CHAPTER 3: Common Pitfalls in AI Programs
Understanding Data Sources & Training/Maintaining AI Models
When building an AI program, it’s essential to understand the primary data types and the roles required for effective
deployment. Companies new to AI typically deal with structured data sets, which are organized in a predefined manner,
often in spreadsheet programs such as Excel or relational databases such as SQL Server. Examples of structured data include
customer records, transaction histories, and inventory logs—all of which can be easily searched, analyzed, and organized.
But, to leverage generative AI models effectively, companies need to be comfortable with unstructured data, which does not
have a predefined format. Unstructured data includes text documents, images, videos, social media posts, and other content
that is not easily searchable or analyzable without specialized tools. This data requires different storage and processing
approaches to train AI models effectively.
Operationalizing these models takes skills and expertise, and often, several people to build and maintain them. Understanding
the nuances of each role will help you build a capable AI team that is prepared to manage model training and
implementation. The three roles needed include:
Business Expert
Engineer
Statistician or Data Scientist
This person understands your
Responsible for the AI model’s coding,
This role involves understanding the
company, the AI strategy, and
operationalization, and maintenance.
underlying mathematics and fine-
generally, how the AI model works.
This person often supports the data
tuning the model. This person builds
Their knowledge bridges the gap
aggregation and pipeline requirements
the algorithm and ensures its
between technical implementation
to “feed” the AI model. This person
accuracy and relevance to your
and business objectives, ensuring
ensures the model is efficiently
company’s unique use case.
that the AI solutions align with your
integrated into your company’s
strategic goals.
systems and performs reliably.
A Blueprint for CMO Success in the High-Stakes AI Marketplace
7
Powered by FlippingBook