Free Website Audit: Discover what's holding your digital presence back
What does AI automation actually mean here?
Automation means software does a step a person used to do. Most of that work needs no AI model. A form submits, a record is created, a notification is sent, a report is generated. Rules do this well, and rules can be tested.
AI automation adds a model to the steps where the input is messy: free text, email, documents, images, speech. A model can classify a request, extract fields from an invoice, summarise a thread or draft a reply. It produces a likely answer, not a guaranteed one.
Most real projects mix both. The rules move the data and enforce the process. The model handles the unstructured part. The design question is which steps belong to which, and where a person signs off.
What does AI automation not do?
State this plainly before you budget.
It does not remove the need for a correct process. Automating a broken process makes the same mistakes faster.
It does not guarantee a correct answer. A model produces a likely output. It can be wrong and still sound confident. Every step where a wrong answer costs money needs a check.
It does not remove the need for people. It moves people from doing the step to reviewing it, handling exceptions and improving the rules.
It does not maintain itself. Models change, APIs change, and your process changes. An automation needs monitoring, error handling and a budget after launch.
It does not come with a published saving. Konzept does not publish a percentage of cost saved, a headcount reduction or an accuracy figure, because those depend on your data and your process. Measure your own baseline first.
Rules or a model: how do they compare?
| Decision criterion | Rule-based automation | AI-assisted step |
|---|---|---|
| Cost | Lower to build. No per-request model fee. | Higher to build and test. Adds a running cost per request. |
| Time to launch | Fast, when the rule is clear and the APIs exist. | Slower. Needs test data, evaluation and a review path. |
| Who can edit it | An operations owner, in the workflow tool, if it is documented. | A developer, with the person who owns the process and the examples. |
| Scale limits | Breaks when the input stops matching the rule. | Handles messy input. Limited by cost per request and by review capacity. |
| Lock-in | Tool lock-in. Moving a workflow tool means rebuilding it. | Model and vendor lock-in. Prompts and behaviour do not transfer cleanly. |
| When it is wrong | The input is free text, documents or images that no rule can cover. | The task has one correct answer that a rule already produces reliably. |
Model and tool fees are not published here. They depend on your vendor, your plan and your volume. Check the vendor’s own pricing page.
Choose rule-based automation when
- The input is structured: a form, a record, a file with known fields.
- The decision has one correct answer that you can write as a condition.
- A mistake has a real cost, so you need the result to be repeatable.
- You must explain every decision to a customer, an auditor or a regulator.
- The systems already expose an API or a webhook for the step.
Choose an AI-assisted step when
- The input is free text, email, a scanned document, an image or speech.
- The task is to classify, extract, summarise or draft, not to decide.
- A person will review the output, or the system can flag low confidence for review.
- The volume is high enough that manual handling is the real bottleneck.
- You accept an error rate and you have a plan to measure and reduce it.
What drives the cost?
Five things move an automation budget more than the model choice.
The number of steps and exceptions. A process with three steps and two exceptions is a small project. A process with twelve steps, four systems and fifteen exceptions is not.
The systems. An automation is only as easy as the worst integration. A modern API is cheap to reach. A legacy system with no API, or a process that runs through email attachments, adds real work.
The data. Clean, consistent data with examples shortens a project. Data spread across spreadsheets, inboxes and one person’s memory lengthens it.
The review path. Deciding who checks the output, how errors are reported and how the system learns from them is part of the scope, not an afterthought.
The volume and the risk. A step that runs twice a month rarely justifies a build. A step that runs a thousand times a month, with a clear cost per run, usually does.
What does Konzept charge?
Konzept publishes starting project prices on the pricing page. Growth starts from EUR 14,500, Scale from EUR 29,500 and Enterprise from EUR 44,000. The Enterprise tier lists AI automation and workflows, full-stack software and API architecture. Monthly partnership plans start at EUR 1,200 for Support, EUR 2,350 for Core and EUR 3,450 for Accelerate.
These are published starting prices for Konzept’s work in Bosnia and Herzegovina. They are not a quote, and they do not include model usage, third-party tool subscriptions or hosting. Konzept works from one office, in Sarajevo.
The AI automation service page describes the scope. Where the automation needs a system built around it, that work sits under software development.
What should you do next?
Pick one process. Write the steps, the people, the systems, the monthly volume, the time each run takes and what happens when it goes wrong. Measure the current cost before you change anything. Without a baseline you cannot tell whether the automation worked.
Then split the process into rule steps and judgement steps. Automate the rule steps first. Add a model only where a rule cannot reach, and put a person on the output until the error rate is known.
Before you commit to a build, read custom software vs off-the-shelf; many processes are automated with tools you already own. If your interest is AI as a search channel rather than an internal process, read the AEO guide.
When the process is written down, get a quote. Send the steps, the systems and the volume. We will tell you what is worth automating, what should stay with a person, and what it takes.