it whether the model gives a predictable result, and only then talk to you about a wider rollout. I work remotely with companies in the UK and the EU, among them a chain of tattoo and piercing studios in the UK and education organisations in Belgium.
- A pilot on one process first
- Limits and oversight agreed in advance
- Data processed in the EU by default, exceptions named before you decide
- Built into the website you already have
Services
Two ways to use AI in your business
An automation takes over a task that someone in your business repeats every week in the same form. An assistant answers the everyday questions that would otherwise reach you by email or phone. Both start the same way, with one process and a decision about where the model’s scope ends.
AI chatbots and assistants
An assistant on the website and on WhatsApp that answers only from your company’s knowledge. For unusual requests it hands the conversation, with its history, to a person.
Where to start
One question to find the right AI service
Choose what should run on its own, and you will see the service that fits, with its scope and the steps of a project.
- Task
- Service
What should run on its own?
Limits
What AI takes over and what stays with you
I start with this list because it decides whether a project ends with a tool that is still in use a year later or one that is put aside after a month. It covers four situations in which the decision stays with a person, and I describe them before you get a quote.
-
Unusual requests go to a person
A model copes well with questions that come back in the same form. When a customer writes about several things at once, such as a complaint with an unusual deadline or an order with a request for an exception to the price list, a good answer needs a decision. I do not leave decisions to a model. In cases like these, the matter goes to you, so the model does not improvise an answer.
-
I sort out the process before automating it
If quotes go out late because nobody knows who approves them, an automation will only speed up a disorganised process. So before I connect anything, I ask how the process works today, step by step. Sometimes it is enough to swap the order of two tasks, and AI is no longer needed.
-
Accuracy depends on oversight
A model can write a sentence that sounds certain and is wrong, and no setting can switch that off. You can limit it, though, by restricting the sources to your documents, forcing an “I don’t know” answer when those documents do not cover the question, having a person approve texts before publication and keeping a log that shows every answer. I set up all four safeguards by default.
-
The model gets its knowledge of your company from me
The model knows nothing about your price list, deadlines, delivery terms or the scope of your services until someone provides that information. A large part of any project is organising this knowledge and keeping it up to date. I can lead that work, but I cannot decide for you what your service costs or when you can deliver it.
These four limits shape how I work. The scope is closed, the sources are named, the route to a person is described in advance, and someone checks the results. That keeps the project predictable, because answers come from your materials, unusual matters reach the right person and you can see how the tool works.
My practice
How I work with AI myself
I do not sell a tool I do not use myself. AI is part of my daily work on code, content and analysis, and that is how I know where it helps and where it costs more time than it saves.
-
Code
When I write code, I use AI for repetitive work: building the skeleton of a module, writing tests, searching a large project or spotting a typo in a configuration file. It is not suited to deciding how to build the whole thing, and I do not let it make those decisions. I check every line of code that goes onto a client’s website myself.
-
Content
I write website texts in conversation with the model, but the source material always comes from the client or from my knowledge of their industry. The model helps me organise the content faster, while fact-checking still takes the most time, because I do it by hand.
-
Analysis
In audits and SEO work, AI helps me process data, for example by grouping keywords, finding a recurring pattern in hundreds of URLs or summarising a long report. The numbers come from measurement tools, and the model only puts them into an order from which I can draw conclusions.
I use AI in my own work before I suggest it to anyone else. The first client projects are in preparation. I cannot show you an assistant running at another company yet, because none is live, and I do not claim otherwise. Instead, I have an established way of working and the relevant knowledge. You can see what I have built for businesses so far in the case studies.
Process
Four steps from the first conversation to maintenance
The pilot matters most, because it shows on real data how much work a full rollout would take off your business. It also gives you a basis for deciding the scope.
-
A conversation about the process
My first questions are what repeats in your business and how long it takes. I want numbers, for example how many enquiries arrive each week, how many quotes go out each month and how many hours go on retyping the same information. Without those numbers, nobody can tell after the rollout whether anything has changed. The conversation is just as likely to end with the answer that AI is not needed there.
-
A pilot on one process
You choose one task, and I work on it for a few weeks alongside your current way of working, which stays in place. You see the results as they come in, including the ones that fail. The pilot is a small first commitment that answers the question “does this make sense?” before you invest in a full rollout.
-
Rollout with limits
If the pilot goes well, the solution stays for good. I write down its scope, name the knowledge sources, agree with you when a matter goes to a person and switch on the operations log. It goes into the tools you already use, most often your WordPress website, your online shop or your company inbox.
-
Maintenance and corrections
Prices, deadlines and the scope of your services change over time. Model providers also release updates, after which the same request can return something different from a week earlier. That is why, after the rollout, I review what the model answered, where it was wrong and what is missing from the knowledge base, on a schedule agreed with you.
Related services
How AI connects with my other services
AI solutions build on your website, your shop and the processes you already have, so these projects overlap with my other services.
An automation rarely ends with the model itself. Usually, data has to come from the system you work in, and the result has to go where someone will see it. That is the same work I describe under WordPress plugins and API integrations, with a model in the middle instead of an inventory system. If your website runs on WordPress, the starting point is usually WordPress maintenance. An automation attached to a neglected website stops working at the first larger update, and the blame then goes to “the AI” when the real cause is a missing backup.
There is also the question of visibility. More and more people read an answer put together by a model instead of looking through a list of results. Writing content that has a chance of appearing in such answers is part of my SEO work and is billed there, in monthly plans. I keep the two topics apart because they share little more than the letters AI. The projects on this page concern work inside your business, while SEO concerns how your business looks from the outside.
FAQ
Questions about AI in your business
What clients ask in the first conversation, most often about the cost, the model’s mistakes and where the company’s data ends up.
How much does an AI rollout cost?
I do not have a price list for it, because the cost depends on how much has to be organised before the start and how many systems the solution has to connect to. The project starts with a pilot on one process, which deliberately keeps the entry threshold low. After it, you know whether a full rollout makes sense, and you do not spend money on something that cannot be judged yet. On top of the cost of the work, there is a fee for using the model, billed by usage. I estimate it in the quote and set a spending limit on your account.
Does AI make mistakes?
Yes, and that cannot be switched off. A mistake usually looks like an answer that sounds certain and is wrong, especially when the question goes beyond the material the model was given. That is why every project gets four safeguards: knowledge sources limited to your materials, a forced “I don’t know” answer, a threshold for handing the matter to a person and a log that shows what was answered. I do not promise error-free answers, because nobody can honestly keep that promise.
Where is my company’s data processed?
I choose the model provider so that the data is processed in the European Union. When the type of data and work requires a data processing agreement, or the contract with your company provides for one, I sign it under Art. 28 GDPR, or Art. 28 UK GDPR for companies in the UK. The model receives only what the task needs, without unnecessary customer data, identity documents or payment details. I keep logs of conversations and operations for a period agreed with you and delete them afterwards. If the most suitable tool for a particular case runs outside the EU, I tell you before you decide, together with the consequences.
Where do I start if I do not know what to automate?
With the process that is most tiresome and repeats most often. It usually comes up within the first ten minutes of the conversation, when you mention a task that is “the same thing every day”. If none of your processes is frequent, rule-based and easy to check, I tell you so instead of looking for a reason to send a quote.
Do I need a new website to use AI?
No. I build the solution into what you already have. On a WordPress website or a WooCommerce shop, it runs as a separate plugin that can be switched off with one click. An assistant on a static website needs a single script. A new website only makes sense if the current one has to be replaced for other reasons anyway, and then it is a separate project with its own quote. An assistant on the website or an automation in the shop is no reason to rebuild your website.
What if the AI project does not work out?
Every project starts with one narrow process, and before it starts, I agree with you how to measure the benefit, for example shorter response times to enquiries, fewer repetitive emails or less retyping of data. If that effect has not appeared after the first month, I tell you so and either adjust the scope with you or close the project, instead of expanding a solution that does not help you. Switching it off means deactivating the plugin and leaves no trace on your website.
Are you an AI automation agency?
No. OneNet is a one-person business. I talk to you about the process, run the pilot, build the solution and review its results myself, so you always deal with the person who does the work. Contact is by email, WhatsApp, phone or video call, all in English.
Can you work on AI projects for a UK or Irish business from Poland?
Yes. I work remotely with companies in the UK and the EU, among them a chain of tattoo and piercing studios in the UK and education organisations in Belgium. You deal with me directly, in English. I am available Monday to Friday, 9:00–17:00 Central European Time (8:00–16:00 UK time).