Takeaways
- The choice between building or buying AI models should align with the company’s business strategy, risk tolerance, and proprietary data value.
- Building AI models offers control but requires substantial resources and maintenance, while pre-built solutions enable faster deployment at the expense of customization.
- Companies should adopt a flexible approach to AI strategy, as the market and technology are evolving too rapidly for long-term "future-proofing."
- To optimize AI investments, businesses should establish a reference architecture that can adapt to various use cases.
- AI deployment decisions will likely need to be revisited as new technological advancements arise.
Summary
This episode of AI360, hosted by Rohan and featuring Baris Sarer from Deloitte, explores the pivotal decision companies face when implementing AI solutions: to build custom models or to purchase pre-built ones from vendors. Sarer outlines four primary options in the AI tooling landscape: large language models from cloud providers, embedded generative AI capabilities in enterprise software, specific point solutions from third parties, and open-source models. Each option has unique advantages, from flexibility in proprietary models to the cost-efficiency and speed of vendor solutions.
Sarer explains that the build-vs-buy decision isn’t binary but rather exists along a spectrum. Off-the-shelf solutions are easier to deploy but may lack the nuance required for complex use cases, while custom-built models offer control but require significant investment and upkeep. To aid businesses in this decision, Sarer recommends a framework involving considerations of business strategy, risk tolerance, costs, and the unique value of proprietary data. Given the rapid pace of AI development, he advises a flexible, toolkit-based approach, allowing businesses to evolve and integrate multiple solutions over time.
