Pyramid Glossary of Terms

Decision Intelligence

Gartner: Decision intelligence is a practical domain framing a wide range of decision-making techniques bringing multiple traditional and advanced disciplines together to design, model, align, execute, monitor and tune decision models and processes. Those disciplines include decision management and decision support as well as techniques such as descriptive, diagnostics and predictive analytics.

Pyramid Extension: It’s the fusion of 3 core disciplines: data preparation; data science and business analytics, to allow a system to offer business users the ability to make data driven decisions. It encompasses many technologies, techniques and functionalities, all with the ultimate goal of driving a decision on any business problem using data.

Descriptive Analytics

Analytical content that shows what happened using data visualizations.

Diagnostic Analytics

An analytical experience, using the same semantic models for descriptive analytics, where a user can slice, dice and analyze the reasons for outcomes and why something happened.

Predictive Analytics

Use and analysis of values derived from predictive models that have been produced using advanced statistical methods and software to generate likely values for a data set. This analytical stage is focused on defining outcomes that COULD happen based on statistical and mathematical models that are machine-trained using historical data and other inputs.

Prescriptive Analytics

Use of tools that take historical and predictive data and values with logical formulations of a business problem (aka business models) to drive decision options to the problem with mathematical tools like optimization (aka decision models). This analytical stage is focused on determining what SHOULD happen based on historical data, business inputs and constraints.

Data or
Semantic Model

A description of an analytical structure of a data set to drive querying. Includes details on the ‘joins’ between data structures; the data types and attribute types of the data structures; the enumeration of measures and their aggregations; the logic and structure of hierarchies.

Semantic models are built on data sets or data bases. In Pyramid, they are a virtual logical definition and are separate from the data store.

Machine Learning or Predictive Model

A predictive model is a mathematical formula that is derived (or “trained”) using specialized software that interprets the interconnection between data values in a data set using statistical mathematics. The formula is then used against real data to derive values. These values are also called “predictions”.

The values are typically consumed by downstream technologies like analytic models, business models, decision models or other applications.

Analytic Model

The collection of selections, measures, calculations and visualization specifications to draw a visual representation of the data. This includes the logic of the calculations and other selection criteria like filters, sorts etc.

Analytic models are built on semantic models, which ultimately reflect the underlying data structures.

Pyramid uses analytic models to differentiate from semantic/data models, because the former describes the ingredients for a specific report, while the latter defines the structure of a dataset.

Business Model

Business models reflect business processes and logic. They do not necessarily reflect a specific data structure, data base or data set.  But they are almost always driven off data queries or query fragments. They are usually expressed in a business-user centric tool like a spreadsheet.

Business models SHOULD extend analytic models by allowing users to easily connect real (live) data into the more fluid process of business modelling.

Decision Model

An extension to the business model, the decision model is an elevated form that produces a prescribed conclusion from a business model. Decision models often involve the use of data science technologies, optimization math and algorithms to solve complex models to derive the prescription.

While all decision models are business models, not all business models are decision models.

Visualizations

Data Visualization is the modern term for building a specific graphical view of data, extracted by a query against a data source. It includes all the usual techniques for drawing data, like grids, column charts and pie charts. The modern era usually allows users to manipulate every aspect of the visualization from its size, shape, colors and tooltips through to fonts, borders and background.

AI

AI (Artificial Intelligence) is a collection of tools, seen and unseen, that are built into the factory to “automagically” help users operate all the different modules and their capabilities. The emphasis on AI functionality is to support functions that are complex or very time consuming. NLQ and Smart Insights are good examples of AI.

The AI tools are based on a variety of technologies, including custom heuristics, classic machine learning algorithms and natural language processing technologies.

Data Science
(DS/ML)

Although all aspects of using data can be grouped under ‘data science’, the more modern definition involves the process of using software and statistical mathematics to reverse engineer reality captured in data through “machine learning”; to mathematically reflect this reality with a formula or model (“training”); and, then to use this formula to recreate this formulaic reality in other venues or with other data sets (“predict”).

Unlike AI, DS/ML is a set of tools for users to build their own models and apply them on their own data sets. AI, as described, may use similar techniques, but it's used to HELP users use the software itself.  “AutoML” – which is the highly automated version of DS/ML is where AI is being used solve DS/ML operations.

No-code

The user is not required to manually enter any logic at all to build and use any of the various models: data/semantic, predictive, analytic, business or decision; and any associated calculations and logic. Given the wide variety of such models and the sophistication of different problems, achieving a pure no-code experience is extremely difficult. Usually no-code techniques come at the expense of bespoke logic and functionality.

Pyramid has extensive modeling options and nearly all of them have no-code coverage – including complicated activities. Competitors offering no-code are generally only able to cover very basic activities.

Low-code

The user is expected to provide light logical inputs to the various models. Typically, needed for calculations, low-code experiences entail easy-to-use frameworks for using tools to enshrine bespoke logic of functionality. Low-code is not necessarily a weakness since tools with no access to customization are often simplistic and useless. Low-code is important for predictive, business and decision modelling.

Most no-code elements in Pyramid offer a low-code option to “upgrade” the functionality to match the user’s exacting needs (like the graphical formula editors).

High-code

The user is expected to provide detailed, script-level code input. Typically, needed for advanced calculations or ML (Machine Learning) scripts. The option for high-code is not always considered a bad thing – depending on the use case. In fact, the lack of high-code can be a major problem especially for DS/ML and advanced analytic formulations.

Application Modules

Data Factory

The data factory or ‘data factory for decisions’ is a concept to explain how the 7 app modules in Pyramid should be connected together into a frictionless  assembly line for taking raw data through all the typical stages to the point that it can give end users a framework for making data-driven decisions across their data estate.

Model

This app module is for the user to design and deploy semantic models. As part of that exercise, users may need to extract, clean, or transform source data (the “prep” part) in a stage called the “data flow”. This is followed by the semantic modeling stage simply called “model”. It is also the best location in the data factory to train, test and apply predictive models for most real world scenarios.

The outputs from Model are called a “data flow” and a “semantic” or “data” “model”.

Discover

This app module allows users to access and use semantic models to create analytic models. In doing so, they typically create visualizations of data that can be consumed in a way to highlight facts and figures from the data. Through this process, users get to “discover” information buried in such data. This alone may lead them to decisions.

Any predictions generated through predictive models can also be used with the original data in the semantic model to generate analytic models and visualizations. In this way, the predicted values are used to drive decisions.

The output from Discover is called a “discovery”.

Formulate

This app module allows users to build advanced calculations on the semantic model. These calculations are used in analytic models within the Discover layer to enrich the resulting visualizations and therefore drive better insights and information. Formulate is really a collection of different interfaces for building calculations (or custom members); lists (or custom sets); KPIs; parameters; and custom visuals;

The output from Formulate is called a “formula”. But it is more specifically formula calculations, lists, parameters, KPIs and custom visuals.

Tabulate

This app module allows users to develop “business models” using the spreadsheet paradigm. Users can integrate one or more analytic model (query) results with  spreadsheet formulas, and manually entered data to construct business models that can drive advanced decisions. Tabulate outputs can be shared with other users via dashboards and publications.

The output from Tabulate is called a “tabulation”.

Solve

The Solve plug-in for Tabulate is focused on decision modelling. It allows users to employ optimization engines to solve complex decision modeling problems. The results are the prescription for a business problem – and in effect produce the “decision model”. This supports the Prescriptive Analytics stage.

The output from Solve is a “decision model”.

Present

This app module allows designers to deliver various models to non-technical consumers of information to drive their decision making. Designers can connect multiple discoveries, often from multiple semantic models, in a singular canvas to tell a fuller story.

Present produces a live query, interactive interface, that allows consumers to play with the prepared analyses, visualizations and models. This becomes the segway to diagnostic analysis, versus merely looking at the descriptive content of what happened. If the content includes predictive values, this interface also offers the opportunity to consume and use predictions. Further, any tabulation artifacts that include business models and decision models can be included in Present. This then becomes a way to deliver prescriptive analyses to consumers.

The output from Present is called a “presentation”. It can take the form of a classic dashboard or the more modern “storyboard” – which includes descriptive text, graphics and non-analytical elements like videos.

Publish

This app module, like Present, allows designers to deliver all the various models to consumers of information to drive their decision making.  However, the outcome is a burst of static reports rather than a live, interactive interface. This delivery mechanism is often needed when addressing the multitudes of users who cannot connect to live systems  for various technical, security or logistical reasons.

The output from Publish is called a “publication”. It can sometimes be referred to as “report bursting” or “enterprise reporting”

Illustrate

This app module allows designers to create data driven infographics that can provide alternative techniques for sharing information rather than through traditional analytic charts and grids. The app also provides an interface for designing data driven natural language analysis. All illustrations and text are embedded in presentations or publications for consumption.

The output from Illustrate is called an “illustration”.

Query Shape

“Query shape” is a Pyramid specific concept describing the way the PYRANA query engine resolves complex analytical models (and queries) and their associated calculations. This includes the process of visualizing the results in an intuitive form.

Resolving most data queries is not always complicated, however, representing them in the right “shape” for proper analysis and executing them efficiently is. Most engines in the market lack many shaping capabilities – which often forces end users to manually shape results in tools like Excel.

Governance

Governance refers to the collection of strategies, processes, and tools that enable an organization to maximize the value generated from analytics while ensuring that the necessary security, data privacy, and audit controls are maintained.

Usually strong governance controls comes at the expense of self-service, and vice versa. Pyramid balances both.

Content Management System

The framework or system that allows content to be stored. The content spans all aspects of the application from the data flow designs and predictive models, to semantic, analytic, and business models.

The key aspects of a strong “CMS (Content Management System)” is that it supports a tight governance framework, security, reusability, shareability, versioning and documentation. Pyramid’s CMS is based on a repository database which provides a single, centralized store, of all analytical content rather than storing content in multiple file locations on a network or local disk. This removes duplication, improves governance and security, and provides the ability to monitor usage and performance.

Sharable Business Logic

Pyramid allows each element in the analytic catalog to be stored atomically. This provides a framework to share business logic and analytic content across all application modules, across different projects and across users - removing duplication and inconsistencies in solutions. This heavily reduces the TCO (Total Cost of Ownership) for developing and maintaining a Decision Intelligence solution.

Collaboration

A key aspect of modern analytics and decision systems is the ability to bring multiple users around the proverbial table to formulate and solve business problems together. For users to collaborate, they need multiple capabilities, that are designed into the CMS and are part of the governance architecture. This includes the ability to share and reuse content (the main aspect of collaboration); the ability to share insights through conversation threads (usually through annotations and workflows); the ability to share insights and knowledge through other social platforms such as Slack or Teams.

Many people mistakenly assume that conversations are the only element of collaboration.

Analytic Lineage

Analytic lineage is a method of tracing the interrelationships and dependencies between the various elements of a data factory. This goes from the raw data, through the various models and finally to the endpoints of the “factory”.

If the elements are highly atomic (like in Pyramid), and are reused multiple times across different projects, analytic lineage and tracing is critical to any governance mechanisms. High levels of reusability, with low levels of redundancy and duplication, is obviously the desired outcome.

Structure Analyzer

A structure analyzer looks at content and ensures that all the atomic elements and the underlying models are structurally correct. Models and definitions often break when either their definitions are changed; when upstream element definitions are changed; or, when the underlying data set structures or values change.

The analyzer uses the lineage framework to work out the tree of dependencies in the CMS.

Workflows

Workflows are used to annotate data points, hierarchy members and reports. The conversation framework is the default out-of-the-box workflow that allows users to converse with each other via threaded discussions and comments which are attached to data points, members or reports - turning analytical content into actionable analytical content.

Custom workflows enable customers to define and implement solutions to support other business processes that go above and beyond a simple conversation thread paradigm  - while being attached to data points, members and reports in the analytic domain (For example, a task workflow tool.)

Actions

Actions are a means of extending the classic flow of analytics by letting users pivot to other content (internal or external ) or trigger events from specific data points or members in a visual. Unlike workflows that are tracked or marked against the item or data model in the system, Actions are “fire and forget” – taking users to the new destinations while injecting the selection context into the targeted activity.

Alerts

Alerts are user configured messages that are sent based on changes in data values. Alerts provide a “push” mechanism to inform the user of the change without the user needing to actively go in and review data values on a regular basis.

Model Mapping

The ability to link separate data models in the context of application modules (Present and Publish) rather than at the data model level. This allows end users to perform cross-model analysis and integration (like filtering and highlighting) at the query level, without the complexity of blending data or building unscalable composite queries.

PYRANA

The core engine in Pyramid that accepts analytic models, calculations and meshes them with the semantic model to run queries against data sets, returning them to the client tools to render visualizations, infographics, and/or dynamic text.

Decision Intelligence Platform

The data & analytics factory for decisions is a collection of different app modules, infrastructural frameworks and content management systems designed to deliver decision intelligence. As such, the entire solution is best described as a “platform” for delivering decision intelligence or a “Decision Intelligence Platform” and encompasses the 3 primary analytical practices; Business Analytics, Data Prep and Data Science.

Services Architecture

This is a design of the application on the server side, where each of the core components of the application are separated out into separate “services” – which can run independently of the other services. This approach is heavily favored in scale out design, because it means that the most used services can be multiplied out as needed for handle incoming demand. Conversely, the services with low demand can be scaled back. “Micro services” means the services in the application are very atomic in function and purpose.

Containerization

A container is a payload of software that packages up code and all its dependencies, so the application runs quickly and reliably from one computing environment to another. It includes a light-weight operating system like Linux. Containers are the modern technique for deploying a specific “service” based application (see above) and scaling it as needed based on demand. There are numerous scaling environment technologies. Kubernetes is the predominant technology.

Kubernetes, with containers, offers a highly effective mechanism to shrink or grow the computing power underpinning an application (aka “Elastic computing”).

Pulse

The standalone Pyramid application connects to an instance of Pyramid offering remote querying of data sources securely without a private network (VPN). Apart from its incredible speed benefits over VPN-based connectivity, its simpler to setup, easier to scale, and fully multi-tenant in architectural design compared to a VPN approach.

Client Applications

Pyramid has several client-facing applications:

·        The Full client – web based, used by admins, pros, scientists and analysts

·        The View client – web based, used by viewers

·        The Direct client –web based, lighter client used to open content directly, without the authoring tools

·        The Mobile client – native client with web content, used for phones and tablets for pros, scientists, analysts and viewers

·        The Embed client – web based, super light interface for consuming Discover and Present content inside a hosted application.

Embedding

Embedded content is live content hosted in another application. The end user accesses the third-party host application to consume the analytic content – so data is live and linked to Pyramid by definition. The content is usually interactive by default but the level of interactivity can be reduced by design and simplify the user experience if required. A business application with Pyramid embedded can be accessed by many anonymous users (which is the typical deployment model).

Since the full, viewer, direct and embed clients are all web based, they can all be technically embedded through iFrames (which is how most competitors describe and implement embedding). However, Pyramid’s separate Embed client can be embedded DIRECTLY into the host, without iFrames – making it lighter, faster and more memory efficient.