The Data, Analytics and AI Glossary

Writing down, in plain business terms, what your core measures mean - an active customer, net revenue - so every number agrees. It is the least technical and most valuable part of getting data right.

Why it matters: Most reporting disputes are really arguments about definitions nobody ever wrote down.

An AI that does not just answer but takes actions on your behalf, stringing several steps together towards a goal. It is the direction much of AI is heading, and the point at which governance stops being optional.

Why it matters: The more an agent can do, the more it matters what it is allowed to touch.

The rules and oversight that keep AI use safe, fair and accountable: what it can access, how its answers are checked, and who is responsible. As AI does more, this stops being optional.

Why it matters: The businesses that adopt AI well are the ones that set the guardrails first, not after.

A Business Central feature for slicing transactions by dimension inside the system, without a separate reporting tool. Useful for quick views close to the ledger.

Turning data into answers you can act on. It spans a ladder, from describing what happened, through explaining why, to predicting what comes next and recommending what to do. Most businesses live on the bottom rung and have far more to gain higher up than they realise.

Hopton's way of delivering analytics in three stages: getting the foundations right, proving value quickly on a real decision, and handing over a platform your own team can run. It is built to make you capable, not dependent.

Why it matters: It is deliberately the opposite of a retainer you can never leave.

Spotting values that do not fit the normal pattern, which can flag errors, fraud or genuine change early. It watches the data so people do not have to.

An engine for processing very large amounts of data quickly across many machines. It sits under platforms like Microsoft Fabric and Databricks, and matters when your volumes are truly large.

Why it matters: It is a heavyweight tool; needing it is a sign of real scale, not a default.

A broad term for software that performs tasks we associate with human intelligence, such as recognising patterns or understanding language. Most useful business AI today is machine learning, or the assistants built on it.

Microsoft's data movement and orchestration service, used to schedule and run the flows that move and transform data, often into a warehouse. Its capabilities also live inside Fabric.

A managed cloud database from Microsoft, often the right-sized warehouse for a mid-market business that does not yet need Fabric. Reliable, well understood, and inexpensive at modest scale.

Why it matters: Plenty of businesses reach for a big platform when a clean Azure SQL database would have done the job for far less.

Microsoft's previous-generation analytics platform, much of which has been folded into Microsoft Fabric. You will still meet it in established estates.

Finding which products are bought together more often than chance would predict, to inform ranging, merchandising and promotions. It works on the till and order data you already keep.

Why it matters: The skill is telling a real pattern from a coincidence; bread and milk sell together because everyone buys both.

Comparing a number against something meaningful: a peer group, an industry norm, or your own past, so you can judge whether it is good, bad or about right. A figure on its own rarely tells you much; a figure in context does.

Why it matters: Without a benchmark, every number is just a number, and targets get set by gut feel.

Data that is too large, too fast-moving or too varied for ordinary tools to handle. The phrase is less fashionable than it was, and most mid-market businesses have a tidy-up problem rather than a big data problem.

Why it matters: Most businesses reaching for big data tools would get further by getting their ordinary data in order first.

Microsoft's ERP for small and mid-sized businesses, running finance, sales, purchasing, inventory and operations. It is where many mid-market businesses create the data they later want to report on.

Why it matters: Reporting on it well is a particular skill, because the data is structured for running the business, not for analysis.

Turning business data into reports, dashboards and answers that support decisions. For most mid-market businesses, in practice, it means Power BI sitting on a clean model.

Why it matters: BI tells you what is happening and why; predicting what comes next is where analytics and AI take over.

Two ways to add calculations in Power BI. A calculated column is worked out row by row and stored; a measure is worked out on the fly as you slice and filter. Choosing the wrong one is a common cause of slow, bloated models.

Marking a model or dataset as trusted and endorsed, so people know which version to build on. It is how you stop the spread of rival copies.

The structured list of accounts a business records its finances against. Consolidating several companies means lining these up first.

Why it matters: Mismatched charts of accounts are the first and largest job in any multi-company consolidation.

A model that scores which customers are drifting away, learned from the history of those who left and those who stayed, so you can act before they go.

Why it matters: Keeping a customer is far cheaper than winning one, and a few weeks of warning is usually enough.

Two common kinds of prediction. Classification puts something into a category; regression predicts a number.

Microsoft's AI assistant built into Power BI and other tools, which summarises, drafts reports and answers questions in plain language. It is helpful, and only ever as good as the model beneath it.

Why it matters: Point it at a clean model and it shines; point it at a mess and it produces confident, wrong answers.

Recorded facts about your business: sales, customers, orders, stock, the lot. On its own it is raw material, neither useful nor useless until something is done with it, and most businesses are sitting on far more of it than they use.

Why it matters: You are almost never short of data. You are short of data that is clean, joined up and trusted.

The agreements and controls that make data trustworthy: what terms mean, who can see what, and how a number can be traced. It is what separates data people trust from data they argue about.

Why it matters: Governance is not bureaucracy for its own sake; it is the difference between confidence and chaos.

A large, cheap store that holds raw data of any shape, before it is cleaned and structured. It is where data can land first and be refined later.

The traceable path from a number back through the model to its source, so any figure can be explained and defended.

Why it matters: An answer you cannot trace is an answer you cannot defend, which is a problem the moment AI starts producing them.

A smaller, focused slice of a warehouse, built for one team or subject area. It gives a department what it needs without the whole warehouse.

How far along a business is in using data well, from spreadsheets and gut feel to governed, predictive analytics. Knowing where you sit decides the right next step.

Why it matters: Most failed projects tried to skip a stage; maturity is the map that stops you doing that.

Sifting large amounts of data for patterns and relationships that are not obvious, often as a first step towards predictive work.

Structuring data so it is easy and fast to report on: deciding the tables, the relationships and the definitions. It is the difference between a model that is a pleasure to work with and one that fights you.

Why it matters: Most reporting problems are modelling problems wearing a different hat.

An automated flow that moves and transforms data from a source into your warehouse or lakehouse, on a schedule, without anyone exporting and pasting.

Whether your data is accurate, complete, consistent and current. Poor quality is the single most common reason analytics and AI projects disappoint.

Why it matters: AI does not fix poor data, it amplifies it - and does so with great confidence.

Where your data is physically stored, which can matter for legal, contractual and compliance reasons.

A short, plain plan for how data will earn its keep: what matters, in what order, and how it will be governed. Best kept to a single page.

Why it matters: A strategy nobody reads changes nothing; brevity is what makes it real.

Presenting data as charts, maps and visuals so a pattern can be seen at a glance rather than read from a table. Good visualisation makes the point; poor visualisation buries it.

Why it matters: The aim is the fastest path to understanding, not the most impressive-looking chart.

A database built and structured specifically for reporting and analysis, as opposed to the systems that run the business day to day. It brings sources together and shapes them for fast, consistent reporting.

A data and AI platform built around large-scale engineering and data science. It is strong at what it is built for, and for most mid-market businesses it is more platform than the work requires.

A lower-code way to pull and shape data in the Microsoft stack, often used to feed a model without building full pipelines. A useful middle ground for smaller needs.

The formula language used to write measures and calculations in Power BI. It is powerful and famously easy to get subtly wrong, which is why calculations belong in one governed model rather than scattered across reports.

Why it matters: Two people writing the same measure slightly differently is how a business ends up with two versions of revenue.

A kind of machine learning using layered networks loosely inspired by the brain, behind much of modern AI including language and image tools. It is powerful and data-hungry.

A storage format that brings reliability and structure to a data lake, underpinning Microsoft Fabric and Databricks. It is plumbing you rarely deal with directly.

Reporting on what has already happened. It is the most common kind of analytics and the place nearly every business starts, and done well it is valuable in its own right.

Going a step beyond description to understand why something happened, usually by slicing and comparing until the cause shows up.

The descriptive context you slice your numbers by: customer, product, region, date. Dimensions are the 'by' in sales by region by month.

Tags such as department, project or region applied to transactions in Business Central, used to analyse the numbers by category later.

A Fabric feature that lets Power BI read large data straight from OneLake at speed, without importing it or running slow live queries. It combines the speed of import with the freshness of live data.

Why it matters: It is one of the real reasons a larger business might move to Fabric, rather than the brochure.

Two ways Power BI gets at data. Import copies it into the model for speed; DirectQuery leaves it in the source and queries it live. Import is usually faster and the right default; DirectQuery suits very large or very current data.

Microsoft's family of business applications, of which Business Central is the ERP for small and mid-sized businesses.

The core system that runs the business: finance, sales, purchasing, stock, operations. For many mid-market firms that system is Business Central.

The three stages of the Analytics Acceleration Programme. Establish gets the foundations and definitions right; Build proves value on a real decision; Continuity hands ownership to your team.

Extract, transform, load: the work of getting data out of source systems, cleaning and shaping it, and landing it where reporting can use it. ELT simply reorders the steps. Either way it decides whether anything downstream is trustworthy.

Why it matters: It is the least glamorous and most important part of any analytics build, and the first thing skimped under time pressure.

The unit you buy Microsoft Fabric in, sized from small (F2) upwards. From F64 and above, report viewers can read content without their own paid licence - which is the threshold that often decides the cost.

Why it matters: Sizing the capacity wrong, in either direction, is one of the most expensive mistakes in the Microsoft data world.

The Fabric feature that lets people ask questions of governed data in plain language and get answers back. The value is the trusted model underneath, not the chat box on top.

The table holding the events you measure, such as sales lines or transactions. It is usually the largest table in the model and the thing your measures add up.

Business Central's built-in tool for building financial statements from the chart of accounts. Good for ledger-accurate finance views, limited beyond them.

Why it matters: It does finance statements well and was never meant to carry the whole business's reporting.

Projecting what is likely to happen, such as demand or cash, from patterns in your history, and expressing it as a range rather than a single figure. It runs on the sales history you already keep.

Why it matters: A range you can plan against beats a single number that is always wrong by some amount.

AI that creates new content - text, images, code - rather than predicting a number or category. The assistants most people now mean by AI are generative.

Why it matters: It is fluent and confident, which is both its strength and its risk; it needs grounding in your own trusted data.

When an AI gives a confident answer that is simply wrong. The risk is highest when it is pointed at ungoverned, missing or contradictory data.

Why it matters: The danger is not that it is wrong - it is that it is wrong fluently, which makes people trust it more.

Refreshing only the data that has changed rather than reloading everything, which keeps large models fast and cheap to update.

Trading between your own companies, which has to be identified and removed before group figures make sense, or you count the same money twice.

Why it matters: Forget it and the group looks bigger and more profitable than it is, which auditors notice.

The handful of numbers that show whether the business is on track. The skill is keeping the list short, because a hundred indicators is the same as none.

Why it matters: Most dashboards track far too much; the value is in choosing the few that change a decision.

A store that combines the flexibility of a data lake with the structure of a warehouse, suited to larger or semi-structured data. It is a core building block of Microsoft Fabric.

The kind of AI behind assistants like Copilot, trained on vast amounts of text to understand and generate language. It is excellent with words and is not, by itself, a source of truth about your business.

Learning patterns from your history to say something useful about what happens next. It is prediction from data, not reasoning, and it learns from the records you already keep.

Why it matters: You usually do not need new data or a data scientist to start - you need a clear question your data can answer.

The core reference data of the business, such as the definitive list of customers or products, kept consistent so everything else lines up.

Why it matters: Much of what looks like a reporting problem is really a master data problem in disguise.

A calculation in the model, such as gross margin or year-to-date sales, defined once and reused everywhere it is needed. Agreeing your measures is most of what makes reporting trustworthy.

A common way to organise data in layers, from raw (bronze) through cleaned (silver) to ready-for-reporting (gold), so quality improves at each stage. It keeps a build orderly and traceable.

Any number you measure. A KPI is a metric that matters enough to act on; the great majority of metrics are not KPIs and never should be.

Microsoft's unified data platform, bringing a lakehouse, warehouse, data engineering, real-time and Power BI together on one capacity and one bill, inside your own tenant. It is where Microsoft is heading, which is not the same as where every business needs to be today.

Why it matters: Adopting it early, before the work demands it, is a common and expensive mistake.

In AI, the thing that has learned the pattern and produces the prediction. It is built once from history, then used again and again, and kept up to date as the world moves.

An emerging standard for connecting AI assistants to your tools and data in a controlled way. It is how agents are given safe, governed access to act.

Asking questions of your data in ordinary words and getting an answer, rather than building or reading a report. It depends entirely on a clean semantic layer to be trustworthy.

Why it matters: The demo always impresses; what decides whether it is safe is the model nobody demos.

The branch of AI concerned with understanding and generating human language. It is what lets you ask questions in plain words and have software understand them.

Hiding whole tables or fields from people who should not see them, as opposed to hiding individual rows. It works alongside row-level security.

A small piece of software that lets Power BI in the cloud reach data that still lives on your own servers, securely. It is the bridge between on-premises systems and cloud reporting.

Fabric's single storage layer, where data lives once and every Fabric tool reads from the same copy, rather than each keeping its own.

Microsoft's reporting and analytics tool. It connects to your data, builds a model with agreed definitions, and turns that into reports people can read and explore. For most mid-market Microsoft businesses it is the natural choice.

Why it matters: It is only ever as good as the model beneath it, which is where the real work sits.

The two halves of Power BI. Desktop is the free Windows application where reports are built; the Service is the cloud where they are published, shared and refreshed.

The capacity-based way of buying Power BI, now largely delivered through Microsoft Fabric capacities. It changes the licensing maths once you have enough viewers.

Why it matters: The crossover point is exactly where buying it right saves real money, and getting it wrong wastes it.

The standard per-user Power BI licence, needed to create content and, below an F64 capacity, to view shared content. It is the default licence for most users.

Why it matters: The maths between licensing everyone on Pro and buying a capacity is exactly where money is won or lost.

Microsoft's low-code family - Power BI, Power Apps, Power Automate and Power Pages - for reporting, building apps and automating work without heavy development.

The tool inside Power BI for cleaning and shaping data before it reaches the model. It is the unglamorous step that decides whether the model is sound.

Using your history to say what is likely to happen next. This is where machine learning starts to earn its keep, and where data you already hold becomes forward-looking.

Why it matters: A few weeks of warning is usually enough to change the outcome, which a backward-looking report can never give you.

A higher per-user licence that adds enterprise features for an individual without the business buying a full capacity. A useful middle option.

The furthest rung: not just what will happen, but what to do about it. It pairs a prediction with a recommended action.

The instruction or question you give an AI assistant. Clearer, more specific prompts get noticeably better answers.

A decision intelligence platform that brings data preparation, analytics and data science together in one governed, no-code environment. In March 2026 it was acquired by ServiceNow, which is folding its semantic and modelling layer into the ServiceNow AI platform.

Why it matters: As Pyramid Partner of the Year, Hopton can give a balanced view on what the new direction means for existing users.

The link between tables that lets the model combine them, so a sale knows which customer and product it belongs to. Relationships are what turn separate tables into a model.

A report is a set of pages you explore and drill into; a dashboard is a single summary view of the headlines. The distinction helps you design for the audience.

Giving an AI assistant your own trusted documents or data to answer from, rather than relying only on what it learned in training, so answers are grounded in your facts.

A simple, robust way of segmenting customers by how recently and how often they buy and how much they spend. It is often the best first step into predictive work.

A rule that limits which rows of data each person can see, tied to their identity, so one report shows each user only what they are allowed to.

Why it matters: Get it wrong and a friendly report becomes the quickest route to someone seeing what they should not.

Letting people answer their own questions from a trusted model, rather than queue for the data team. It works well when the model underneath is governed, and descends into chaos when it is not.

Why it matters: Self-service on an ungoverned estate just multiplies the number of conflicting versions of the truth.

The shared layer where your data is modelled and each measure is defined once. It is the real asset in any Power BI estate, because every report, and every AI assistant, reads from it.

Why it matters: The dashboards are disposable; the model is the thing worth investing in and protecting.

One agreed place each number comes from, so two people asking the same question get the same answer. In practice it usually means one governed model that every report and assistant reads from.

Why it matters: Most reporting disputes are not about the data - they are about there being several versions of it.

A cloud data warehouse known for scaling storage and compute separately and working across clouds. Like Databricks, it answers a problem of scale that most mid-market businesses do not yet have.

A way of structuring data for reporting, with fact tables in the middle surrounded by dimension tables. It is the foundation of a model that is both fast and easy to understand.

Why it matters: Most slow, confusing models are slow and confusing because they ignored this simple shape.

Structured data sits in neat rows and columns, like a sales ledger. Unstructured data does not, like emails, documents, photos or call notes.

Two broad styles of machine learning. Supervised learning trains on examples with known outcomes; unsupervised learning finds structure without being told the answer.

The history a model learns from. For most businesses this is the sales, service and order data already sitting in their systems, not something new to gather.

Why it matters: You have been collecting the training set for years; the barrier is rarely volume, it is whether it is clean and joined up.

The rules governing personal data in the UK. They apply in full to anything AI does with personal data; the technology gets no exemption.

Why it matters: The most common and avoidable breach is people pasting business or client data into public AI tools.

A store designed to find things by meaning rather than exact match, which is how AI assistants retrieve the right information to answer from.