Whenever talking to clients about utilizing AI, the conversation must turn to data. This is because data access, quality and consistency are a critical ingredient to developing an AI application – whether for decision-support or as part of a business process re-design effort.
Below is a summary of the insights and considerations around data and AI:
1) Artificial Intelligence (AI) and machine learning (ML) are intrinsically connected. Every AI application will need to have a database of “knowledge” to access when it is running utilizing the AI’s machine learning capabilities. This AI/ML knowledgebase is at the core of an AI application’s critical components
2) The need for how much and the type of data required for the AI application output/results to be “reliable” on Day 1 can be immense depending on what the output of the AI is expected to be
3) AI applications that are expected to provide decision-support may be the ultimate consumers of data, while business applications to support automation that requires analysis of documents, electronic input and other defined sets of data may need less data to satisfy use case requirements
4) Tools are available to assist with assessing the quality of an organization’s available data for a conceptualized AI application’s use case
5) These tools can also point data analysts to “data gaps”, issues with inconsistent data quality that is critical for reliability of an AI’s output, and identify data that is not currently accessible internally or externally to develop the AI envisioned
6) Cleaning up or augmenting organizational data may be a major step in the development of an organization’s AI application
7) Creating, augmenting, correcting inconsistent data, adding missing data to existing databases can be a time-consuming and even costly practice for an AI project
8) Never forget an AI/ML application is a commitment to “continuous learning” which requires keeping data current, making corrections to existing data, and augmenting its accessible data to optimize and maintain its knowledgebase for improved, enhanced output results or another use case
9) Even if an organization is contemplating utilizing a third-party vendor-supported AI application, you may not have all the data required to complete the vendor application’s data models to process transactions or have adequate historical data to enable the AI to execute adequately to produce the desired results
The net takeaway of the above real-world discussions around data and AI is that early assessment of the data available to meet the output requirements of a proposed AI application is critical. Therefore, an organization must be flexible in adjusting its use case, possibly revising scope, and expected processing capabilities and/or planned results for generating information, recommendations, and insights.
If you would like to discuss our perspectives, our responses to the above discussions for your organization, and our approach to Enterprise AI, please contact us using the form below or at [email protected].