How to Spot an AI Imposter - Part 4

Three Foundational AI Enablers for Business Intelligence.

At Ferraro Consulting we have come up with three technology innovations that separate the leaders from all others in their use of AI for business intelligence: a business semantic layer, an embedded AI engine, and advanced human in the loop feedback.

A Business Semantic Layer

If your business intelligence AI doesn’t speak business like you speak business, it will fail. For this reason, we recommend an architecture that puts generative AI on top of a business semantic layer.

In Diagram 3, the business semantics layer communicates directly with the compute engine and the compute engine does high performance analytics directly on the data. The analytical compute engine then feeds the results back up into the user’s desktop, through the generative AI interface.

This architecture works, first because generative AI does not speak the language of your business. It needs some training. The business semantic layer provides the language needed by the generative AI engine. Second, the architecture works because generative AI does not have an analytical compute engine. It needs some help. The compute engine provides the power necessary to get answers quickly. Third, this architecture works because generative AI has problems with hallucination. The last thing you want is a generative AI engine that directly connects to your data.

A business semantic layer solves the challenges you might face if you just tacked a generative AI engine onto a business intelligence platform. The generative AI engine only has to communicate with the semantic layer. Only the business semantic layer defines the query. The query runs directly from the semantic layer to the data, and therefore it returns accurate results to generative AI user. The business semantic layer provides greater accuracy, and it provides the analytical compute power necessary for enterprise BI.

An Embedded AI Engine

The second innovation that distinguishes BI leaders from BI laggards and newbies is the existence of a core AI engine embedded. In the platform generative AI by itself is not AI. Generative AI is a component in an ecosystem of AI modules that can all work together.

Diagram 4 shows how generative AI works well with an existing, embedded AI engine in the business intelligence platform. Instead of expecting a generative AI engine to do all the work, the engine is able to pull from work that is already being done by core AI capabilities already embedded in the BI platform.

In this architecture, the presence of an embedded analytical compute engine multiplies the number of capabilities supported by generative AI. Several business intelligence platforms already include an advanced analytical engine for complex algorithms, a recommendation engine for next best action, a simulation engine for forecasting and impact analysis, a natural language processing engine, and other AI capabilities. Generative AI by itself cannot do all that AI is capable of doing. So then, to the extent which a BI platform has already been doing AI enablement, they will exceed the capabilities of new entries, laggards, and add-ons.

Advanced Human in the Loop Feedback

The third innovation that separates BI leaders from laggards is human in the loop feedback. A comprehensive implementation of human in the loop feedback will continuously increase the value generated by AI enabled BI. Ferraro Consulting has identified six different elements of advanced feedback.

First, the most basic human in the loop feedback is the simplistic positive-negative or thumbs up-thumbs down option. While bilateral feedback is a good start, it does not provide the level of detail necessary to fine tune generative AI for BI.

Second, human in the loop feedback should have language corrections, the ability of the of the user to go into the semantic model, change the language, and redirect both query and search as needed.

Third, in order to ensure accuracy, it should be possible to break every question down into phrases. Solid human in the loop feedback allows for the association of business phrases with the language in the data model for a specific organization.

Fourth, the model steward should be able to update the business glossary to reflect the best possible association of terms and definitions with the semantic model. The level of granularity is commensurate with the continuing improvement of accuracy.

Fifth, administrators and subject matter experts should be able to see all corrections and a single control plane. By so doing, they can determine which corrections to prioritize in the fine tuning of the model.

Sixth, transparency for all human in the loop feedback is extremely important for the governance of AI for BI. In addition, transparency ensures a better understanding of how generative AI is being used by your organization, and how the impact is improving over time.

For a complete analysis of what really matters for AI enabled BI, read the Ferraro Consulting POV paper, Hype Alert: Not All AI Enabled BI Makes the Grade.