The Thrillworks Digital Delivery Glossary

The software a team logs into to create, edit and publish content without needing a developer. Everything else in this section is a variation on that idea.

A CMS that stores and organises content but does not produce the finished web pages. It hands content out through a data feed, and a separate application turns that into a website, an app, a kiosk screen or anything else.

The name is unfortunate. "Headless" sounds like something is missing. What is actually missing is the assumption that content exists to become one specific web page.

A CMS where the content storage and the page rendering are the same system. WordPress and Drupal in their default configuration are the familiar examples. Simpler to start with, harder to reuse content beyond the website it was built for.

An approach where a digital platform is assembled from independent, replaceable services rather than bought as a single suite. The CMS handles content, a separate service handles search, another handles commerce, and each can be swapped without rebuilding the others.

A vendor category describing a suite that bundles content management with personalisation, analytics, commerce and campaign tools. Whether a DXP is a product or a marketing category depends largely on who is selling it to you.

Content stored as separate labelled pieces of information rather than as one blob of formatted text. An article stored as headline, author, publish date, body, topic, related items is structured. The same article pasted into a rich text field is not.

The complete set of rules describing every kind of content a platform holds: what types exist, what information each contains, which pieces are required, and how they relate to each other.

One kind of thing in the content model. Article, Event, Person, Product, Location, Policy Document.

One labelled piece of information inside a content type. Publish date, alt text, summary, hero image.

One actual item of a content type. A specific article. A specific event.

A link from one entry to another, so information can appear in many places without being duplicated. An event references a venue rather than restating the address.

The organised set of labels and categories used to classify content so it can be filtered, grouped, related and found.

Information about a piece of content rather than the content itself: its title for search, its description, its language, its author, its publish date, its owner.

A field that allows formatted text: bold, links, lists, headings. Useful and dangerous in equal measure, because it is where structure goes to die when a content model is too loose.

A file managed by the platform: image, video, document, audio.

A dedicated system for storing, organising, rights-managing and distributing media at scale. Distinct from the basic media library inside a CMS.

The principle that a piece of content should be authored a single time and delivered to every channel that needs it, rather than rewritten per destination.

Delivering a coherent experience across every channel a person might use: web, app, email, in-store screen, voice, kiosk. Frequently claimed, rarely achieved, and almost entirely dependent on whether the content was structured in the first place.

The people, processes and tooling that produce and maintain content over time. The unglamorous discipline that determines whether a platform stays good after launch.

A systematic review of the content that exists: how much, what kind, what condition, what is current, what is duplicated, what should not survive the move.

The resulting list. Even a rough count is enormously valuable, because it converts the largest unknown in a replatform into a range.

Moving existing content into a new system, including mapping every old address to a new one.

Older content where the words and the layout are tangled together in raw markup rather than stored as separate labelled fields. The main driver of migration cost.

An agreed date after which editors stop updating the old system, so content stops moving underneath a migration.

Deciding deliberately what content is carried forward, what is preserved somewhere quieter, and what stops being published at all.

The accumulated cost of content that is out of date, duplicated, unowned or structurally wrong. Like technical debt, it compounds quietly and gets paid in a lump.

The principle that any given piece of information lives in exactly one place, and every appearance of it elsewhere is a reference back to that place.

The part people see and use.

The systems behind it: data storage, business logic, integrations.

A structured way for two systems to exchange information automatically, without a person copying anything between them.

Two common styles of API. REST returns a fixed shape of data per address. GraphQL lets the requester specify exactly which fields it wants in a single call. GraphQL tends to suit content-heavy applications with many different front ends.

An automatic notification sent from one system to another when something happens, for example telling a website to rebuild a page the moment an editor publishes.

Custom software built to sit between two systems that cannot talk to each other directly, usually because one of them was never designed to be integrated with.

Connecting two systems so they exchange information. Worth specifying the direction, the frequency, the system of record and what happens when it fails, because "integration" on its own commits to none of that.

The single system that holds the authoritative version of a given kind of data. Getting this explicit for every data type prevents the most common and most expensive class of integration argument.

Building every page in advance as a finished file, so a visitor is served something that already exists rather than something assembled on request.

Assembling the page on the server at the moment it is requested. Necessary when content is genuinely personalised or changes by the second.

A hybrid: pages are pre-built, but individual pages can be rebuilt on a schedule or on demand without rebuilding the whole site. In practice this is where most large content sites land.

An architectural pattern built around pre-rendered front ends, APIs and managed services rather than a traditional application server. The term has faded, the pattern has not.

A global network of servers holding copies of your site close to your visitors, so pages load quickly and traffic spikes hit the network rather than your origin.

Computing that happens on that distributed network, physically near the user, rather than in one central location. Useful for redirects, personalisation, access control and localisation without a round trip to origin.

The authoritative server the CDN pulls from.

Storing a copy of something so it does not have to be produced again, and the far harder problem of knowing when that copy is stale. Most "why hasn't my change appeared" support tickets are cache invalidation questions.

A separate copy of a platform for a particular purpose. Typically development for building, QA for testing, staging for final review, and production for the live site.

The automated pipeline that tests changes and moves them safely between environments, so releasing is routine rather than an event.

Defining servers, networks and services in version-controlled files rather than by configuring them by hand. Makes environments reproducible and changes reviewable.

The version-controlled store where a project's code lives, with a complete history of every change.

Work deferred to move faster now, which accrues interest. Sometimes a deliberate and correct trade. Dangerous when it is unrecorded, because then it is not a decision, it is a surprise.

One large application that does everything, versus many small services that each do one thing. Both are legitimate. Microservices trade internal simplicity for operational complexity, which is a good trade only if you have the operational capacity to pay for it.

One shared platform serving several distinct organisations or brands at once, where each sees only its own material and cannot affect anyone else's.

One organisation inside such a platform.

A multi-tenant pattern where a central property shares content outward to connected properties, each of which keeps control of its own local material.

When a connected property automatically displays content published centrally, without anyone copying it. Update the source and every property showing it updates too.

When a connected property replaces part of an inherited item with something local, while the rest continues to come from the source.

A term used for two genuinely different things, which is the cause of more scope disputes than any other word in digital delivery. The two definitions follow.

The look: brand expression, colour, typography, photography direction, spacing, page compositions. Usually owned by a brand or creative partner.

The working parts: every reusable building block, its variants, its states, its responsive behaviour, its accessibility requirements, what an author is permitted to assemble with it, and what may be customised.

One reusable building block of an interface. Pages are assembled from components rather than designed one at a time.

The documented catalogue of available components, so designers and engineers work from the same set.

An alternative version of the same component. A card with an image and one without.

How a component looks and behaves in different conditions: default, hover, focus, active, disabled, loading, error, empty.

Named values (colour, spacing, type scale, radius) held in one place and referenced everywhere, so a change propagates rather than being hunted down.

A way of organising components from smallest to largest: atoms, molecules, organisms, templates, pages. A useful mental model. Not a law, and not worth arguing about.

A deliberately plain layout showing what goes where and in what priority, before any visual treatment.

A finished visual design of a specific page or screen. A set of comps is not a component system, and treating them as interchangeable is the most expensive misunderstanding in this section.

An interactive version used to test flow and behaviour before it is built.

The point at which design work transfers to engineering. Better thought of as an ongoing collaboration than a moment, because the questions that matter tend to arrive after the files do.

Building so one experience adapts across screen sizes rather than maintaining separate mobile and desktop products.

Building a baseline that works everywhere, then layering on capability for environments that support it. Deeply unfashionable and consistently correct.

What it actually feels like for a person to create and publish content in the system.

The whole of a person's experience of a product, not just its interface. Includes how easy it is to find, understand, complete and recover from.

The specific screens, controls and interactions a person uses.

How information is organised, labelled and structured so people can find things and understand where they are.

The path a person takes to accomplish something, across whatever channels and systems that requires.

A described archetype of a user group, based on research, used to keep decisions anchored to real needs rather than internal assumptions.

A framing that focuses on what someone is trying to accomplish rather than who they are demographically. Often more useful than personas for prioritisation.

Watching real people attempt real tasks. The cheapest way to be proven wrong before it becomes expensive to be wrong.

Research methods for validating information architecture. Card sorting learns how people group things. Tree testing checks whether they can find things in a proposed structure.

An expert review against established usability principles. Fast and cheap. A complement to user research, not a substitute.

Designing the whole service including the parts the customer never sees: the staff process, the systems, the handoffs. Where digital problems often turn out to be operational problems.

Designing and building so people with disabilities can use what you make. Roughly one in five people, which makes it a mainstream requirement rather than an edge case.

The Web Content Accessibility Guidelines, the international standard. Organised into three levels: A, AA and AAA. AA is the level normally required of public bodies and referenced in law.

Regional accessibility legislation: Ontario, the United States and the European Union respectively. Each references WCAG rather than restating it.

Software that reads a page aloud and lets a person navigate it without seeing it. The most common assistive technology, and the one most often broken by well-intentioned design.

Writing markup that describes what each part of a page is, not just how it looks. A heading marked as a heading rather than as large bold text. The foundation of everything else in accessibility.

A written description of an image, read aloud by screen readers and shown if the image fails to load.

The measurable difference between text and its background. One of the most commonly failed and most easily fixed accessibility requirements.

The visible indication of which element the keyboard is currently on. Removing it because it looks untidy is one of the most common accessibility failures in modern design.

A third-party script that claims to make a site compliant automatically.

Designing for the full range of human variation from the outset, rather than designing for a notional average user and remediating afterwards.

The work of making content findable through search engines, covering technical health, content quality and authority.

The infrastructure side: crawlability, indexation, site speed, structured data, canonical rules, sitemaps, redirects.

The equivalent discipline for AI assistants and AI-generated answers, where a person may receive a response without visiting your site at all.

A search that is answered on the results page, with no visit to a source. Already the majority of searches for many query types.

A distinct thing a search or AI system recognises: a company, a person, a product, a concept. Search engines increasingly reason about entities and their relationships rather than matching keywords.

Machine-readable labelling embedded in a page that states what the content actually is: an article, an event, a product, a defined term. The most direct way to tell both search engines and AI systems how to interpret you.

The declared authoritative address for a page that may be reachable at several addresses. Prevents versions of the same page competing against each other.

Markup declaring which pages are the language equivalents of each other, so the right version is shown to the right person and the versions do not compete.

An instruction that sends a visitor from an old address to a new one. A 301 declares the move permanent, which is what allows the page to retain accumulated search value.

The complete list matching every old address to its new one. The single biggest factor in whether a site keeps its search performance through a replatform.

The accumulated authority a page has built up over years. Real, fragile, and lost by default when a migration is handled carelessly.

Google's measures of loading performance, interaction responsiveness and visual stability. Both a ranking input and a reasonable proxy for whether a site feels good to use.

An emerging convention: a plain file at the root of a site that gives AI systems a curated, structured guide to its most useful content.

The measurement of how people use a digital product. Useful in proportion to how clearly you decided what to measure before you started.

Recording specific user actions rather than only page views. What people actually do, rather than only where they went.

The documented definition of every event, its properties and its purpose. The thing that separates analytics you can trust from a dashboard nobody believes.

Assigning credit for an outcome across the touchpoints that contributed to it. Genuinely difficult, frequently oversimplified, and worth treating as directional rather than precise.

A system that unifies customer data from multiple sources into a single profile, usually to power personalisation and segmentation.

Dividing an audience into groups with different behaviours or needs so they can be served differently.

Adapting what someone sees based on who they are, what they have done or where they are.

Showing two variants to comparable audiences to learn which performs better. Requires enough traffic to reach significance, which is a bar many sites quietly fail to clear.

Information collected directly from your own audience with their knowledge. Increasingly the only durable kind.

Central stores for analytical data. A warehouse holds structured, modelled data. A lake holds raw data of any shape. Most organisations end up with both and a strong opinion about it.

Authentication proves who someone is. Authorisation determines what they may do. Conflating them is the origin of a large share of access-control bugs.

Signing in once and being recognised across multiple connected systems.

The system holding the accounts and performing the authentication.

Granting permissions according to a person's role rather than individually. The only approach that stays manageable past a handful of users.

Information that identifies a specific person. Subject to legal obligations that vary by jurisdiction and are usually stricter than teams assume.

The requirement that data be physically stored within a particular country or region. A common procurement blocker, and one worth resolving early because the answer shapes architecture.

Protecting data both while stored and while moving across a network.

Independent audits of a vendor's security practices and controls. Standard evidence requests in enterprise procurement.

Authorised simulated attack against a system to find weaknesses before someone else does.

The plan for restoring service after a serious failure. RTO is how quickly service returns. RPO is how much recent data could be lost in the worst case. Both should be decided by the business, not assumed by the vendor.

A durable record of who did what and when. Distinct from version history, which tracks the content rather than the actor.

Capturing and honouring people's choices about tracking and data use, in line with applicable law.

The opening phase, spent understanding the organisation, the users, the constraints and the content before anything is built.

Agile is a set of principles. Scrum and Kanban are specific practices. Many organisations adopt the ceremonies without the principles and conclude that agile does not work.

A fixed period, typically two weeks, in which a defined amount of work is completed and reviewed.

The prioritised list of everything wanted but not currently scheduled. The mechanism by which a request is sequenced rather than refused.

The smallest version that delivers real value and generates real learning. Frequently misused to mean "the version we could afford", which is a different and less useful idea.

The agreed standard a piece of work must meet before it counts as complete. Prevents the slow accumulation of things that are finished except for the parts that are not.

A live register of Risks, Assumptions, Issues and Dependencies, reviewed regularly.

Something the work requires from outside itself in order to proceed.

What is included, stated precisely enough that everyone can tell what is not.

The documented process for adding to scope, with cost and schedule impact made visible before approval rather than discovered afterwards.

The single named person empowered to make decisions on behalf of the organisation without convening a committee.

The recurring senior forum where direction is set and escalations land.

The phase where the organisation's own people use the system on real tasks and confirm it does what they need.

Re-testing existing functionality after a change, to confirm that fixing one thing did not break another.

The moment the new system becomes the live one.

The defined period immediately after launch for finding and fixing what only appears under real use. A phase, not an afterthought.

The deliberate handover of understanding, not just documentation, so the organisation can operate and evolve what was built.

A way of expressing how much of a person's time a role requires. 0.5 FTE is roughly half a working week.

The document defining scope, deliverables, assumptions, exclusions, timeline and commercial terms for an engagement. The exclusions section is the one worth reading twice.