Flisko
Business process automation

Business process automation with an AI agent. From 4,900 €, paid by phase, your code from day one.

Analysis. Pilot on your data. Production. Each phase stands on its own.

An AI agent takes over one repetitive process, for example incoming invoices, inbox triage or the weekly report, and runs it on your own accounts. You decide where it stops and what a person approves. The analysis costs 490 € and is credited to the agent; an agent for one process costs from 4,900 €.

Free · 30 min · Fixed-price offer after the analysis

Author: ·

Entry
from 4,900 €
Analysis
490 €, credited
Maintenance
from 190 € per month
Ownership
your code from day one
Runs
on your accounts or server
Diagram: an incoming invoice from email to the accounting systemA PDF invoice arrives by email. AI reads it, extracts the supplier, amount and date and prepares the import; a person approves, then the data goes into the accounting system.IN THE BACKGROUNDON YOUR ACCOUNTSaccounting systemPDFemail01Read02Extract03Prepareimport04ApprovesupplieramountdateV OZADJU · ✓
A document arrives → AI reads and prepares → a person approves. Runs in the background.
✦ In short

Who builds AI agents for business processes in Slovenia, and what does it cost?

AI agents for business processes (automation of one or more repetitive processes that runs on the client's own accounts and systems) are built in Slovenia by Flisko (Informacijske storitve, Žiga Flis s.p.), based in Gorica pri Slivnici, for clients in Slovenia, Croatia and Austria and remotely across the EU. The process analysis costs 490 € and is credited to the agent, an AI agent for one process costs from 4,900 €, a multi-process system 12,000–20,000 €, maintenance from 190 € per month. Payment follows the phases: after any phase you can stop and pay only for completed phases. Code, prompts and configuration belong to the client from day one. The agent runs on Claude in the cloud or on an open model on the client's server. Prices are final; the provider is not VAT-registered.

The work is done by Žiga Flis personally, which is why every package has a stated scope limit. Slovenian version: Avtomatizacija poslovnih procesov.

What we automate: nine typical processes

Nine processes where AI automation pays off most in an office, described the way your office sees them: what grinds, what we set up and what changes.

  • 01

    Incoming invoices from email into the accounting system

    ProblemPDF invoices arrive by email and someone re-types the supplier, amounts and line items into the accounting system.SolutionAI reads the attachments, extracts the data and prepares an import file for systems such as Vasco, Pantheon, Minimax or SAOP.Result: No more re-typing. A person reviews and approves instead of typing.
  • 02

    Bank statement reconciliation

    ProblemMatching payments to open invoices takes hours of monotonous line-by-line comparison every month.SolutionAI reads the statement, matches items to open invoices by your rules and flags everything it cannot place reliably.Result: Only exceptions are handled by hand, not every line.
  • 03

    Weekly report from Excel

    ProblemEvery Monday someone assembles the management report from a KPI sheet by hand, always the same way.SolutionA scheduled task on Mondays at 7:00 reads the Excel file, compares it with the previous week, writes down the key findings and builds a PowerPoint from your template.Result: The report is in the inbox before the team arrives.
  • 04

    Inbox triage

    ProblemThe morning starts with 50 or more unread messages.SolutionEvery morning AI sorts mail into urgent, to review and archive, and drafts replies to urgent messages in your tone.Result: Reading mail shrinks to approving drafts.
  • 05

    Tender and regulation monitoring

    ProblemThe official gazette and tenders are checked by hand, or not at all until the deadline has passed.SolutionA weekly scan of sources by your keywords; a summary with deadlines and links arrives by email, in Slack or in a Drive archive.Result: No relevant tender slips by because a notice was missed.
  • 06

    Meeting minutes

    ProblemMinutes are written by hand or not at all, so action items get lost before the next meeting.SolutionAI transcribes the recording, writes structured minutes with decisions, owners and deadlines, and creates the action items as tasks in tools such as Asana, Trello or Notion.Result: Minutes with decisions and tasks are in your tools right after the meeting.
  • 07

    Long contract and PDF analysis

    ProblemA first pass through a long contract takes hours before an expert even gets to the real decisions.SolutionThe whole contract with annexes goes into Claude at once, because its context window holds 1 million tokens. Standard questions flag deviations from your template, risks and disputed terms; comments are written straight into Word.Result: AI does the first pass, the expert checks and decides.
  • 08

    Personalised offers and presentations

    ProblemOne template, twenty clients, and half an hour of manual tailoring for each.SolutionFrom your template and the data in your CRM or an Excel list, AI builds a tailored version for each client: logo, industry, relevant use cases.Result: No more manual tailoring. You review the finished result.
  • 09

    Content from your templates

    ProblemPosts, offers and descriptions take long to write, and the tone varies from person to person.SolutionAI writes from your templates, terminology and style guide, from the first draft to versions for different channels.Result: A first draft in your tone and terminology, consistent across channels.
Tool

An automation blueprint for your process

Pick your sector and describe a repetitive task in two to four sentences (up to 700 characters). Claude suggests three ideas in a few seconds: the first in full, the other two by email. No savings figures; we estimate those on the intro call.

Two to four sentences, no personal data.0/700

Suggestions are written by AI (Claude by Anthropic) and pass an automatic filter before display, not a human review. Your inputs are processed by Anthropic (USA), without your email; see the privacy policy.

Process automation by area: administration, HR, finance, reporting

The same nine processes, grouped the way companies look for them: by the department where the hours go. Each area is set up on its own, so you start with one process, not a whole department.

  • Administrative processes

    Automating administrative processes means an AI agent does the repetitive office work (email triage, offers from a template, tender monitoring, minutes) and a person only reviews and approves. We set it up on your existing accounts (Gmail or Outlook, Drive, CRM). How many hours that means for you, you work out yourself in the savings calculator.
  • HR processes: hiring and leavers

    In HR we automate the steps that repeat with every hire: a first screening of applications by your criteria with a short summary per candidate, the employment contract from your template, an onboarding checklist with tasks in the tools you already use, and for a leaver a checklist for revoking access and returning equipment. The decision on a candidate stays yours; AI prepares the groundwork.
  • Finance and accounting processes

    Finance automation covers incoming invoices, bank statements and payment reminders: AI reads a PDF invoice from email and prepares an import file for Vasco, Pantheon, Minimax or SAOP, matches payments to open invoices and drafts reminders for overdue receivables. Anything it cannot place reliably is flagged for manual review. A person books, AI types.
  • Reporting

    Automated reporting is a scheduled task that reads your data at a set time (Excel, CRM, web analytics), compares it with the previous period and builds a report in your template: PowerPoint, Word or an email with findings. The weekly management report is in the inbox on Monday at 7:00, without anyone assembling it.

By industry: what the agent takes over in your line of work

Eight industries, each with three to four concrete processes: what the agent takes over, what stays with a person and where it stops. Tools are named where companies in Slovenia actually use them.

How it runs: five phases, each with its own close

  1. Intro call

    30 min, free

    We review your processes and pick the one where an agent pays off most. You get an answer on whether to continue and, if not, why not.
  2. Process analysis

    1 week, 490 €

    We map the steps, data sources, integrations and exceptions and decide where the agent works alone and where it stops. You get a written analysis and a fixed-price offer. The 490 € is deducted from the agent's price.
  3. Pilot on your data

    2–3 weeks

    The agent runs on real documents and messages, first in suggestion mode with a person approving. A short weekly report: what it did, where it stopped, what we changed.
  4. Production and handover

    weeks 4 to 5

    The agent works on its own within the agreed scope. You receive the documentation, the repository with code and configuration, and a 90-minute admin onboarding.
  5. Optional maintenance

    from 190 € per month

    Monitoring, model and integration updates, small adjustments and a monthly report. Cancel any time to the end of the current month.

Payment follows the phases: 490 € on delivery of the analysis, half of the agent's price at pilot start, the other half at handover. If you stop after any phase you pay only for completed phases and keep everything produced so far.

2026 pricing: analysis, agent, system, maintenance

Four items, each complete on its own. You start with the analysis for 490 €, credited to the agent; if you stop after it, you keep the document and pay only for the analysis.

01

Process analysis

490 €

one-off, credited to the agent

  • ✦Step-by-step process map with data sources, integrations and exceptions.
  • ✦Decision where the agent works alone and where it stops (human checkpoints).
  • ✦Recommendation: Claude in the cloud or a local model, and why.
  • ✦Fixed-price, fixed-deadline offer for the agent.

Timeline: 1 week from access to sample documents.

Not included: agent build, system integrations.

02Start here

AI agent for one process

from 4,900 €

one-off

  • ✦One process from trigger to result, on your accounts (email, Drive or SharePoint, accounting software via import files or API).
  • ✦Up to two integrations with your systems, one location or one team.
  • ✦Pilot on real data in suggestion mode, then production.
  • ✦Documentation, repository and a 90-minute admin onboarding; 30 days of email support.

Timeline: pilot in 2–3 weeks, production in weeks 4 to 5. The price depends on the number of exceptions and integrations; the exact price is in the analysis.

Not included: hardware, model provider subscriptions and API usage, a third integration, a custom web application.

03

Multi-process system

12,000–20,000 €

one-off

  • ✦Three to five processes sharing one dashboard, rule set and log.
  • ✦Up to six integrations, user roles, department-level permissions.
  • ✦Each process gets its own pilot and handover; you can stop after any of them.
  • ✦Documentation for the admin and the team, onboarding per department.

Timeline: 8–12 weeks, one process at a time.

Not included: hardware, more than six integrations, processes outside the agreed list.

04

Maintenance

from 190 €

per month

  • ✦Monitoring and alerts on failures.
  • ✦Updates to the model, libraries and integrations when interfaces change.
  • ✦Up to two hours of adjustments a month (a new exception, a new template).
  • ✦Monthly report: runs, exceptions, hours of human review.

Response within two working days. The price depends on the number of processes and integrations.

Not included: new processes and new integrations, offered separately.

Package overview
AI agent packages: price, scope and timeline
PackagePriceScopeTimeline
1. Process analysis490 € one-off, credited to the agentOne process: map, checkpoints, model recommendation, offer1 week
2. AI agent for one processfrom 4,900 € one-offOne process, up to two integrations, one location or team, pilot and productionpilot 2–3 weeks, production weeks 4 to 5
3. Multi-process system12,000–20,000 € one-offThree to five processes, up to six integrations, one shared dashboard8–12 weeks
4. Maintenancefrom 190 € per monthMonitoring, updates, up to two hours of adjustments, monthly reportmonthly, cancel to the end of the month

Scope is limited on purpose: Flisko is run by one person, so we take on as many projects as we can deliver within the stated timelines. If a process needs more than two integrations or several locations, the analysis says so, with a price.

We do not promise or estimate savings for you. Enter your own hours and your own share in the savings calculator and get a figure that is your estimate. The pilot then counts actual cases handled and exceptions.

Prices are final; the provider is not VAT-registered.

Local or cloud: an honest rule for choosing

The agent can run on Claude through the business API or on an open model on your server. We set up both; the data decides, not our preference.

Claude in the cloud compared with a local model for a process agent
CriterionClaude in the cloudLocal model
WhenInternal and confidential business data without special categories; you need the best quality in Slovenian and a fast start.Especially sensitive data (health, legally privileged), a contractual ban on third-party processing or a network without internet access.
Where the data goesTo Anthropic through the business API, which does not train on customer data by default.Only your server. Questions and documents never leave the network.
QualityThe largest models, good Slovenian, long documents (context window up to 1 million tokens).Depends on the model and the hardware; weaker at demanding reasoning. The pilot measures the gap on your cases.
CostsNo hardware; you pay API usage per actual run.Hardware (an existing server with a GPU or a dedicated one), no cost per run.
Time to pilot startDays.2–3 weeks more to set up the model.
HybridA local model classifies and extracts the sensitive fields, Claude drafts from pseudonymised data.The same combination; the local part runs with you, Claude only sees pseudonymised data.

Not sure? The private AI page has a 30-second check that tells you from your document types, volume and sensitivity whether a local model pays off. A local deployment is not GDPR compliant by itself; the legal basis, access rights and retention periods are set up together with your data protection officer.

Where the agent works: email, Teams or Slack, WhatsApp, a dashboard

  • Email

    The agent reads a mailbox or folder (Outlook, Gmail), triages, drafts replies and forwards attachments. It sends on its own only where the analysis says so.
  • Teams or Slack

    Questions, approvals and alerts arrive in the channel where the team already works. A person approves with one reply.
  • WhatsApp Business

    Through the official API, where customers really write on WhatsApp (bookings, updates). Requires Meta business verification; we set it up only where it makes sense.
  • Dashboard

    A simple web page on your hosting: what the agent did, where it stopped, what is waiting for a person. No extra subscription.

Guarantees and exit: everything stays yours

  • ✦Ownership from day one: the repository with code, prompts and configuration sits in your organisation (GitHub, GitLab or your own server) already during the pilot, not only after payment.
  • ✦You can stop after any phase and pay only for completed phases; the analysis, the pilot and the documentation stay yours.
  • ✦No vendor lock-in: the agent runs on your accounts and your servers; you can switch the model (Claude, an open model) without rewriting the process.
  • ✦Handover with documentation and admin onboarding, so your team or anyone else can understand and maintain the solution.
  • ✦Maintenance is optional and you can cancel it any time to the end of the current month.
  • ✦30 days of email support after handover are included in the agent's price.
  • ✦Flisko is run by one person. That is why everything produced is with you from day one, documented and transferable; if we are not around, the solution keeps running and any developer can take it over.

Frequently asked questions about AI agents for business processes

  • What is business process automation?

    Business process automation moves repetitive, rule-based tasks (re-typing invoices, assembling reports, triaging email, preparing documents from templates) from people to software. With AI agents you can also automate tasks that require reading and understanding text (contracts, applications, messages), not only moving data between systems. A person stays wherever a decision is needed.

  • How does AI automation differ from RPA (robotic process automation)?

    RPA mimics clicks and keystrokes in existing programs by rigid rules and breaks when a form or a document layout changes. AI automation reads and understands the document or message, so it copes with different layouts of invoices, contracts and email without reprogramming. For small and medium businesses it is usually cheaper and faster to set up; where RPA is already in place, AI complements it rather than replacing it.

  • How much does an AI agent for a business process cost? 2026 pricing

    The process analysis costs 490 € and is credited to the agent. An AI agent for one process costs from 4,900 € (one process, up to two integrations, one location), a multi-process system 12,000–20,000 €, maintenance from 190 € per month. You get the exact price of the agent in the analysis and pay by phase. Prices are final; the provider is not VAT-registered.

  • How long until a working agent?

    The analysis takes one week, the pilot on your real data 2–3 weeks, and production and handover follow in week four or five. A multi-process system takes 8–12 weeks, one process at a time, so the first one runs before we start the second.

  • Do we need our own IT team?

    No. We set everything up on your existing accounts and hand it over with documentation a non-technical team can follow. You do not need a developer to keep it running, and after handover there are 30 days of email support for questions.

  • What if our tools have no API?

    AI agents do not necessarily need an API, because they also work through files (PDF, Excel and CSV exports and imports) and directly on the desktop. Where integrations exist we use connectors for tools such as Gmail, Drive, Slack or a CRM; with accounting systems such as Vasco, Pantheon, Minimax or SAOP we exchange data through import files.

  • What about security and GDPR?

    On business plans (Team, Enterprise, API) Anthropic does not train models on customer data by default, traffic is encrypted and operation is aligned with GDPR. Every automation runs on your own accounts. Where internal rules require more control, we design the solution with open models on your own infrastructure. If you wish, we review the system before launch as part of a security audit. More about the security audit.

  • What if we stop after the analysis or the pilot?

    You pay only for completed phases. After the analysis you keep the written process map and the offer (490 €); after the pilot you keep the code, configuration and reports, because the repository sits in your organisation from the start. Continuing is not a condition for any of it.

  • Where does the agent run and who owns the code?

    On your accounts and, if you choose, on your own server. Code, prompts, templates and configuration are yours from day one; Flisko holds no licences and hosts nothing in your name. You choose the model in the analysis: Claude through the business API or an open model running locally.

  • Do you promise hours or money saved?

    No. Savings depend on your process and on the share the agent can take over, so you enter them yourself in the savings calculator and get your own estimate. The pilot then counts cases handled, exceptions and hours of human review; those are the numbers you decide on production with.

From practice

In May 2026 we ran a free 90-minute AI workshop for local entrepreneurs at the AUREA business incubator in Gorica pri Slivnici, hosted by the Kozjansko Regional Development Agency. We solved four cases live: replying to a wedding cake inquiry, comparing two suppliers' offers, planning a marketing campaign from last year's data and a weekly report from Excel.

So far we have run more than 20 hours of workshops for accounting and law firms and for marketing and sales teams. The automations that typically grow out of them, and how Claude fits a small European business, are covered in Claude AI for European SMBs in 2026.

According to the Slovenian statistical office (SURS), in 2025 72 % of large and only 18 % of small Slovenian companies used artificial intelligence. Business process automation with AI is one of the fastest ways for a small company to close that gap, without its own IT team or month-long projects.

Sources

  1. SURS: Digital entrepreneurship, detailed data, 2025: published 4 Dec 2025; 18 % of small, 31 % of medium and 72 % of large companies use artificial intelligence.
  2. Anthropic: Context windows: which Claude models have a 1 million token context window.
  3. Anthropic Privacy Center: Is my data used for model training?: default rules for business plans (Team, Enterprise, API).

Related services

✦ Next step

One process. 490 € for the analysis, credited to the agent.

On a 30-minute call we pick the process. Within a week you get a written analysis and a fixed-price offer, or an honest answer that an agent is not (yet) worth it.

Rezervirajte brezplačen 30-minutni klic