B2B Automation Glossary

Definitions, examples, and practical notes on AI, APIs, and workflow automation.

A

  • Agent

    A system in which a language model is placed in a loop with tools, memory and a goal, so that it decides the next action based…

  • Agentic Flow

    The runtime behaviour of an agentic system: the sequence of observations, decisions and actions that unfolds as the model pursues a goal. Where agentic workflow names…

  • Agentic Workflow

    A hybrid pattern in which a deterministic orchestration layer defines the stages, and language models make bounded decisions inside each stage. Unlike a free-running agent, the…

  • AI Image Artefact Taxonomy

    A classification of the ways AI image processing fails visually, such as halos, seams, over-smoothing, fabricated text, repetition, colour shift and geometry distortion, used to diagnose…

  • AI Image Duplicate Detection

    Identifying images that are the same or near-identical, using perceptual hashing or embedding similarity, to avoid redundant processing and storage and to detect re-uploads. Duplicate detection…

  • AI Image Model Selection

    Choosing the right model for an image task by measuring performance on your own content rather than by benchmark reputation. Image models differ substantially by content…

  • AI Image Pipeline Documentation

    The record of a pipeline's steps, their order, their parameters and the rationale for each choice, so that the pipeline can be understood, modified and reproduced…

  • AI Image Pipeline Monitoring

    Instrumentation of an image pipeline's throughput, error rate, latency and quality, so that degradation is detected before users notice. Image pipelines fail silently, producing output that…

  • AI Image Pipeline Rollback

    Reverting an image pipeline to a previous model or parameter set, and reprocessing the affected assets, when a change is found to degrade quality. Rollback is…

  • AI Image Processing Retry Policy

    The rules governing how a failed image processing job is retried, including which errors are retryable, the backoff, and the maximum attempts. Image processing failures are…

  • AI Image Reference Conditioning

    Using one or more reference images to guide generation, controlling subject identity, style or composition rather than relying on text alone. Reference conditioning is what makes…

  • AI Image Storage Tiering

    Storing image assets across storage classes based on access frequency, keeping frequently served variants on fast storage and originals or rarely used variants on cheaper archival…

  • AI Image Throughput Planning

    Sizing the compute and concurrency needed to process an image workload within its required time, accounting for per-image processing time, batch efficiency and GPU memory limits.…

  • AI Workflow Definition Review

    A structured review of a workflow's design before build, covering the happy path, every branch, every failure mode and every compensation. Design review is far cheaper…

C

  • Context Window

    The maximum number of tokens a model can attend to in a single request, comprising the system prompt, the full message history, any retrieved documents, and…

E

  • Embedding

    A dense numeric vector, typically 384 to 3,072 dimensions, that represents the meaning of a piece of text in a continuous space where geometric proximity corresponds…

F

  • Fine-Tuning

    Further training of a pretrained model on a task-specific dataset to adjust its weights so that it reliably produces a desired style, format or behaviour. Supervised…

  • Function Calling (Tool Use)

    A model capability in which the model, instead of answering in prose, emits a structured request to invoke a named function with typed arguments. The application…

H

  • Hallucination

    A generated statement that is fluent, plausible and unsupported by any source, produced because the model optimises for likely continuations rather than for truth. Hallucination includes…

L

  • Large Language Model (LLM)

    A neural network trained on very large text corpora to predict the next token in a sequence, which after instruction tuning and alignment behaves as a…

  • LoRA (Low-Rank Adaptation)

    A parameter-efficient fine-tuning technique that freezes the base model and injects small trainable low-rank matrices into selected layers, typically the attention projections. Instead of updating billions…

M

  • Model Context Protocol (MCP)

    An open protocol that standardises how AI applications connect to external data sources and tools, separating the host application, the client and the server that exposes…

P

  • Prompt Engineering

    The discipline of designing the instructions, context, examples and output constraints that steer a language model toward reliable behaviour. It is not a trick but a…

R

  • Retrieval-Augmented Generation (RAG)

    An architecture in which a language model's answer is conditioned on documents retrieved at query time rather than on parametric memory alone. A RAG pipeline has…

  • RLHF (Reinforcement Learning from Human Feedback)

    An alignment technique that trains a reward model on human preference comparisons between candidate outputs, then optimises the language model against that reward using reinforcement learning,…

S

  • Semantic Search

    Search that ranks results by meaning rather than lexical overlap, implemented by comparing query and document embeddings. It handles synonyms, paraphrase, and natural-language questions that keyword…

  • System Prompt

    The privileged instruction block sent before any user turn that defines the assistant's identity, scope, rules, tone, available tools and output contract. It is the closest…

T

  • Temperature

    A sampling parameter that scales the logits before the softmax, controlling how much probability mass is spread across alternatives. At temperature zero the model effectively takes…

  • Tokenization

    The process of splitting text into the subword units a model actually consumes. Modern LLMs use byte-pair encoding or a variant such as SentencePiece, learning a…

  • Transformer Architecture

    The neural network design in which every token attends to every other token in the context through scaled dot-product self-attention, replacing recurrence with parallel matrix operations.…

V

  • Vector Database

    A storage system purpose-built for high-dimensional vectors that supports approximate nearest-neighbour search at scale, usually through an index such as HNSW, IVF-PQ or DiskANN. Beyond the…

W

  • Workflow Change Impact Analysis

    Determining what will be affected by a change to a workflow before making it, including downstream consumers, dependent workflows and historical data. Impact analysis prevents the…

  • Workflow Concurrency Control

    Limiting how many workflow instances or steps execute simultaneously, to protect downstream systems and to keep resource usage predictable. Unbounded concurrency is a common cause of…

  • Workflow Data Retention

    The policy governing how long a workflow retains its inputs, intermediate state, outputs and logs, balancing operational need against data protection obligations. Retention is a legal…

  • Workflow Error Classification

    Categorising failures as transient, permanent, or requiring human intervention, so that each is handled appropriately rather than uniformly retried. Uniform retry is the most common workflow…

  • Workflow Notification Design

    Deciding who is notified of what, when and through which channel, so that notifications are actionable rather than noise. Poor notification design causes alert fatigue, where…

  • Workflow Operational Runbook

    The documented procedure for operating a workflow, covering routine checks, common failures and their resolution, escalation and the location of logs and dashboards. The runbook is…

  • Workflow State Persistence

    Storing a workflow instance's state durably so that it survives process restarts, deployments and failures, and can be resumed. Without persistence, a long-running workflow cannot survive…

  • Workflow Step Dependency

    The relationship between steps where one requires the output of another, determining the order of execution and the opportunities for parallelism. Explicit dependencies are what allow…