Breaking Down Enterprise AI, Part I: Insight Track

2023 was an incredibly exciting year for artificial intelligence. Several technology trends – distributed computing, transformer models, and large repositories of data – converged to produce powerful generative AI products like ChatGPT and Midjourney. These products have helped generate excitement for the field of AI overall, but the vast spectrum of available AI approaches may leave you unsure of what type of AI to pursue for your own enterprise goals. 

Enterprise Knowledge’s Enterprise AI (e-AI) division has experience delivering enterprise solutions throughout the field of AI, which we break down into the following categories: 



In this blog we will focus on the insight track, which includes Descriptive, Diagnostic, Predictive, and Prescriptive AI. These categories of AI are focused on analyzing and modeling data to produce insights. A sequel to this blog will focus on the action track, including Expert, Generative, and Agentic AI, which focuses on systems that can utilize data to perform actions.

Enterprise AI begins primarily with data. Making good decisions based on past data is a core function of human intelligence, and AI techniques from the insight track are designed to augment this human capacity or build it into automated systems. 

Descriptive AI is the most basic level of the insight track. Describing and analyzing patterns in data is the most fundamental activity AI can assist with, and forms the foundation of more advanced forms of AI. Diagnostic AI assists with the identification of normal and abnormal patterns in the data, indicating where time and resources should be directed. Predictive AI extrapolates trends from past data into the future, assisting with sound decision-making about the future. Finally, Prescriptive AI makes use of techniques from the previous types of AI to recommend the best actions based on your enterprise goals. 

Let’s take a quick tour through each category of the insight track to understand what drives them and how they can be applied. 


Descriptive AI

Descriptive intelligence is the capability of assessing the shapes of data by its statistical features like size, distribution, or variance and presenting the assessments in relatable ways. Considered through this lens, Descriptive AI is surprisingly widespread. Automating these analyses and applying them to vast pools of data is where these simple techniques augment human capability. For example, visualizing datasets of enterprise transactions and customer relationships can be performed by an AI system for interpretation by human agents. 

Bringing these representations closer to the point of need for enterprise employees can enhance the immediate decision-making process, ideally resulting in faster and more frequent success. Building systems that allow access to descriptive AI at the point of need is one of the simplest ways an organization can begin to utilize artificial intelligence. 

See this case study as an example of our experience implementing descriptive AI in systems that can help organizations make the most of their data. 


Diagnostic AI

Diagnostic intelligence is the capability of distinguishing between normal and abnormal patterns in data. While descriptive AI is best applied to pools of data, diagnostic AI makes the most sense when applied to streams of data that must be monitored. Automatically distinguishing between normal and abnormal patterns in the data can help identify how well enterprise systems are functioning. This will help people allocate their attention more effectively in what could be a vast sea of enterprise data. Diagnostic AI systems have been used to look for patterns that are characteristic of malicious behavior like fraud or network intrusions, or to look for patterns that indicate a malfunction like the poor vital signs of a sick patient or the overconsumption of energy in an unoccupied building. 

Diagnostic AI can help identify issues in a system faster and more frequently than human monitoring, as these systems can operate consistently around the clock and detect patterns that might escape the attention of a human. This makes Diagnostic AI well-suited for augmenting defensive or conservative functions like risk mitigation and resource allocation. 

EK has led the design of Digital Twin knowledge graphs, which are a great example of a diagnostic AI form factor.


Predictive AI

Predictive intelligence involves estimating future outcomes based on past data. From anticipating other drivers on the road to choosing the next word in a sentence, predictive intelligence is extremely useful for determining what actions to take in the future. While a perfect oracle has not been built (and probably never will), imperfect predictions are essential for the process of planning and are even a feature of certain industries like entertainment. 

Creating a predictive AI system involves creating some sort of model of past data and using it to generate predictions based on some variable or set of variables. This can range from logical models where the rules and logical relationships represent the outcomes of past analyses to dynamic neural network models that can create a wide range of predictive models with the same underlying architecture. 

Applying predictive AI in the enterprise involves creating these models from past datasets and applying them in augmentative tools for people or as integrated modules of an action-oriented AI system like generative or agentic AI. This can help people leverage the information stored in past data in their decision-making process, or bring predictive intelligence to automated workers like self-driving cars or virtual customer service agents.

We have often developed predictive AI systems in collaboration with clients. For example, we worked on a solution that would help our client predict their energy consumption for sustainability practices.


Prescriptive AI

Prescriptive intelligence takes us to the event horizon of action without actually crossing. Prescriptive AI involves creating a system that can integrate the factual information from previous levels of the insight track and output a suggestion on what should be done. Not only do we have some idea of abnormal data or a prediction of the future to base our action on, but we also have the judgment necessary to make a decision based on that information. 

Some of the simplest examples of this sort of system exist in the recommendation algorithms that power our social media and entertainment industries. These sorts of prescriptions look like the combination of multiple data interpretations: “We noticed you have watched several sci-fi movies in the past (descriptive AI), and we have also noticed that many sci-fi movie enthusiasts  have enjoyed this particular film (diagnostic AI). We think you would like it! (predictive AI) You should watch it! (prescriptive AI)”. Prescriptive AI isn’t just the suggestion that you should do something, it’s the process of combining these prior conclusions into the suggestion that’s sent to the user.

Prescriptive AI can be extremely useful for growing engagement with an AI system because if it is accurate it is likely to significantly decrease the amount of friction users experience with their tools. On one hand, this allows the knowledge and information stored in your system to reach users more frequently; on the other hand, over-optimizing for the engagement of your users can be risky in consumer products. Complex trade-offs like these abound in the field of AI, which highlights the importance of defining a strategy for its use. 

EK has extensive experience developing prescriptive AI systems based on semantic recommendation knowledge graphs. For an example, see this case study. In addition to recommendation systems, knowledge graphs and semantic technology can be used to help improve the explainability of AI systems, provide contextual information to LLMs, and synthesize data from across the organization into a unified semantic layer for improved findability and data management, which will benefit your AI efforts even further.



We hope this article has been helpful for understanding how you can leverage artificial intelligence. If you have an idea for your organization and need help building it, or if you are looking for help strategizing your organizational approach to the entire spectrum of AI please contact us! We’ll be happy to help.

Ethan Hamilton Ethan Hamilton is a curious and persistent Data Engineer with experience designing and developing infrastructure for semantic and generative AI solutions. Ethan approaches client problems by earnestly discussing challenges with clients, designing AI systems that can solve the most pertinent issues, and implementing the data infrastructure for such systems in collaboration with other teammates. More from Ethan Hamilton »