This Artificial Intelligence glossary is designed for anyone in the mortgage industry, providing a clear, jargon-free introduction to the key AI terms you’re likely to encounter. It focuses on what these terms mean in a real-world context, not just in theory.
You don’t need to be technical to benefit from AI. By becoming familiar with the language and concepts in this glossary, you’ll be better equipped to have informed conversations with technology providers and feel more confident in how AI could improve your business operations.
Algorithm
A set of rules or instructions followed to analyse data and make decisions, such as assessing borrower risk.
Artificial Intelligence (AI)
Technology that simulates human intelligence to perform tasks like underwriting, customer service, and document processing.
AI Agent
A type of AI that can carry out specific tasks autonomously, such as managing a mortgage case or following up with clients.
Agentic AI
Using technology to perform repetitive tasks like data entry, document collection, and client follow-ups without manual effort.
Generative AI
AI that creates content such as emails, reports, or marketing materials for client communication and outreach.
API (Application Programming Interface)
A connection that allows different software systems (e.g. CRM, lender platforms, sourcing systems) to communicate and share data.
Bias
When an AI system unfairly favours or disadvantages certain people or products because of the data it was trained on or the way it was designed.
Chat bot
An AI-powered tool that interacts with clients via text or chat to answer questions, complete fact-finds, qualify leads, or book appointments.
Context
The background information an AI uses to understand a request properly. In a mortgage setting, this could include the client’s circumstances, the stage of the application, previous communications, and the documents already provided.
Data Privacy
Protecting client and business information so it is handled securely and in line with regulations. Most large public LLMs do not define strict data privacy provisions around where business information resides, and whether that data is used for training future iterations of models.
Document Recognition (OCR)
Optical Character Recognition – technology that converts scanned documents into readable data.
Deterministic Systems (and Probabilistic Systems)
Deterministic systems that always produce the same output for the same input.
- Predictable and consistent
- No randomness involved
- Example: A calculator … 2 + 2 will always equal 4
Probabilistic Systems
Deterministic systems that always produce the same output for the same input.
- Predictable and consistent
- No randomness involved
- Example: A calculator … 2 + 2 will always equal 4
Probabilistic systems that use probability to make decisions or predictions.
- Outputs may vary even with the same input
- Handle uncertainty and ambiguity
- Common in AI and machine learning
- Example: A chatbot generating slightly different responses to the same question
In summary:
- Deterministic = fixed and predictable
- Probabilistic = flexible and based on likelihood
Explainability
How clearly a system can show why it produced a particular result or recommendation.
Fine-Tuning
Further training an AI model so it performs better for a specific business use case or type of data (such as mortgage product or criteria data).
Guardrails
Rules, limits, and controls put in place to help AI behave safely, accurately, and within business or compliance requirements.
Hallucination
When an AI produces incorrect or misleading information that sounds plausible but is not accurate.
Human in the Loop
A process where a person still checks, approves, or overrides AI outputs, especially for important decisions.
Large Language Model (LLM)
An AI model trained on vast amounts of text data to understand and generate human-like language.
Major LLM providers of AI models;
- OpenAI – GPT (used in ChatGPT)
- Google DeepMind – Gemini
- Anthropic – Claude
- Microsoft – Copilot (powered by OpenAI models)
- Meta – Llama
- DeepSeek – DeepSeek
- xAI – Grok
Models may also include smaller LLMs that are not provided by the major LLM providers, and that have been trained to perform a task, such as sourcing mortgage products or identifying lenders’ underwriting criteria.
Machine Learning (ML)
A type of AI that improves over time by learning from past data, used in underwriting and risk assessment.
MCP (Model Context Protocol)
A standard way for AI to connect to other systems, read information, and use tools. It is like an AI-friendly way of linking systems together.
Natural Language Processing (NLP)
AI that understands and processes human language, used in chatbots, email drafting, and voice tools.
Predictive Analytics
Using historical data to forecast future behaviour, such as predicting the next words in a sentence that you are typing.
Parsing
The process of reading information and breaking it into useful parts. For example, AI might parse an email, application form, or uploaded document to identify names, addresses, income figures, loan amounts, and other key details.
Prompt
The input or instruction you give to an AI system to generate a response.
Prompt Engineering
Writing clear instructions or questions so the AI gives more useful and accurate responses.
RAG (Retrieval-Augmented Generation)
A method where AI looks up relevant information from approved documents before producing an answer, helping make responses more accurate and grounded in your own content.
Structured Data
Information already organised in fixed fields, such as names, loan amounts, property values, and dates in a CRM or application form, or criteria in mortgage sourcing.
Tokens
Small units of text that AI reads and processes. A token can be a whole word, part of a word, a number, or even punctuation. AI systems break text into tokens so they can understand and generate language.
The number of tokens affects how much information an AI can handle at one time, how detailed a response can be, and the cost of using the system.
Temperature
A setting that controls how cautious or creative an AI’s responses are. In a mortgage setting, a lower temperature is usually better for factual tasks such as summarising documents, drafting case notes, or answering policy questions, while a higher temperature may be more suitable for creative tasks such as writing marketing copy.
Higher temperature can increase the chance of the AI giving an answer that sounds plausible but is less accurate, because it takes more freedom in how it responds. Lower temperature can reduce that risk, but it does not remove hallucinations completely. AI can still be wrong even when temperature is low.
Training Data
The past information used to teach an AI model. This means the model’s knowledge may not always include the latest lender criteria, product changes, or regulation updates until it has been retrained on newer data.
Major public LLMs do not have complete training data for the UK mortgage industry.
Unstructured data
Information that is not neatly organised, such as emails, call notes, PDFs, or scanned documents.
Visual AI
AI that can understand, analyse or create images and video. It can identify what is shown, extract useful information, and spot patterns or issues.
Voice AI
AI that can transcribe conversations or voice inputs.
Workflow automation
Technology that manages a process, such as a mortgage case – sending reminders, requesting documents, and tracking case progress.