AI and software glossary
Plain definitions for the terms that come up when you plan, buy or build AI and business software.
A
Agentic workflowAn agentic workflow is a multi-step process where AI decides which step to take next, calls approved tools and checks its progress toward a defined goal.AI acceptable use policyAn AI acceptable use policy is a written company rule that states which AI tools employees may use, what data they may enter into them and what still requires human review.AI agentAn AI agent is software that uses a language model to choose its next step and act through approved tools within set permissions, while a chatbot only replies to questions.AI gatewayAn AI gateway is a layer between your applications and AI model providers that routes requests, enforces access rules and records usage and cost in one place.AI governanceAI governance is the set of policies, roles and controls that decide which AI uses are allowed, what data they may touch and who is accountable for the results.AI guardrailsAI guardrails are the technical and policy controls that keep an AI system inside agreed limits, such as blocking certain topics, capping spend or requiring approval before an action runs.AI inferenceAI inference is the step where a trained model processes new input and produces an output, such as a classification, a prediction or a piece of generated text.AI observabilityAI observability is the practice of logging and monitoring an AI system in production, tracking things like inputs, outputs, latency, cost and errors so problems can be found quickly.AI readiness assessmentAn AI readiness assessment reviews a company's workflows, data, systems and people to decide where AI can help, what must be fixed first and what to leave alone.AI workflow automationAI workflow automation uses software to move work between people and systems, with AI handling steps that need reading, classifying or drafting rather than fixed rules.Apache KafkaApache Kafka is an open-source event streaming platform that stores messages in an ordered, replayable log, so many systems can read the same stream independently.API (application programming interface)An API is a defined way for one software system to request data or actions from another, such as creating an order or reading a customer record.Approval workflowAn approval workflow is a defined sequence of steps that routes a request, document or transaction to the right person for review and sign off before it takes effect.Audit trailAn audit trail is the chronological record of who did what, when and through which system, kept so an action can be reconstructed and checked later.
B
Bill of ladingA bill of lading is the document a carrier issues to a shipper that records the goods received, the terms of carriage and, in some forms, who owns the cargo.Business process automation (BPA)Business process automation is the use of software to run a defined business process, such as onboarding or invoice approval, with less manual handling at each step.
C
Carrier vettingCarrier vetting is the process of verifying, before the first load, that a trucking company is who it says it is and holds the operating authority and insurance it claims.Cloud migrationCloud migration is moving applications, data or infrastructure from on-premises servers or another provider into a cloud platform such as AWS or Azure.Confidence scoreA confidence score is a number an AI or machine learning model produces alongside its output that estimates how likely that output is to be correct.Conflict checkA conflict check is a search of a firm's client, matter and party records, run before taking on new work, to find relationships that could create a conflict of interest.Context windowA context window is the maximum amount of text, measured in tokens, that an AI model can consider at once across a prompt, conversation history and its response.CRM (customer relationship management)A CRM is the system a business uses to record customers, contacts, deals and interactions, such as Salesforce, HubSpot or Microsoft Dynamics 365.
D
Data governanceData governance is the set of rules, roles and processes that decide how a company defines, owns, protects and maintains the quality of its data.Data pipelineA data pipeline is an automated process that collects data from source systems, cleans and reshapes it, and delivers it to the place where it is used.Data residencyData residency is the requirement that certain data be stored and processed within a specific country or region, usually to satisfy a law or a customer contract.Data warehouseA data warehouse is a central database built for reporting and analysis that combines cleaned data from systems such as the ERP, CRM and finance tools.
E
EDI (electronic data interchange)EDI is the standardized electronic exchange of business documents, such as purchase orders and invoices, directly between the computer systems of two trading partners.EmbeddingsEmbeddings are lists of numbers that represent the meaning of text or images, so software can find content that is similar in meaning rather than matching exact words.ERP (enterprise resource planning)An ERP is the core system that runs finance, purchasing, inventory and order management from one database, such as NetSuite, SAP, Microsoft Dynamics 365 or Odoo.ETL (extract, transform, load)ETL is the process of extracting data from source systems, transforming it into a consistent format and loading it into a destination such as a data warehouse.Evals (AI evaluation)Evals are structured tests that measure how well an AI system performs on a fixed set of realistic examples, so changes can be compared before they reach users.Exception handlingException handling is the part of an automated process that detects a case it cannot handle confidently and routes it to a person instead of guessing or failing silently.
F
Fine-tuningFine-tuning is further training of an existing AI model on your own examples so it follows a specific format, style or task more consistently.FinOpsFinOps is the practice of managing cloud spending as a shared responsibility of finance, engineering and business teams, so costs are visible, owned and tied to value.
G
Generative AIGenerative AI is AI that creates new content, such as text, images, audio or code, rather than only classifying or scoring information that already exists.Global Trade Item Number (GTIN)A GTIN is the globally unique number, issued under GS1 standards, that identifies a trade item and appears in the barcode on its packaging.Golden datasetA golden dataset is a fixed set of real examples with agreed correct answers, used to test whether an AI system gets the work right before and after every change.Grounding (AI)Grounding is the practice of connecting an AI system's answers to specific, verifiable sources, so its output can be checked and traced rather than taken on faith.
H
HallucinationA hallucination is output from an AI model that sounds confident and plausible but is false, unsupported by its sources or invented outright.Human-in-the-loopHuman-in-the-loop is a design where a person reviews, corrects or approves an AI system's work at defined points before it takes effect.
I
Intelligent document processingIntelligent document processing uses OCR and AI to read documents such as invoices, contracts and forms, extract the needed fields, check them and route exceptions for review.iPaaS (integration platform as a service)An iPaaS is a cloud platform for connecting business applications and automating data flows between them, such as Workato, Boomi, MuleSoft or Zapier.
K
L
Large language model (LLM)A large language model is an AI model trained on large amounts of text to understand and generate language, such as OpenAI's GPT models, Anthropic's Claude and Google's Gemini.Lease abstractionLease abstraction is the process of reading a lease and its amendments and recording the key dates, rent terms and clauses in a structured summary or lease register.Least privilegeLeast privilege is the security principle that every person, application or AI agent gets only the minimum access needed to do its job, and nothing more.
M
Master data management (MDM)Master data management is the discipline of keeping one agreed, accurate version of core business records, such as customers, products, suppliers and prices, across every system.Model Context Protocol (MCP)The Model Context Protocol is an open standard that lets an AI application connect to external data sources and tools through one common interface instead of a custom integration per source.Model driftModel drift is the gradual drop in an AI system’s accuracy that happens when the real-world inputs, or the model behind it, change after it was tested.Model-agnostic architectureModel-agnostic architecture is a way of building AI systems so the underlying model or provider can be changed without rebuilding the application around it.Multi-agent systemA multi-agent system is a design where several specialized AI agents each handle part of a task and coordinate with each other, rather than one agent handling everything.
O
Optical character recognition (OCR)Optical character recognition converts text in scanned documents, photos and image-based PDFs into machine-readable text that software can search and process.Order falloutOrder fallout is when an order drops out of its automated path, such as provisioning or fulfillment, and needs a person to fix it before it can continue.Order management system (OMS)An order management system tracks every order from checkout or receipt through allocation, fulfillment, shipping, returns and refunds, usually across several sales channels.
P
Personally identifiable information (PII)Personally identifiable information is any data that can identify a person on its own or when combined with other data, such as a name, an email address or an ID number.PollingPolling is when one system asks another on a schedule whether anything has changed, rather than waiting to be told.PrefetchingPrefetching loads data or a page before the user asks for it, so the next screen appears almost immediately when they do.Process miningProcess mining is a technique that analyzes the event logs a company's own systems already produce to show how a process actually runs, rather than how it is documented.Product information management (PIM)Product information management is the system and process for collecting, checking and publishing product data, such as attributes, descriptions and images, to every sales channel.Prompt engineeringPrompt engineering is the practice of writing and refining the instructions given to an AI model to make its output more accurate, consistent and useful for a specific task.Prompt injectionPrompt injection is an attack where instructions hidden in content an AI system reads, such as an email or web page, try to override its rules.Proof of conceptA proof of concept is a small, time-limited build that tests whether a specific idea works with real data before a company commits to full development.Publish-subscribe (pub/sub)Publish-subscribe is a messaging pattern where a sender publishes a message to a topic and every subscribed system receives its own copy, without the sender knowing who they are.
R
Requirements briefA requirements brief is a short decisions document that states the business problem, the users, what the first release must do and how success will be measured.Retrieval-augmented generation (RAG)Retrieval-augmented generation is a technique where an AI system first finds relevant passages in approved sources, then uses them to write an answer with references.Robotic process automation (RPA)Robotic process automation uses software bots that copy a person's clicks and keystrokes in applications to complete repetitive, rule-based tasks.
S
Segregation of dutiesSegregation of duties is the control principle of splitting related tasks, such as approving a payment and recording it, across different people so no one person controls a whole process.Semantic searchSemantic search finds content that matches the meaning of a query rather than only the exact words typed, using embeddings to compare meaning instead of matching keywords.Server-sent events (SSE)Server-sent events are a web standard that lets a server push a stream of updates to a browser over one ordinary HTTP connection, in one direction only.Shadow AIShadow AI is the use of AI tools at work without the approval or knowledge of IT and security.Shadow modeShadow mode is a way to test an AI system by running it on live inputs while its output is recorded for comparison but never acted on.ShardingSharding splits one database across several servers, with each server holding part of the data and a shard key deciding where each record lives.Single sign-on (SSO)Single sign-on lets employees use one company login, usually managed in Microsoft Entra ID, Okta or Google Workspace, to access many applications.Speed to leadSpeed to lead is the time between a prospect submitting an inquiry and a person or system from the business making a useful first response.Straight-through processing (STP)Straight-through processing completes a transaction from start to finish through automated systems with no manual rekeying.System of recordA system of record is the single system a company treats as the authoritative source for a given type of data when other systems disagree or fall out of sync.
T
Technical debtTechnical debt is the future cost of shortcuts in software or systems, paid as slower changes, more errors and higher maintenance until the shortcuts are fixed.Three-way matchingThree-way matching is an accounts payable control that compares the purchase order, the goods receipt and the vendor invoice before a payment is approved.TokensTokens are the small pieces of text, often parts of words, that an AI language model reads and generates.Total cost of ownershipTotal cost of ownership is the full cost of a system over its useful life, including purchase or build, licenses, infrastructure, integration, support and training.Trade promotion management (TPM)Trade promotion management is the process of planning, budgeting, approving and measuring the discounts and in-store promotions a consumer goods company runs with its retail partners.Transformer modelA transformer is the neural network design behind modern large language models, using a mechanism called attention to weigh how each word relates to the others.Transportation management system (TMS)A TMS is software that plans, executes and tracks the physical movement of goods, covering carrier selection, route planning, load building and freight cost audit.
V
Vector databaseA vector database stores embeddings and quickly finds the entries closest in meaning to a query, which makes it a common component of AI search and assistants.Vendor lock-inVendor lock-in is a state where switching away from a supplier has become so costly or difficult, due to proprietary formats or deep integration, that a company stays by default.
W
WebhookA webhook is an automatic message one system sends to another the moment an event happens, such as a new order being placed or a form being submitted.WebRTCWebRTC is an open standard for sending live audio, video and data between browsers and apps with very low delay, encrypted by default.WebSocketA WebSocket is a long-lived, two-way connection between a browser or app and a server, so either side can send a message at any time.
Z
See how an idea becomes a working solution.
Explore how we scope, build and hand over AI projects, with clear decisions at each stage.