Machine learning is software that learns from data to make predictions or generate content, instead of being programmed rule by rule. According to Google's developer training material, machine learning is the process of training a piece of software, called a model, to produce useful predictions or content from data. By how they learn, ML systems fall into four main types: supervised, unsupervised and reinforcement learning, and generative AI.
Business use in Türkiye is still narrow. According to Eurostat, in 2025, 2.9% of enterprises in Türkiye with 10 or more employees used machine learning for data analysis, compared with 5.1% in the EU. Among enterprises with 250 or more employees, the shares were 12.2% in Türkiye and 25.1% in the EU.
In short:
- Machine learning is a subfield of AI; the system derives the rule from data rather than from a person (EU AI Act, recital 12).
- The most common business uses are prediction (a number) and classification (a category); grouping is also frequent (Google).
- The starting point is not the algorithm but the question itself and the quality of historical data.
What is machine learning and how does it work?
In traditional software, a developer writes the rules explicitly, for example "send for approval if the amount exceeds this limit". In machine learning, the system is given many past examples; the model finds the relationship in those examples itself and applies it when a new input arrives.
Google's material illustrates this with rainfall prediction. The traditional approach builds a physics-based model of the atmosphere and computes huge numbers of equations. In the ML approach, the model is given large amounts of weather data, learns the mathematical relationship between weather patterns and rainfall, and then predicts from current data. In the material's definition, a model is a mathematical relationship derived from data and used to make predictions.
In practice the process has three steps: historical data is collected and prepared, the model is trained on it, and its performance is measured on data it has not seen before. If this last step is skipped, there is no way to know how accurate the model will be in real work.
What is the difference between machine learning and AI?
AI is the umbrella term; machine learning is one of the methods under it. Recital 12 of the EU AI Act (Regulation 2024/1689) describes the key characteristic of AI systems as their capability to "infer" outputs from inputs, and names two families of techniques that enable this: machine learning approaches that learn from data, and logic- and knowledge-based approaches that infer from encoded knowledge or symbolic representation.
This is also the short answer to the frequently searched question of how machine learning differs from symbolic AI: in the symbolic approach, rules and knowledge are encoded by people; in machine learning, the relationship is learned from data. The same recital states that systems based solely on rules defined by people to execute operations automatically are not AI systems.
Deep learning is a subtype of machine learning. In Eurostat's enterprise survey the indicator is defined as "machine learning (e.g. deep learning)", so deep learning is measured as an example of machine learning. Generative AI is also one of the ML types in Google's classification; for details see what is generative AI.
What are the types of machine learning?
According to Google's material, ML systems fall into four categories based on how they learn:
- Supervised learning: the model learns from many examples that contain the correct answers. The material compares this to a student studying old exams that include both questions and answers. Two common uses are regression, which predicts a numeric value (e.g. travel time or the market value of a home), and classification, which predicts which category something belongs to (e.g. whether an email is spam).
- Unsupervised learning: the data has no correct answers; the model looks for meaningful patterns. The most widely known technique is clustering, which groups similar records. According to the material, the clusters are not defined in advance; interpreting and naming them is left to people.
- Reinforcement learning: the model receives rewards or penalties for actions in an environment and develops the strategy that earns the most reward. Examples in the material include teaching robots tasks and software that plays Go.
- Generative AI: creates new content such as text, images, audio or video from user input.
Where do companies use machine learning?
The examples below are practical suggestions showing how the model types described by Google map to business work; they are not based on any sector data:
- Demand and sales forecasting (regression): predicting next period's order volume from past sales, seasonality and campaign data; this feeds stock and production planning.
- Delivery and duration forecasting (regression): predicting a shipment's arrival time or how long a job will take from past records.
- Classification and prioritisation: sorting incoming requests, emails or documents by type; flagging suspicious transactions or faulty records for human review.
- Customer and product grouping (clustering): grouping customers with similar purchasing behaviour or stock items with similar movement; the business unit then interprets the groups.
- Quality and maintenance signals: classifying the likelihood of a failure or defect from sensor or production records.
What these examples share is the need for sufficient and reliable historical data on the same question. A model built before data ownership, definitions and quality are settled will quickly repeat bad data; on this see data governance for AI.
How much do companies in Türkiye use machine learning?
Eurostat's survey on AI in enterprises measures "machine learning (e.g. deep learning) for data analysis" as a separate technology. According to 2025 data, the share of enterprises using it in Türkiye was 2.4% among those with 10-49 employees, 3.7% with 50-249 employees and 12.2% with 250 or more. In the EU, the shares for the same groups were 3.8%, 8.7% and 25.1%.
Compared with a year earlier, the share among enterprises with 10 or more employees rose from 2.3% to 2.9% in Türkiye and from 4.2% to 5.1% in the EU. Among large enterprises in Türkiye, however, it stayed at 12.2% in both years. According to the same survey, the share of enterprises using at least one AI technology in 2025 was 7.4% in Türkiye and 20.0% in the EU. Eurostat's analysis shows that the most frequently used technology in the EU in 2025 was text mining at 11.8%; machine learning also trailed technologies generating images, audio and text.
Where should a machine learning project start?
The questions below are not a sourced standard but a checklist that can be used before a project:
- Is the business question clear? Pick a measurable question such as "what will next month's orders be?" or "is this record faulty?".
- Is there historical data? Supervised learning needs enough examples with known correct answers.
- How will success be measured? Define up front the metric against which the model will be compared with the current method (e.g. manual estimates).
- What happens if it is wrong? If a wrong prediction is costly, the result should be used as a recommendation rather than a decision, with the final say left to a person.
Conclusion: how should a company prepare for machine learning?
Machine learning is the AI method that learns the rule from data; in companies it is used mostly for prediction, classification and grouping. Use in Türkiye lags behind the EU and is concentrated in large enterprises. The right start is a clear business question and reliable historical data. To assess whether machine learning fits your processes, you can reach us through the contact page.
This article is for general information only. Google is a trademark of its respective owner.
#interimmanagement #forwarddeployedengineering