Glossary Explorer

Comprehensive glossary of AI and commerce terminology to help you navigate the evolving language of artificial intelligence.

Browse all glossary terms

Filters:
General AI

Natural Language Generation (NLG)

Natural language generation (NLG) is the subfield of artificial intelligence concerned with producing coherent, contextually appropriate human-readable text from structured data, internal representations, or other non-linguistic inputs. NLG systems take inputs such as database records, numerical summaries, knowledge graph facts, or semantic representations and generate written or spoken language that communicates that content in a natural, fluent form. Modern large language models have made high-quality NLG widely accessible, enabling generation of everything from product descriptions and financial summaries to personalized email drafts and API documentation. In commerce and enterprise AI, NLG delivers value at scale by automating content that would otherwise require significant human writing effort. Retailers use NLG to generate unique, SEO-optimized product descriptions for catalogs with hundreds of thousands of SKUs—a task impossible to complete manually at that volume. Business intelligence platforms use NLG to translate dashboards and query results into narrative summaries that non-technical stakeholders can act on. Personalization engines use NLG to craft individualized marketing messages framed in natural prose. The primary risks of production NLG are factual inaccuracy (hallucination) and brand voice inconsistency, both of which require human review processes and quality guardrails when the output represents the organization to customers.

General AI

Natural Language Processing (NLP)

Natural language processing (NLP) is the branch of artificial intelligence and computational linguistics concerned with enabling computers to understand, interpret, manipulate, and generate human language. NLP encompasses a wide range of tasks—including tokenization, parsing, named entity recognition, sentiment analysis, machine translation, summarization, and question answering—each addressing a different dimension of bridging the gap between the ambiguity of human language and the precision required for computational processing. Modern NLP is dominated by transformer-based neural architectures trained on large corpora, which have dramatically improved performance across virtually all language tasks. NLP is the foundational technology underlying the majority of AI applications that interact with customers or process textual data in commerce and enterprise environments. Search engines, customer service chatbots, product review analysis systems, contract review tools, voice assistants, and email triage systems all rely on NLP capabilities to function. For commerce specifically, NLP enables understanding of unstructured customer feedback at scale, extraction of product attributes from supplier data sheets, and intelligent search that interprets shopper intent rather than matching keywords literally. As language models grow more capable, NLP increasingly handles tasks previously requiring specialized domain expertise—making it one of the highest-leverage AI investments for organizations handling large volumes of text.

General AI

Natural Language Understanding (NLU)

Natural language understanding (NLU) is the component of natural language processing focused specifically on machine comprehension of the meaning, intent, and context within human language, rather than its surface-level form or generation. Where NLP is a broad umbrella covering all language tasks, NLU addresses the harder problem of semantic interpretation: determining what a user actually means (intent), identifying the relevant entities and facts in an utterance, resolving ambiguities, and maintaining coherent understanding across multi-turn dialogue. NLU systems produce structured semantic representations—intents, entities, relationships, and context—that downstream systems can act upon. In conversational commerce and enterprise AI deployments, NLU is the capability that determines whether a system actually understands a customer or employee versus merely pattern-matching on keywords. A well-functioning NLU layer allows a virtual assistant to correctly interpret a modification request (not a cancellation), identify the relevant order number from context, and route to the appropriate workflow—even when the user phrases the request in dozens of different ways. NLU quality is the primary determinant of customer satisfaction with AI-powered service channels; failures manifest as frustrating misunderstandings that damage trust and drive customers to human agents. Investing in domain-specific NLU training data and evaluation frameworks is essential for any organization deploying conversational AI at scale.

AI Models & Technology

Neural Interface

A neural interface (also called a brain-computer interface or BCI) is a direct communication pathway between the brain and an external computing system. Neural interfaces capture neural signals — typically via electrodes implanted in or placed on the scalp — and translate them into digital commands, or conversely deliver stimulation back to neural tissue. Research applications span motor rehabilitation, sensory restoration, and neuroscience. While neural interfaces remain largely in research and early clinical stages, their relevance to enterprise AI is emerging as companies like Neuralink and Synchron pursue commercial products. In future commerce contexts, neural interfaces could enable hands-free, thought-driven interaction with digital systems, with significant implications for accessibility and human-machine workflow design. For most enterprise AI practitioners today, neural interfaces are an important technology to monitor on the horizon but are not yet a deployment consideration for commerce applications.

AI Models & Technology

Neural Networks

Neural networks are computational models loosely inspired by the structure of biological neurons, consisting of layers of interconnected nodes (neurons) that transform input data through sequences of weighted operations and nonlinear activation functions. During training, these weights are adjusted via backpropagation to minimize prediction error. Deep neural networks — those with many layers — are the foundation of modern AI, powering image recognition, natural language processing, speech synthesis, and most other AI capabilities in production use today. Neural networks underpin virtually every AI capability deployed in modern commerce platforms: the embedding models that power semantic search, the transformer architectures behind LLMs, the convolutional networks used in image-based product classification, and the reinforcement learning systems used in dynamic pricing. Understanding neural network fundamentals — including concepts like overfitting, training data requirements, and the role of architecture choices — helps business and technical leaders make more informed decisions about AI system design, vendor selection, and realistic capability expectations.

General AI

Non-Linear ROI

Non-linear ROI describes investment return patterns in which the relationship between input (capital, time, effort) and output (value, revenue, efficiency) is not proportional—where returns accelerate, compound, or exhibit threshold effects rather than scaling in a straight line. AI investments frequently exhibit non-linear ROI because AI systems improve with more data (data network effects), AI capabilities compound when integrated across multiple workflows (platform effects), and AI can unlock entirely new capabilities or business models that produce step-change rather than incremental returns. The flip side is that non-linear investments can also deliver zero or negative returns until a critical threshold of data quality, model maturity, or organizational capability is crossed. For business cases involving AI and advanced technology, recognizing non-linear ROI patterns is essential for setting realistic expectations and making sound investment decisions. Early-stage AI deployments often underperform linear projections because data pipelines, integration work, and model training represent sunk costs that precede any return. Once operational, well-designed AI systems can compound value across the organization—a single trained demand-forecasting model improving inventory, procurement, marketing, and supply chain decisions simultaneously. Leaders who evaluate AI investments with linear ROI frameworks systematically undervalue platform-building and overvalue quick-hit automations, leading to portfolios of isolated point solutions that fail to capture the compounding returns that justify transformational AI investment.