The question managing directors put to us most is not “what can AI do?” but “where do we start without wasting money or losing control of our data?”. The answer barely changes: with one specific task that comes up every day, a success measure written down beforehand, and rules the team understands. Apply supports small and mid-sized companies, accounting firms, law practices, clinics, hotels and online shops from the first decision through to everyday use. Together we establish what is worth doing, try it on cases from your own archive, set up an environment where data stays under your control, and teach staff to trust the output only as far as it has earned. All of it remotely, by video call and remote access.
The services are designed to build on one another, yet none is a prerequisite for another. If you already know what you want to automate, start with adoption. If AI is still at the talking stage in your company, strategy is the cheaper first move.
An AI use policy, a list of the tools colleagues already rely on, and idea cards scored for value and risk. For leaders who want to decide on evidence rather than trend articles.
A single task, such as turning enquiries into quotes, gets a draft produced by a model and approved by a person. The result lands in Pantheon, Minimax, Metakocka or Outlook, not in yet another app.
Search over your internal files with every answer sourced, inside your own EU cloud tenancy. SharePoint permissions carry over to the AI, and spending has a ceiling. No hardware purchase required.
For firms that bought the licences but never saw the habit take hold. Online workshops built on each department's real tasks, plus a simple rule on which data never gets typed in anywhere.
What they share: lots of repetition, text or a document as input, mistakes that are easy to spot, and a person who checks the result before it has consequences. If your task lacks these traits, you will hear so in the very first conversation.
No deadline stays buried in the inbox. An accounting firm in Ljubljana gets a hundred client emails a day: invoices, payroll questions, letters from FURS. The model identifies the client and the type of matter, pulls out any deadline and routes the message to the right accountant. A human still writes the reply.
Specifications come from the catalogue, never from the model. A children's goods shop in Celje wants to sell on Amazon in Germany and Italy. AI turns the Slovenian descriptions and feature table into German and Italian drafts, which a fluent speaker reviews before anything is published.
Every statement points to its source. On each project an architecture studio in Kranj wades through utility providers' conditions and municipal spatial plans. AI lists the requirements that shape the design, citing document and page, and the lead architect verifies them.
Management still makes the calls. A 60-room hotel collects several hundred reviews a month on booking sites, in five languages. AI groups them by theme, say road noise, breakfast or parking, and shows how often each comes up month by month.
Mismatches are flagged, not silently fixed. A warehouse in the Koper hinterland receives delivery notes and consignment papers in all sorts of layouts. The model reads sender, pallet count and order reference, checks them against the advance notice in the WMS and shows any gap to the warehouse operative.
Three questions before every trial. One: will the text contain personal data, and if so, can it be stripped out before processing? Two: who is the processor, where does it process the data, and do we have the data processing agreement GDPR requires? Three: is the risk to individuals high enough to call for an impact assessment? The answers go into your record of processing activities. We draw on the Information Commissioner's guidance on AI, which broadly stresses that a controller must know where data goes and why. Your data protection officer or legal adviser makes the legal judgement; we provide the technical answers.
Each stage ends with a go or no-go, so the most a failed attempt can cost is known in advance. We work over video calls and remote access, Monday to Friday, 9:00-17:00.
With the head of the department we pick one task and measure what it costs today in time and errors. From the archive we collect about fifty real cases together with the answers your people gave back then.
The model processes the archived cases and we compare its output with what actually happened. The pass mark is set beforehand, for example “pallet count right on 45 of 50 delivery notes”. Miss it, and we stop here.
For a few weeks the model handles new cases in parallel while the team carries on the old way. Two or three colleagues compare the drafts with their own work and note what they accepted, edited or threw away.
The saving is calculated from the logged cases, not from impressions. On that basis you choose a roll-out with a guide and GDPR records, another round of improvements, or a clean stop that costs far less than a year's subscription to unused software.
The Artificial Intelligence Act (EU AI Act) is an EU regulation, so it applies in Slovenia directly, with no need for a separate national law to bring it in. What it asks of you depends on what you use an AI system for and on whether you develop and sell it or simply use it in your work. The first step never changes: know which tools run in your company, for what purpose and with which data. We build that overview with you; your legal adviser makes the legal assessment.
Three safeguards. The model only gets jobs where an error is quick to spot, such as reading fields from a document, not working out a price. Answers drawn from your files always cite their source so they can be checked. And until the figures prove the quality, a person approves every result. Errors do not disappear, but they become visible and cheap.
Towards customers and authorities, your company is, just as with a mistake by an employee. That is why every project spells out who approves a result before it leaves the building and which cases bypass the model and go straight to a person. For the technical operation of the solution we built, we are responsible to the extent set out in the contract.
No. We handle the integrations, the cloud environment and quality monitoring. From you we need someone who knows the task inside out and can judge whether a draft is right, plus a contact person for access. If you have an IT team, we hand them the documentation at the end so they can maintain the solution themselves.
For writing, summarising and searching your own mail and documents, Copilot is often a good start. Before buying licences, though, we check who can see what, because Copilot uses everything a user can reach, forgotten folders shared with everyone included. For tasks that reach into your ERP, online shop or WMS it is rarely sufficient on its own; that is where adoption with proper integrations comes in.
Our work is charged at a fixed price or €50 per hour plus VAT, always with a cap stated in the quote. Model usage is billed to you directly by the vendor according to the volume of text processed; you get an estimate up front and we track consumption throughout the trial. If a dedicated cloud environment is needed, you pay the cloud provider for it. Ongoing support after go-live can be included in a Start, Business or Premium plan.
Describe the work your team repeats every day and the software it is done in. We will look at real examples and propose a pilot with a measurable goal.
Enquiry received
An engineer will write back no later than the following working day. Faults that leave your staff unable to work are always looked at first.
No town by that name here. Check the letters, or settle for the nearest larger place. Our engineers work only over the internet, so Kočevje and Koper get precisely the same treatment.
