Insights · Glossary
The terms, in plain language.
What the words used in data, business intelligence, AI and decision systems mean, and where each one shows up in our work.
Glossary
Defined once, used the same way everywhere.
33 terms in five groups. Each definition stands on its own, and each links to the service or the layer of the stack where it applies.
Data foundations.
The layer every report, model and decision rests on.
Data engineering
The work of collecting data from the systems that produce it, cleaning and reshaping it, and delivering it in a form that reports, analysis and AI can rely on. It covers architecture, pipelines, storage, quality checks and the operations that keep them running.
In our work. Work on analytics or AI starts here, because a dashboard or a model is only as reliable as the data beneath it.
Data warehouse
A database organised for analysis rather than for running transactions. Data from finance, sales, operations and other systems is loaded into agreed structures, so the same question gets the same answer whoever asks it.
In our work. We build on the platform an organisation already runs, from SQL Server estates to cloud warehouses, with reconciliation against the source systems built in.
Data lake
Low-cost storage that holds data in its raw form, structured or not: files, logs, documents and sensor readings. It keeps everything, which suits data science, but without structure and governance it becomes hard to search and harder to trust.
Data lakehouse
An architecture that combines the open, low-cost storage of a data lake with the table structure, transactions and governance of a warehouse. One copy of the data serves reporting, analytics and machine learning.
In our work. We layer a lakehouse as bronze, silver and gold: data as it arrived, cleaned and conformed data, and business-ready tables that dashboards and models read.
ETL and ELT
Two ways to move data into an analytical platform. ETL extracts data, transforms it, then loads it. ELT loads it first and transforms it inside the platform, which suits cloud warehouses and lakehouses with plenty of compute.
In our work. Our pipelines are reusable and carry reconciliation checks, so a total in the warehouse can be traced back to the ERP or point-of-sale system it came from.
Data integration
Bringing data from separate systems, such as ERP, point of sale, CRM and spreadsheets, into one consistent view. The hard part is rarely the connection. It is agreeing which system is right when two of them disagree.
Data quality
How far data can be relied on for the use it is put to: complete, accurate, consistent, timely and free of duplicates. Quality is measured with rules that run automatically, not checked by hand before a board meeting.
In our work. Profiling, validation and monitoring run inside the pipelines we build, so a failure is visible before it reaches a dashboard.
Master data management
Also: MDM
The discipline of keeping one agreed record for the things a business depends on: customers, products, suppliers, stores and the chart of accounts. Without it, every system holds its own version and every report disagrees.
Data catalogue and lineage
A data catalogue lists what data exists, what it means and who owns it. Lineage shows where a number came from and every transformation it passed through. Together they let someone trust a figure without asking the person who built it.
Business intelligence and analytics.
From data to a view of performance that people accept.
Business intelligence
Also: BI
The practice of turning operational data into a shared, current view of how the business is performing, through agreed measures, reports and dashboards. Its value depends less on the tool than on whether people accept the numbers.
In our work. We build on Power BI, Tableau or Looker, on top of a governed model, so finance and operations read the same figure.
KPI dictionary
A written, agreed definition of every measure that matters: what it counts, the formula, the source, the owner and how often it updates. It ends the meeting that starts with two departments presenting different revenue numbers.
In our work. A KPI dictionary is one of the first things we produce, agreed with the people who own each measure.
Semantic layer
Also: semantic model
A layer between the data platform and the tools people use, where business terms and calculations are defined once. Every dashboard, spreadsheet and AI assistant that reads through it gets the same answer to a question such as what margin was last month.
Descriptive, diagnostic, predictive and prescriptive analytics
Four kinds of analytics. Descriptive reports what happened. Diagnostic explains why. Predictive estimates what is likely to happen next. Prescriptive recommends what to do about it, within the limits the business sets.
Self-service analytics
Letting business users explore data and build their own reports without waiting for a specialist. It works when the underlying model is governed. Without that, it multiplies the versions of the truth.
Conversational analytics
Asking a question of business data in plain language and getting an answer with its source, instead of building a report. It depends on a semantic layer, so that the words in the question map to agreed measures.
In our work. Assistants that answer over governed enterprise data are part of our AI work. In CogniGraph, natural-language querying is on the roadmap and is not a released feature.
AI and machine learning.
Intelligence that has to work in operations, not in a demo.
Machine learning
Software that learns patterns from historical data and applies them to new cases, for example forecasting demand or scoring risk. It produces estimates with a degree of confidence, not certainties.
In our work. Our models are trained on the governed data foundation, give confidence ranges, and are monitored once live.
Generative AI and large language models
Also: LLMs
Generative AI produces new text, code, images or summaries in response to a request. Large language models are the systems behind most of it. They are fluent by design, which is why enterprise use needs them tied to trusted sources.
Retrieval-augmented generation
Also: RAG
A way to make a language model answer from an organisation's own documents and data. The system first retrieves the relevant passages, then asks the model to answer using them, and shows where the answer came from.
In our work. Our language-model applications answer over enterprise data and documents with sources shown, and keep the choice of model provider open.
AI agents and agentic AI
An AI agent is software that pursues a goal through several steps: it gathers information, uses tools and systems, and prepares a result. Agentic AI describes systems built from such agents.
In our work. We build AI agents for decision support. They gather the evidence, apply business rules and prepare a recommendation for a person to act on, with a record of how they reached it.
AI & Machine Learning servicesIntelligent Decision Systems services
MLOps
The engineering practices that keep models working after launch: versioning, deployment, monitoring for drift, retraining and rollback. Many AI pilots that never reach operations fail here, not in the model.
AI governance
Also: responsible AI
The rules and controls that decide what an AI system may do, how confident it must be, when a person must review its output, and what is recorded. It is what lets an organisation explain an automated decision to a board or a regulator.
In our work. Automatic where policy allows, human judgment where it is required, escalation when confidence is low, and an explanation every time.
Decision intelligence.
Where data, analytics and AI meet the decision they were meant to improve.
Decision intelligence
A discipline that treats a business decision as something to design: the question being asked, the evidence it needs, the rules that apply, who is accountable and what action follows. Business intelligence shows what is happening. Decision intelligence carries that through to who decides, on what basis, and what happens next.
Intelligent decision system
A system that carries a recurring decision from question to action. It assembles the evidence, applies the rules, prepares a recommendation, routes it to the right person or approves it where policy allows, and records what happened.
Decision path
The route one decision takes: the question, the evidence, the rules, the recommendation, the approval and the action. Designing the path before automating it is how a decision stays explainable.
Knowledge graph
A model of how things relate: customers to orders, products to suppliers, measures to the data behind them, rules to the decisions they govern. Because the relationships are stored explicitly, a system can follow them to answer questions a table cannot.
In our work. CogniGraph is CogniverseAI's graph of how an enterprise's data, metrics, models, rules and people relate.
CogniGraph is in active development. The core graph and its API are in place; natural-language querying and the visual explorer are on the roadmap. We work with a small number of design partners.
AI orchestration
Coordinating several models, agents, data sources and rules so they work as one governed process, with confidence scoring and escalation to people. The value is in the coordination, not in adding more agents.
Human in the loop
A design in which a person reviews or approves an AI system's output before it takes effect, always or above a threshold. The person stays accountable. The system makes that accountability faster, consistent and explainable.
Governance, strategy and delivery.
The agreements and the plan that make the rest hold.
Data governance
The agreements that make data dependable: who owns each dataset, what each term means, who may see what, how quality is measured and how long data is kept. Good governance is mostly decisions about people and definitions, supported by tooling.
Data residency
Where data is physically stored and processed, and the rules that restrict moving it across borders. Data-protection rules and sector regulators in Jordan and the Gulf states affect where personal and regulated data may be hosted.
In our work. We build in the client's environment: their cloud, on-premise or hybrid, aligned to the data-residency requirements that apply in Jordan and the Gulf.
Data and AI strategy
A plan that ties data and AI investment to the decisions and measures a business cares about: the current state, the use cases worth doing, the target architecture, the operating model and the sequence.
In our work. Ours is set out in three horizons, crawl, walk and run, and is built by the team that then delivers it.
AI readiness assessment
Also: maturity baseline
An honest baseline of how ready an organisation is to use data and AI: the state of its data, platforms, skills and governance, and the way decisions are made today. It shows what has to be true before a use case can work.
Strategy & Transformation servicesAI & Machine Learning services
Vertical slice
A thin, working piece of the whole system, from source data to a decision someone uses, built first to prove the architecture on real data before the rest is built.
Strategy & Transformation servicesData & Engineering services
Capability transfer
Ending an engagement with the client's own team able to run what was built. Documentation, training and handover are part of the work, not an extra.
Ready to turn enterprise intelligence into better decisions?
Let's identify where data, AI and orchestration can create measurable value in your organization.
01
Send a message
Tell us what you are trying to decide and what your data looks like today.
02
A 60-minute conversation
Where your decisions are made today, what they rest on, and where the gaps are.
03
A written summary you keep
What we heard, what we would look at first, and what an engagement could look like.