AI Glossary for eCommerce Teams. 26 terms defined. Company: The organisation that builds and sells the AI. This wiki covers five: OpenAI, Google, Anthropic, xAI, and Meta. Product: The interface or platform where you use a model. Examples: ChatGPT, Gemini app, Claude.ai, Codex, Claude Code, an API. Model: The underlying AI engine that processes your input and produces output. Examples: GPT-5.2, Gemini 3 Pro, Claude Sonnet 4.5. Mode: A setting within a product that changes how the model behaves. Examples: Thinking mode, Deep Think, extended thinking. Agent: A product or workflow where the model can take multi-step actions autonomously, such as reading files, running code, browsing the web, or calling tools. API: Application Programming Interface. A way for developers to send prompts to a model programmatically and receive responses, rather than using a chat UI. Tier: A classification of models by their depth of reasoning. This wiki uses four tiers: Fast, Balanced, Deep Reasoning, and Specialist. Fast tier: Models optimised for speed and low cost. Best for high-volume repetitive tasks like bulk tagging, classification, and simple copy. Balanced tier: Models that balance speed, cost, and quality. Suitable for strategy, analysis, campaign work, and most day-to-day tasks. Deep Reasoning tier: Models that spend more time thinking before answering. Best for high-stakes decisions, pricing logic, and policy-sensitive work where mistakes are expensive. Specialist tier: Models purpose-built for a specific task type such as coding, image generation, or video generation. Not general-purpose. Rating: The assessment given to a model for a specific eCommerce task. This wiki uses three ratings: Recommended, Caution, and Not Applicable. Recommended: The model is a strong fit for the task. It has the right capabilities and should produce good results with reasonable prompting. Caution: The model can handle the task, but there are trade-offs such as cost, speed, or the need for more careful prompting. Evaluate before committing. Not Applicable: The model is not designed for this task. For example, an image model cannot do ticket triage, and a text model cannot generate video. Speed band: A qualitative indicator of how fast a model responds. This wiki uses three bands: Fast, Balanced, and Slow. Cost band: A qualitative indicator of how expensive a model is to use. This wiki uses three bands: Cheap, Mid, and Premium. Task Matrix: A feature in this wiki that maps 28 common eCommerce tasks against 24 AI models, showing which models are recommended, which require caution, and which are not applicable. Task Routing: A decision framework for choosing the right model tier based on the nature of your task, from fast and cheap to deep and expensive. Open-weight: A model whose weights are publicly released so it can be self-hosted. Useful when data privacy, compliance, or vendor independence is a priority. Meta's Llama models are open-weight. Prompt: The text instruction you send to a model. Prompt quality significantly affects output quality across all models. Token: The unit of text that models process. Roughly four characters or three-quarters of a word in English. Pricing is typically per token. Context window: The maximum amount of text (measured in tokens) a model can process in a single request, including both the prompt and the response. Fine-tuning: The process of further training a model on your own data to specialise it for a particular task, tone, or domain. Hallucination: When a model generates information that sounds plausible but is factually incorrect. More common with fast models on knowledge-intensive tasks. Multimodal: A model that can process multiple types of input or output, such as text, images, audio, or video. This page also explains the methodology behind the eCommerce AI Wiki ratings.