Explaining how Machine Learning uses Pattern Recognition

Traditional software follows rules that people explicitly define.

For example, an ERP system might contain an if/then rule such as: If an invoice exceeds $50,000, then route it for additional approval.

A developer or configuration expert manually defines this rule for the software system.

Machine learning works differently. Instead of defining every rule manually, we give the system many examples and allow it to identify patterns. Most modern AI systems are built using machine learning.

For example, we might give a machine-learning model historical invoice data, including:

• Invoice amount
• Supplier
• Purchase order
• Payment terms
• Posting date
• Previous exceptions
• Whether the invoice was ultimately identified as unusual

The model may learn that certain combinations of values are more likely to indicate a duplicate, an error, or a possible exception. It can then assign a risk score to a new invoice, even when no single predefined rule has been violated.

At it's core, much of AI is advanced pattern recognition. An AI system examines large amounts of data and identifies relationships that occur repeatedly.

The AI does not necessarily understand the business process in the same way an experienced consultant does. It recognizes statistical relationships in the data.

For example, it may learn that invoices from a particular supplier, submitted without a purchase order and posted near the end of a financial period, have historically required manual correction. When a similar invoice appears, the system can flag it for review.