Glossary

Machine learning is the branch of computing in which a program works out its rule from examples instead of being handed that rule by a programmer, and is judged on how well the rule holds on cases it was never shown.

A neural network is a stack of layers of very simple arithmetic units whose connection strengths are adjusted by training until the whole arrangement turns the inputs you have into the outputs you want.

Deep learning is machine learning built on neural networks of many stacked layers, which work out for themselves which properties of the raw data matter instead of being handed a list by a person.

Generative AI is the market's name for models whose output is a piece of content, such as text, images, audio, video or code, as opposed to the much larger body of models that sort, score or forecast things that already exist.

Large language models are neural networks trained on a vast quantity of written text to predict the piece of text that comes next, an objective narrow enough to state in a line and broad enough to yield writing, translation, code and summary as side effects.

A transformer is a neural-network design that takes a whole sequence in at once and, for every element in it, works out how strongly each of the other elements should influence that one.

Prompt engineering is the practice of composing what you send a generative model, the instruction, the supporting material and the worked examples, so that it returns something you can use.

Retrieval-augmented generation is a pattern that searches a body of documents when a question arrives and places the passages it finds into the model's prompt, so the answer is drawn from those sources rather than from training alone.

Natural language processing is the area of AI that makes human language readable and writable by machine: sorting text into categories, pulling structured facts out of it, translating it, generating it and measuring whether any of that came out right.

Computer vision is the branch of artificial intelligence that turns images and video into structured answers: what is in the frame, where it sits, which pixels belong to it, and where it moves next.

Agentic AI is software that wraps a language model in a loop where it can request actions from real tools, read what each action returns, and choose the next move until the job is finished or a rule stops it.

MLOps is the engineering practice of keeping a trained model useful in production: recording what produced it, serving it, watching for the day its answers stop matching the world, and retraining or reverting before anyone downstream is harmed by the gap.

Responsible AI is the practice of building and running systems whose decisions you could defend to the person on the receiving end of them, covering fairness, transparency, privacy and accountability.

Data analysis is the practice of interrogating data to answer a specific question, and of establishing how much weight the answer can bear.

Data wrangling is the work of repairing and reshaping real records, and of deciding what their gaps and inconsistencies mean, until a table can be trusted with the question being asked of it.

Data visualisation is the practice of encoding numbers as position, length and colour so that a comparison a reader would otherwise have to calculate becomes something they can simply see.

Data engineering is the discipline of moving records from the systems that produce them into a store where they can be queried, and of keeping that supply correct while the sources underneath keep changing.

SQL is the declarative language for asking questions of relational databases: you state which rows and columns you want and how tables relate, and the database engine works out how to retrieve them.

ETL is the pattern for moving data from the systems that produce it into the one place it is analysed: extract the records from each source, transform them into a common, cleaned shape, and load them into a warehouse; ELT keeps the same three steps and runs the transformation inside the warehouse after loading.

A data warehouse is a database built for analysis rather than for running the business: it holds copies of records from many operational systems, integrated under one set of definitions, kept as history rather than overwritten, and arranged so that questions across years and departments run fast.

A data lake is a store that keeps data in its original form, files of any shape in cheap object storage, loaded before anyone has decided what questions will be asked of it, so that the structure is applied when the data is read rather than when it is written.

Business intelligence is the practice and the tooling for turning an organisation's data into reports and dashboards that people use to run it: connecting to the sources, modelling the data into measures everyone agrees on, and presenting it so that a manager sees the state of the business without asking an analyst.

Data governance is the set of decisions and responsibilities an organisation puts in place over its data: who owns each dataset, what each field means, who may read or change it, how long it is kept and how its quality is measured, recorded so that the answers do not depend on who is asked.

Data quality is the degree to which a dataset is fit for the purpose it is used for, assessed along named dimensions, whether the records are complete, unique, consistent, timely, valid and accurate, and measured against the requirements of the decision the data supports rather than against an absolute standard.

Data modelling is the design of how data is structured for a given use: which entities are recorded, which attributes describe them, how the tables relate through keys, and, for analytical use, which table holds the measurements and which hold the context, decided before the data is loaded so that the questions the model must answer run correctly and fast.

Hypothesis testing is the statistical procedure for deciding whether an observed difference, between two groups, two periods or a sample and a claimed value, is larger than chance variation would produce if there were no real difference, by computing how probable a result at least that extreme would be under that assumption.

Python is a general-purpose programming language that became the standard interface for driving numerical and machine-learning code written in other languages.

Git is a distributed version control system that records the history of a set of files as a graph of snapshots, lets many people work on branches of that history at once, and merges their work back together.

HTTP is the request-and-response protocol of the web: a client sends a method, a path and headers, a server answers with a status code, headers and a body, and each exchange stands alone unless the two sides agree to carry state in a header.

A REST API exposes an application's data as resources at stable addresses and lets clients read and change them with HTTP's standard methods and status codes, so that the protocol itself carries most of the meaning.

JSON is a text format for exchanging structured data: values are strings, numbers, booleans or null, arranged in ordered arrays and in objects of name and value pairs, and every mainstream language reads and writes it without a library of its own.

GraphQL is a query language for APIs and a server-side runtime for answering those queries: a client asks for precisely the fields it needs from a typed schema, in one request, instead of calling several fixed endpoints and discarding most of what comes back.

Microservices is an architectural style in which one application is built as a set of small services, each running in its own process, owning its own data and talking to the others over lightweight network calls, so that each can be built, deployed and scaled on its own.

Continuous integration is the practice of every developer merging their work into a shared mainline at least once a day, with an automated build and test run on each merge, so that integration problems surface within minutes of being created rather than weeks later at release time.

Unit testing is the practice of writing small automated checks that call one piece of code in isolation, a function, a class or a module, with known inputs and assert on its outputs, so that a change which breaks that behaviour fails a test within seconds of being made.

Full-stack development is the ability to build both halves of a web application: the front end that runs in the browser (HTML, CSS and JavaScript) and the back end that runs on a server (a language, a framework and a database), plus the API that joins them.

A framework is a body of code that imposes a structure on the application built with it: it decides how files are organised, when the application's own code is called and how its parts talk, whereas a library is code the application calls when it chooses.

Object-oriented programming is a way of organising a program as objects, each holding its own data and the operations allowed on it, with classes as the templates objects are made from, so that the code models the things it is about rather than the steps the computer takes.

Data structures are the ways a program arranges data in memory (arrays, lists, hash tables, trees, graphs) and algorithms are the step-by-step procedures that work on them; together they decide whether an operation takes a millisecond or an hour as the data grows.

Digital marketing is the trade name for every commercial activity that reaches a buyer through a screen, gathered under one heading because the work is bought, delivered and counted through the same few platforms rather than because the activities resemble one another.

SEO is the work of being retrievable and worth retrieving: legible to the systems that crawl, index and select sources, and useful enough at the moment of the query that somebody still opens the page instead of accepting the answer printed above it.

Marketing automation is software that sends a decided message when a contact does a decided thing, which makes it a delivery mechanism for judgements somebody has already formed rather than a way of avoiding them.

Pay-per-click is advertising bought by the click rather than by the impression: an advertiser bids on the searches or placements it wants to appear for, and pays only when someone clicks the ad, with each click priced by an auction run at the moment the page loads.

Keyword research is the work of finding out which queries people type when they want what a business offers, how many of them there are, what each query means and what competing for it costs, before any page is written or any ad is bought.

Conversion rate optimisation is the practice of raising the share of visitors who complete the action a page exists for, a purchase, a sign-up, an enquiry, by changing the page and the path to it on evidence rather than opinion.

A/B testing is a controlled experiment on a live audience: visitors are assigned at random to two or more versions of a page, an email or a feature, and the versions are compared on one pre-chosen outcome so that the difference can be attributed to the change rather than to who happened to see it.

Retargeting is advertising shown to people because of something they did before: visited a site, viewed a product, abandoned a basket or appeared on a customer list, with the ad platform recognising them later on other sites, in search results or in a video feed and serving an ad tied to that earlier action.

Content marketing is the practice of earning customers by publishing material they want for its own sake, guides, comparisons, tools, data, video, so that the people who need what a business sells find it, trust it and come back, instead of being interrupted by an advertisement.

The marketing funnel is a model of how strangers become customers in stages, awareness, interest, consideration, purchase, with fewer people at each stage than the one before, used to decide which message, channel and measure belong to which stage.

Marketing attribution is the assignment of credit for a sale or a sign-up to the advertisements, searches, emails and visits that preceded it, according to a rule called an attribution model, so that each channel's reported results reflect its share of the outcomes it helped cause.

Return on ad spend is the revenue a campaign produced divided by what it cost, expressed as a ratio or a percentage: five pounds of sales for every pound of advertising is a ROAS of 5, or 500 per cent, and the number only becomes a decision once it is compared with the margin on those sales.

Shared hosting puts your website on a machine that is simultaneously running hundreds of other people's, pooling its processor, memory and disk between all of you, which is both why it costs so little and why your speed partly depends on strangers.

A VPS is a software-defined slice of one physical machine that behaves as a server in its own right, with its own operating system, reserved CPU, memory and disk, and administrative access down to the kernel.

A dedicated server is an entire physical computer rented by a single customer, with no hypervisor taking a share and no other tenants on the hardware, chosen when consistent performance, specific components or provable isolation matter more than the ability to resize in seconds.

Managed hosting is any plan where the provider runs the server on your behalf, covering updates, hardening, monitoring, backups and tuning, so the thing being sold is administration time with a machine attached rather than the machine itself.

Cloud hosting is the model in which servers, storage and databases are created on demand through an API from a provider's pooled hardware and billed for the time they exist, rather than rented as a fixed box by the month.

A web server is the program listening on a network port for HTTP requests and answering them, either by returning a file from disk or by handing the request to application code behind it.

Load balancing is the practice of putting one entry point in front of several interchangeable servers and distributing incoming requests among them by a chosen policy, so that no single machine carries the traffic and no single machine's failure takes the service down.

A CDN is a worldwide network of servers that answers requests for your site from a location near the person asking, instead of sending every request back to the single machine your site runs on.

DNS is the distributed lookup system that turns a hostname such as getting-digital.net into the IP address a browser can connect to, by walking a chain of nameservers whose answers are then cached for a lifetime the domain's owner publishes.

SSL/TLS is the protocol that encrypts the connection between a browser and a server and, by way of a certificate signed by a certificate authority, proves the server is entitled to answer for the domain in the address bar.

A domain name is an address in the internet's public namespace that you rent through a registrar for a yearly fee, and it remains yours only while the registrar account holding it stays paid up and under your own control.

Caching keeps the result of expensive work, such as a rendered page or a database query, and serves that stored copy until it expires, making it the largest single source of speed on any busy website.

Uptime is the proportion of time a service answers requests, quoted in nines, and a service level agreement is the contract clause setting out how that proportion is counted and what the provider owes when it falls short.

IaaS is the layer of cloud computing where you rent raw infrastructure through an API, meaning virtual machines, disks and networks, and take responsibility for the operating system and everything you install above it.

PaaS is a hosting layer that takes your source code and turns it into a running application for you, owning the machines, the runtime, the build, the routing, the certificates and the restarts, so that shipping becomes a push rather than a project.

Serverless computing runs your code only while an event is calling it: the provider conjures an execution environment on demand, throws it away afterwards, and charges for work performed instead of for capacity sitting idle.

A container is an application packaged together with the libraries and configuration it needs, run as an isolated process on a host that is already switched on, so the same package behaves the same way on a laptop, in a build pipeline and in production.

Object storage keeps every file as a self-contained object (the bytes, a unique key and some metadata) inside a flat bucket you read and write over HTTP, with no drive to mount and no directory tree underneath it.

Kubernetes is an open-source system that runs containers across a cluster of machines, keeping the number and placement of running copies matched to a declared desired state, and restarting, rescheduling and scaling them without a person intervening.

A stock licence is the permission you buy when you download an asset, naming who may use it, inside which products, at what scale and until when, while copyright stays with whoever made it.

Royalty-free is a pricing arrangement in which one payment replaces the per-copy, per-view and per-year fees a licence could otherwise charge, and it decides nothing else about how the asset may then be used.

A web template is a pre-built layout sold as permission to use it in one end product, delivered either as flat files you adapt or as a theme wired into a content management system and its release cycle.

A design system is a versioned set of tokens, coded components and usage rules that a team builds interfaces from, so an interface decision is made once and then imported rather than redrawn on every screen.

Font licensing sells permission for one named context of use at a time, so installing a family to make artwork, serving it to browsers, embedding it in software and using it on air are separate purchases rather than one.

A client brief is the short written statement of what a piece of work has to achieve, for whom, and under which fixed constraints, agreed before either side commits to a solution.

A statement of work is the document that commits one bounded piece of work to paper: what gets built, on which assumptions, to what deadline, for what money, and what sits outside it.

Scope creep is the quiet expansion of a priced piece of client work through additions that each look too minor to renegotiate, until what gets delivered is substantially larger than what was sold.

A retainer is a recurring fee that holds a supplier's time or availability open across a period, so what the client buys is first call on somebody's attention rather than any named artefact.

A day rate sells a bounded unit of a supplier's working time rather than a finished result, so the client rather than the supplier carries the risk that the job turns out bigger than it looked.

Zero trust is a security model in which no request is trusted because of where it comes from, so every access to a resource is authenticated, authorised and checked against policy each time.

Encryption transforms readable data into ciphertext with a key so that only someone holding the right key can turn it back, protecting the data while it is stored or in transit.

A hash function turns input of any size into a fixed-size fingerprint in a way that cannot be reversed, so that the fingerprint can verify data or store a password without holding the original.

A public key infrastructure is the system of certificates, certificate authorities and revocation mechanisms that lets a party verify which identity a public key belongs to before trusting it.

Authentication is the process of confirming that a user, device or service is who it claims to be, typically by checking something it knows, has or is.

Multi-factor authentication requires proof from at least two different kinds of factor, such as a password and a device-generated code, so that stealing one is not enough to log in.

Identity and access management is the set of policies and systems that create, verify, govern and retire identities and decide what each one may reach.

Least privilege is the principle that every user, process and system should hold only the access required for its current task, and no more.

A firewall is a control that allows or blocks network traffic according to rules about its source, destination, protocol and, in modern forms, its content and the application it belongs to.

A virtual private network carries traffic through an encrypted tunnel over a public network so that a remote device or site behaves as though it were connected to a private one.

An intrusion detection system watches network traffic or host activity for signs of attack and raises alerts, while an intrusion prevention system sits inline and can block what it recognises.

A SIEM collects logs and events from across an organisation's systems into one searchable store, correlates them into alerts, and retains them for investigation and compliance.

Vulnerability management is the continuous process of finding weaknesses in systems, deciding which ones matter in context, fixing or mitigating them, and confirming the fix.

A penetration test is an authorised, scoped attempt by a skilled tester to exploit weaknesses in a system the way an attacker would, ending in a report the owner can act on.

Incident response is the prepared, practised process by which an organisation detects a security incident, contains it, removes the cause, recovers, and learns from what happened.

Threat modelling is the structured practice of asking what could go wrong with a system, who would make it go wrong and how, while the design can still be changed cheaply.

Phishing is an attack that deceives a person into revealing credentials, approving a payment or running malicious content, usually through a message that impersonates someone they trust.

Ransomware is an attack in which criminals gain access to an organisation's systems, encrypt or steal its data, and demand payment to restore access or withhold publication.

A security operations centre is the team, tooling and process that monitors an organisation's systems for security events around the clock, triages alerts, and starts the response to those that are real.

A security risk assessment identifies the threats to an organisation's assets, estimates how likely each is and how much harm it would do, and ranks them so that limited effort goes where it reduces the most harm.

TCP/IP is the family of protocols on which the internet runs: IP moves packets between addresses without guarantees, and TCP or UDP ride on top to deliver either a reliable ordered stream or fast unordered messages.

The OSI model is a seven-layer reference framework that separates networking into physical transmission, link, network, transport, session, presentation and application concerns, used to reason about where a problem sits rather than to describe any real protocol stack.

IP addressing is the scheme that gives every interface on a network a numeric address, with IPv4's four-byte addresses now supplemented by IPv6's much larger space, and with private ranges reserved for use inside networks that reach the internet through translation.

Subnetting divides an address block into smaller networks by choosing how many leading bits identify the network, written as a prefix length or a subnet mask, so that each subnet has its own range of host addresses and its own routing entry.

Routing is how packets move between networks: each router holds a table of known destinations and the next hop toward each, learned by configuration or by a protocol, and forwards each packet to the most specific match.

Switching is the forwarding of frames within a single network segment by a device that learns which hardware address sits on which port and sends each frame only where it needs to go.

A VLAN divides one physical switched network into several logical segments, each with its own broadcast domain, by tagging frames with a VLAN identifier so that ports and devices in different VLANs cannot reach each other without a router.

Network address translation rewrites the addresses and ports in packets as they cross a network boundary, most commonly so that many devices with private addresses can share one public address to reach the internet.

DHCP is the protocol by which a device joining a network asks for and receives an address, a subnet mask, a default gateway and DNS servers from a server, for a limited lease period, so that nobody has to configure addresses by hand.

A wireless LAN connects devices to a network over radio using the 802.11 family of standards, through access points that bridge the wireless medium onto the wired network and enforce the authentication and encryption that the shared airwaves demand.

A project charter is the short document that formally authorises a project, names its sponsor and manager, states its purpose, its intended outcomes and its boundaries, and gives the manager the authority to spend the organisation's resources on it.

Project scope is the agreed description of what a project will deliver and what it will not, stated precisely enough that a change to it can be recognised as a change and decided rather than absorbed.

A work breakdown structure decomposes a project's total scope into a hierarchy of deliverables and, at the bottom, work packages small enough to estimate, assign and track, so that everything to be produced appears exactly once.

The critical path is the longest chain of dependent activities through a project's schedule, which sets the earliest possible finish date and on which any delay delays the whole project.

A Gantt chart is a schedule drawn as horizontal bars against a calendar, one bar per activity, with dependencies drawn between them and progress shown along each bar.

A risk register is the project's living list of things that might happen and would matter, each with an estimate of likelihood and impact, a planned response, an owner and a review date.

Stakeholder management is the practice of identifying everyone who affects or is affected by a project, understanding what each wants and can do, and engaging them deliberately so that the project keeps the support it needs and hears the objections while they are cheap.

Change control is the process by which a proposed change to a project's agreed baseline is recorded, assessed for its effect on scope, schedule, cost and risk, and then approved, rejected or deferred by a person holding the authority to do so.

Earned value management measures project performance by comparing the budgeted value of work completed with the value planned to be complete by now and with the cost incurred, producing variances and indices that forecast the final cost and date.

Agile is a family of delivery approaches that build a product in short iterations, each producing something usable, so that plans are corrected by what was learned rather than defended against it.

Scrum is a lightweight framework for delivering complex work in fixed-length iterations called sprints, with three accountabilities, five events and three artefacts, defined in a short public guide.

A sprint is a fixed-length iteration of one month or less in which a team plans a goal, builds a usable increment toward it, shows the result, and reviews how it worked, before the next sprint begins immediately.

A product backlog is the single ordered list of everything that might be done to improve a product, owned by one person who decides the order, refined continuously, and drawn from when each iteration is planned.

Kanban is a method for managing knowledge work by making it visible on a board, limiting how much is in progress at each stage, and improving the flow of items from start to finish rather than working in fixed iterations.

A user story is a short description of a piece of functionality from the perspective of the person who needs it, stating who they are, what they want and why, and carrying acceptance criteria that say when it is done.

The definition of done is the shared, written standard that every piece of work must meet before it counts as complete, covering the quality, testing, documentation and release conditions that make an increment usable.

A retrospective is a regular, structured meeting in which a team examines how it worked over the last period, decides what to change, and commits to one or two concrete improvements for the next one.

Incident management is the service-management practice of restoring a disrupted IT service to normal operation as quickly as possible, through a defined process of logging, categorising, prioritising, resolving and closing each incident, separate from the later work of finding and fixing its root cause.

A recording method in which every transaction is entered at least twice, as a debit to one account and an equal credit to another, so that total debits always equal total credits and the ledger checks its own arithmetic.

Under the accruals basis, income and expenses belong to the period in which they are earned or incurred rather than the one in which cash moves; an accrual is the adjusting entry for a cost incurred or income earned that has not yet been paid or invoiced.

Financial statements are the structured report of an entity's financial position, performance and cash movements over a period; under IFRS the complete set is a statement of financial position, a profit or loss and comprehensive income statement, a statement of changes in equity, a cash flow statement and notes.

Movement of cash and cash equivalents into and out of an entity over a period; IAS 7 reports it in three groups, operating, investing and financing, so readers can see whether the business funds itself from trading or from selling assets and raising money.

The difference between current assets and current liabilities: the short-term resources, such as cash, receivables and inventory, that remain once obligations falling due within the operating cycle or the next twelve months are set against them.

The principle that a sum available now outweighs an equal amount received later, because money in hand can be invested to earn a return; present and future values are linked through an interest or discount rate and the number of periods between them.

The sum of a project's expected cash flows, each discounted to today at the required rate of return, minus the initial outlay; a positive result estimates the value the project adds to the firm, and a negative one the value it would destroy.

A valuation method that estimates what an asset or company is worth today by forecasting the free cash it will generate, adding a terminal value for the years beyond the forecast, and discounting both at a rate that reflects their risk.

WACC blends the returns a company's lenders and shareholders require, each weighted by its share of the capital structure, with debt counted after tax; it is the rate for discounting firm-level cash flows and the usual benchmark for new investment.

Earnings before interest, taxes, depreciation and amortisation: net profit with financing costs, tax and the non-cash charges for using up tangible and intangible assets added back, used as a rough gauge of operating earnings across companies financed in different ways.

How an investment portfolio is divided between asset classes such as shares, bonds, property and cash, chosen to fit the investor's time horizon and capacity to bear losses; it sets the portfolio's broad risk and return profile before any individual holding is picked.

Spreading money across holdings whose prices do not all move together, within an asset class and between classes, so that a loss in one position is partly offset elsewhere and no single failure decides the outcome for the whole portfolio.

A pooled fund, whether a mutual fund, an ETF or a unit investment trust, that aims to match the return of a chosen market index, before fees, by holding the securities in that index rather than selecting investments on a manager's judgement.

Using borrowed money, or instruments that need only a small deposit, to control a larger position, so that gains and losses are measured against a smaller base of one's own capital; also the ratios, such as debt to equity or a bank's capital to exposures, that measure it.

A contract whose value tracks a named underlying variable, such as an interest rate, share price, commodity price or currency, which needs little or no money up front compared with buying the underlying outright, and whose settlement falls due later.

The laws, controls and supervision that keep criminal proceeds from being passed through the financial system to look legitimate: identifying customers, understanding their activity, monitoring transactions and reporting suspicion, to the international standard set by the Financial Action Task Force's Recommendations.

Interest worked out on the original sum and also on the interest already added to it, so the balance grows by a larger amount each period; the more often interest is credited and the longer the term, the stronger the effect.

An arrangement, grounded in law and shared technical standards, under which customers can let an authorised third party read their bank account data or start payments from their account through secure interfaces, without ever handing over their banking login details.

A pivot table is an interactive summary of a list of records: the user drops fields into rows, columns, values and filters, and the spreadsheet groups the records and aggregates them, so the same data can be regrouped without writing a single formula.

A cell reference is the address, such as B4, that a formula uses to point at another cell, and it comes in three forms: relative, which shifts when the formula is copied; absolute, which dollar signs pin in place; and mixed, which pins only the column or only the row.

Data validation is a rule attached to spreadsheet cells that limits what may be entered in them, for example whole numbers within a range, dates after a given day or one item from a list, and responds to anything else with a warning or a refusal.

A macro is a stored sequence of commands inside a document or workbook that replays a routine task on request, usually captured by recording the user's clicks and keystrokes and saved as code that can later be edited.

Mail merge combines one template document with a list of recipients, filling placeholders such as name and address from each row of the list, to produce a personalised letter, email, envelope or label for every person on it.

Track changes is a review mode in which a word processor records each insertion, deletion and formatting change as a proposal marked with its author, so that the document owner can accept or reject every edit individually instead of comparing two versions by eye.

A slide master is the top-level slide in a presentation file that holds the shared design, such as fonts, colours, background, logo and placeholder positions, and passes it down to the layouts and slides beneath it, so one edit changes every slide based on it.

Co-authoring is several people editing the same stored file at the same time, each seeing the others' presence and changes as they happen, instead of passing copies back and forth and merging them afterwards.

Version history is the record a cloud storage service or editor keeps of a file's earlier saved states, letting anyone with the right permission open, compare and restore a previous version without a separate backup copy.

A cloud flow is an automation that runs on a provider's servers rather than a user's computer: a trigger such as an arriving email, a button press or a schedule starts it, and a chain of actions then reads and writes data in other services through connectors.

In workflow and low-code tools, a connector is a packaged adapter around another service's API that exposes that service's triggers and actions as ready-made building blocks, so a flow or app can use it without anyone writing the underlying calls.

A sensitivity label is a classification tag, such as Public, General or Confidential, stored in a file's or email's metadata that travels with it and can switch on protection like encryption, watermarks or sharing limits wherever the item goes.