AI glossary for banking and financial services
This glossary is designed to help executives in banking and financial services understand key concepts in Artificial Intelligence (AI), Generative AI (GenAI), Large Language Models (LLMs), agentic systems, and other emerging technologies. Each term includes a clear definition, a summary of its role, and specific use cases in the financial sector.Glossary
Foundational Concepts
Artificial Intelligence (AI)
Definition: Simulation of human intelligence by machines.
Summary: Enables automation and decision-making across banking operations.
Use Case: Real-time fraud detection systems that flag suspicious transactions based on behavioral patterns.
Machine Learning (ML)
Definition: A subset of AI where systems learn from data to improve performance.
Summary: Powers predictive models for risk and customer behavior.
Use Case: ML models assess creditworthiness using transaction history and alternative data.
Large Language Model (LLM)
Definition: The outcome of a type of Machine Learning process intended to understand text and meaning in a variety of situations – typically developed by large software companies.
Summary: Understands and generates human-like language.
Use Case: Summarises client interactions and suggests next steps for relationship managers.
Generative AI (GenAI)
Generative AI
Definition: AI that creates new content (text, multimedia, code).
Summary: Automates content creation for internal and client-facing documents; chats with end users.
Use Case: Drafts personalised investment summaries for wealth management clients.
Hallucination
Definition: When GenAI produces incorrect or fabricated information, often in a convincing way.
Summary: A risk in compliance-sensitive environments.
Use Case: A GenAI-generated compliance report might cite non-existent regulations.
Prompt Engineering
Definition: Crafting effective questions to guide GenAI outputs.
Summary: Ensures accuracy and relevance in generated content.
Use Case: Prompts generate accurate summaries of quarterly earnings calls.
Context Engineering
Definition: Complements Prompt Engineering by providing relevant background information so that Large Language Models can give the optimum answers.
Summary: Ensures accuracy and relevance in generated content.
Use Case: Understand user’s background and level of understanding before providing guidance.
Context Window
Definition: The amount of text an LLM can consider at once.
Summary: Determines how much information the model can process.
Use Case: Enables analysis of lengthy regulatory documents or loan agreements.
Retrieval-Augmented Generation (RAG)
Definition: Use the power of an LLM but on information – typically documents – it has not been trained on, especially internal documentation.
Summary: Synthesise responses based on an underlying – typically private – knowledgebase.
Use Case: Chatbots use RAG to enable users to explore proprietary market insights accumulated by a bank.
Fine-Tuning
Definition: Customising an LLM for specific tasks or industries.
Summary: Improves performance on domain-specific tasks.
Use Case: Fine-tuned models understand internal jargon for compliance automation.
Agentic AI and Agents
Agentic AI
Definition: AI systems that can adapt to changing circumstances, act autonomously toward goals, making use of teams of specialised agents.
Summary: Goes beyond content generation to perform multi-step tasks and reasoning.
Use Case: Automates end-to-end mortgage processing, from application to approval.
Agents
Definition: Autonomous or semi-autonomous entities that perform tasks and may interact with APIs.
Summary: Help to complete workflows, sometimes in collaboration with other agents or humans.
Use Case: An agent handles onboarding by verifying documents, checking KYC, and setting up accounts; a human in the loop may perform a final check.
Comparison with GenAI Assistants
Definition: GenAI Assistants interpret complex instructions and generate content, in response to user interactions.
Summary: Agentic AI is like a digital employee — it takes action, makes decisions, and completes tasks, harnessing GenAI technology.
Analytics and Observability
Predictive Analytics
Definition: Uses historical data to forecast future outcomes.
Summary: Supports proactive decision-making.
Use Case: Predicts loan default risk based on customer behavior and macroeconomic indicators.
User Analytics
Definition: Tracks how users interact with IT systems.
Summary: Improves user experience and system performance.
Use Case: Optimises chatbot flows by identifying drop-off points.
Observability
Definition: Monitoring IT system health and performance, including AI systems.
Summary: Ensures reliability and compliance.
Use Case: Monitors AI-driven trading algorithms for anomalies and uptime.
Governance and Ethics
Bias
Definition: Systematic errors due to skewed training data.
Summary: Can lead to unfair or discriminatory outcomes – or simply wrong decisions being made.
Use Case: A biased credit scoring model may disadvantage certain demographics.
Explainability
Definition: The ability to understand and interpret AI outcomes.
Summary: Critical for trust and regulatory compliance.
Use Case: Regulators require explanations for AI-driven loan denials.
AI Governance
Definition: Frameworks for managing AI risks and ensuring ethical use.
Summary: Includes data privacy, fairness, and accountability but also other areas such as technology selection and approval.
Use Case: A bank establishes an AI governance board to oversee ethical deployment and risk management.