Analysis

    Enterprise use of foundation models is very time-consuming and expensive, and it is not limited to just training the models but includes all aspects of working with them. Recent research from TechTarget’s Enterprise Strategy Group indicates several key enterprise preferences for dealing with foundation models.1

    First, there is a widespread desire among organizations for customized outcomes and to leverage foundation models to produce outcomes that are distinctly unique to the organization. According to Enterprise Strategy Group research, 40% of organizations with generative AI projects in either the production or proof-of-concept (PoC) stage strongly agree that they see value in training their own generative AI models.

    Organizations are customizing foundation models through grounding foundation model data with their proprietary data. Methods include fine-tuning, prompt engineering, and Retrieval Augmented Generation (RAG) techniques. Only 6% of organizations with generative AI projects in either the production or PoC stage say they are not customizing the foundation models they are using.

    Organizations are increasingly experimenting with a variety of models from a range of sources in a quest to fit the best foundation model to a particular use case. 43% of respondents to Enterprise Strategy Group research with generative AI projects in either the production or PoC stage say that a combination of proprietary, commercial, and/or open source foundation models will best align with their organization’s approach to leverage foundation models to support generative AI initiatives in the next 12 to 24 months.

    Further, there are growing signs that, with the right cost controls, organizations are willing to customize open source models to the point that the models truly become proprietary, with organizations willing to run their own model training in addition to model inference workloads. These trends of experimentation and customization mean organizations have to consider ways to run AI workloads cost-effectively.

    1. Source: Enterprise Strategy Group Research Report, The State of the Generative AI Market: Widespread Transformation Continues, September 2024. All Enterprise Strategy Group research references in this brief are from this report.