Comprehensive glossary of AI and commerce terminology to help you navigate the evolving language of artificial intelligence.
Glossary Explorer
Browse all glossary terms
Lagging Indicators
Lagging indicators are performance metrics that reflect outcomes already produced by past actions—they confirm trends and results after the fact rather than predicting or driving them. Revenue, net promoter score, annual churn rate, and customer lifetime value are classic lagging indicators: they measure what has already happened and are valuable for assessing whether strategies worked, but they cannot be acted upon in time to change the outcome they describe. Lagging indicators are typically more reliable and objective than predictive measures because they represent completed events. In AI and commerce strategy, over-reliance on lagging indicators creates a reactive posture: by the time revenue decline or increased churn is visible in reporting, the underlying causes have often been entrenched for weeks or months. AI systems can help organizations shift toward leading indicators—early signals like search abandonment rates, add-to-cart-to-purchase ratios, or model confidence score distributions—that predict future lagging outcomes while there is still time to intervene. Understanding the distinction between lagging and leading indicators is essential for designing AI-powered dashboards, setting up early-warning alerting, and evaluating whether AI investments are actually driving business outcomes or simply correlating with them.
Large Language Model (LLM)
A large language model (LLM) is a deep learning model trained on vast amounts of text data to predict and generate natural language. LLMs are built on transformer architectures and contain billions to trillions of parameters that encode statistical patterns about language, knowledge, and reasoning. Prominent examples include GPT-4, Claude, Gemini, and Llama. They can perform a wide range of tasks — summarization, translation, question answering, code generation, and more — without task-specific training. In commerce and enterprise contexts, LLMs serve as the foundation for capabilities including AI-powered search, product content generation, customer service automation, and intelligent document processing. They can be accessed via API (hosted models) or deployed on-premises. Their primary limitations — hallucination, context window constraints, and sensitivity to prompt phrasing — require careful system design. LLMs are often augmented with retrieval systems (RAG), tool use, and fine-tuning to adapt them to specific business domains and workflows.
Leading Indicators
Leading indicators are metrics that signal future performance or outcomes before they fully materialize. Unlike lagging indicators (which measure results after the fact, such as revenue or churn), leading indicators measure behaviors or conditions that predict those outcomes — for example, feature adoption rates, session depth, or model confidence scores that correlate with downstream conversion. In AI-powered commerce systems, identifying reliable leading indicators is essential for proactive management. If an AI recommendation engine shows declining click-through rates on top suggestions, that is a leading indicator of potential revenue impact before it appears in sales data. Similarly, tracking model input distribution shifts, user feedback scores, or query failure rates can surface model degradation early. Teams that instrument their AI systems around leading indicators can intervene before business outcomes are materially affected.
Learning Velocity
Learning velocity refers to the speed at which a model, system, or organization is able to improve its performance through experience, feedback, or new data. At the model level, it describes how quickly a model converges during training or fine-tuning. At the organizational level, it describes how rapidly teams iterate on AI systems based on real-world performance signals and user feedback. In enterprise AI programs, learning velocity is a competitive differentiator. Organizations that establish tight feedback loops — collecting outcome data, labeling it efficiently, retraining or fine-tuning models, and measuring impact — are able to compound improvements faster than those with slow, manual processes. For commerce applications such as personalization, search ranking, or demand forecasting, a higher learning velocity translates directly into more accurate models, better customer experiences, and faster adaptation to market shifts.
LoRA (Low-Rank Adaptation)
LoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning technique that adapts a pre-trained large language model by injecting small, trainable rank-decomposition matrices into the model's existing weight layers, rather than updating all parameters. Instead of fine-tuning billions of weights, LoRA trains only the low-rank matrices — typically representing less than 1% of total parameters — and keeps the original model frozen. The trained adapter can be merged back into the base model or swapped at inference time. LoRA has become a practical standard for enterprise AI customization because it dramatically reduces the GPU memory and compute required for fine-tuning. A commerce team can create domain-specific adapters for tasks like product taxonomy classification, brand-voice content generation, or customer intent detection — without provisioning the infrastructure required to fine-tune a full model. Multiple LoRA adapters can be maintained for different use cases and applied selectively, enabling flexible, cost-effective model specialization at scale.