AI and Machine Learning — The Basics in a Nutshell

AI and machine learning (ML) are positioned as major drivers of change for business models and processes, but success depends on choosing appropriate methods, aligning them with strategy, and managing data, privacy, and risk. A clear distinction is made between artificial intelligence (AI) as the broader goal of enabling computers to perform tasks requiring human intelligence, machine learning as systems that learn from data without explicit programming, and deep learning as a neural-network-based subset of ML. The focus is on applied AI, since general AI is described as unavailable and potentially decades away.

Model selection depends on the task and available data. Supervised learning is suited to labeled datasets and is often used to predict future outcomes from historical data, including regression (predicting values) and classification (predicting categories such as true/false). Examples include dynamic pricing based on user attributes and predicting shopping cart abandonment to trigger targeted discounts, alongside an explicit warning that certain inputs (e.g., age-based pricing) can lead to discriminatory outcomes. Unsupervised learning addresses unlabeled data through structuring and clustering, illustrated by segmenting customers into groups for differentiated newsletter content. Related approaches include anomaly detection to identify unusual events such as suspicious login attempts, and recommender systems that propose products based on behavioral patterns, commonly using unsupervised methods.

Data quality and quantity are emphasized as central, with diagnostic and debugging work required to ensure reliability in production. ML pipelines may rely on batch processing, stream processing, or hybrid approaches that must treat both data types consistently. Because many frameworks and cloud services already exist, building from scratch is often unnecessary; however, teams must understand models and match algorithms to business goals. Privacy and security are critical when using personal data, requiring anonymization and privacy-by-design practices aligned with legislation such as GDPR. Execution guidance stresses strategy first, tight collaboration between strategists and technologists, meaningful metrics, and starting early with agile learning supported by solid project and risk management.

• Machine learning algorithms have existed for many years but have only been practical to use for the last 10–20 years due to increased processing power.
• Supervised learning uses labeled datasets for training, cross-validation, and testing.
• Historical data is often used in supervised learning to predict future outcomes.
• Dynamic pricing can be implemented as a regression problem that predicts a product price from user and context features.
• Shopping cart abandonment prediction can be implemented as a classification problem producing a true/false outcome.
• Unsupervised learning can cluster unlabeled customer data into groups for differentiated marketing content.
• Anomaly detection is often based on unsupervised learning techniques.
• Fraud detection can use anomaly detection to identify unusual login attempts from authentication data.
• Recommender systems can be built using supervised learning or, in many cases, unsupervised learning.
• Data processing in ML commonly uses batch processing, stream processing, or a combination of both.

• “AI and ML will be game changers and will strongly influence business models and business processes.”
• “It is essential to choose an appropriate model.”
• “Understand what you want to do and how.”
• “ML depends on data and will create new data.”
• “Don’t forget privacy and information security issues.”
• “Start soon, learn from experience and failure, and always apply solid project and risk management.”
• “AI is not an end in itself.”
• “Avoid project management approaches that are too theoretical.”
• “Don’t try to reinvent the wheel, but understand what existing solutions you could use.”
• “An accuracy of 99.9 % looks good at first glance, but could actually significantly hinder your business processes.”