AI Ház

AI implementation for companies

The first step in introducing AI differs from company to company. Some organisations need to start by organising their data, others by selecting a provider and setting rules for its use. Where AI is already part of daily work, the next task is building more complex workflows.

The three starting points are described below. In practice they often build on one another, but an introduction can begin with any of them. Governance and training are part of the work in all three cases.

1Data organisation and data maturity

I'm interested

2Provider selection and governance

I'm interested

3Building more complex AI workflows

I'm interested

1. Data organisation and data maturity

The role of data

An AI solution can only work with the data it has access to, and its results cannot be more accurate than its input. If product data exists in several spreadsheets in different versions, or documents are stored as scanned images across different folders, a solution built on that data will not be reliable either.

In such cases the first phase of the introduction is a data organisation task. The work done here also forms the basis for later automation and analysis.

Data assessment and organisation

The data assessment records what data exists where, who is responsible for it, and in what format and quality it is available.

The data is then organised. We remove duplicates, standardise names, digitise paper and scanned documents, and designate for each data domain the system that serves as the authoritative source.

Data maturity

Data maturity describes how orderly, reliable and accessible an organisation's data management is. At a low level, data is kept in personal spreadsheets and e-mails; at a higher level, in central systems with designated owners and access rules.

After assessing the current level, we prepare a scheduled plan for the next steps. Organising all data is not a precondition for the first AI project; it is sufficient to start with the data domain the first workflow relies on.

2. Provider selection and governance

Assessing current AI use

In most organisations, AI use begins without a management decision. Some employees use free chatbots to draft e-mails, translate or summarise documents, often with personal accounts and without the company's knowledge.

A growing share of the software already in use also includes AI features. Office applications, CRM systems, accounting software and meeting recorders often process company data by default.

The introduction therefore begins with an assessment of who uses which tools, for which tasks and with which data.

Provider selection

The range of AI models and tools changes quickly, and provider offers are difficult to compare. The same task can be solved at low cost and good quality with one model, and at higher cost or with less accuracy with another.

Beyond the model's capabilities, the decision must take into account the data handling terms, namely where the data is stored and whether it is used for model training. The pricing model, compatibility with existing systems and the effort required to change provider later also matter.

Cloud and on-premise operation

Cloud services can be introduced quickly, require no hardware of your own, and offer the most advanced models. Their cost is proportional to usage.

With on-premise operation the model runs on the company's own server, so the data does not leave the organisation. This involves a hardware investment and operating tasks, but costs are more predictable and stricter data protection requirements can be met.

In many cases a combination of the two is the most practical, with sensitive data processed on-premise and other tasks in the cloud.

3. Building more complex AI workflows

Process mapping and AI workflows

AI delivers measurable results when it operates as part of a defined process. To that end, we first map how the work is done today, where the data comes from, who performs which step, and where waiting time or repetitive manual work occurs.

On this basis we design the workflow, defining which steps the AI performs, which the existing systems handle, and where human approval is required.

Rule-based automation

Not every automation requires artificial intelligence. Many processes can be automated reliably and at low cost with simple rules and conventional integration.

If a task can be described by clear rules, using AI only adds cost and uncertainty, so in such cases we do not recommend it.

Cost analysis

The fees for AI services usually depend on usage: the volume of data processed, the chosen model and the number of requests. In a test this amount is negligible, but in production at high volume it can become a significant cost item.

Before the introduction we therefore calculate the expected monthly cost from the real data volume and compare it with the expected savings. The decision is then based on figures, and the monthly invoice holds no surprises.

Implementation steps

  1. 01

    Assessment

    A review of current AI use, processes and data.

  2. 02

    Setting the direction

    Deciding whether to begin with data organisation, provider selection or process development.

  3. 03

    Plan and costs

    A scheduled implementation plan with expected costs and savings.

  4. 04

    Delivery and training

    Carrying out the plan and preparing employees.

Governance and training

AI policy

An AI policy is an internal regulation that defines which tools employees may use, for which tasks and with which data. It may be short, but it must be unambiguous.

The policy lists the approved tools and accounts, classifies data as public, internal, confidential or personal, and specifies which of these categories may be entered into an AI tool. It also defines who reviews AI-generated content before it leaves the company.

Training

A policy works when employees know it and apply it. The training covers the use of the approved tools, effective prompt writing and the limitations of AI, including the fact that a model can give an incorrect answer with confidence.

Data security is treated as a separate topic. It covers what counts as confidential data, what must not be copied into an AI tool, and how AI-generated content can be verified.

Frequently asked questions

  • In a short data assessment we review where the data is located and how consistent and accessible it is. On that basis we determine which AI solutions can be started immediately and where data organisation is needed first.

Contact

Let us discuss introducing AI

AI implementation for companies | AI Ház