Case Study: Implementation and Adoption of Generative AI at Goldman Sachs


The advent of generative AI and large language models (LLMs) presents a series of innovative opportunities for companies in the technology sector. Goldman Sachs, a multinational investment bank, has embraced this change, with various proof of concepts in place. The company’s Chief Information Officer, Marco Argenti, explains his perspective on the adoption of generative AI and discusses the benefits and challenges of integrating these models into their operations.

Key Observations

  1. A survey by KPMG LLP reveals that 65% of executives believe generative AI will significantly impact their organization within three to five years, and 77% anticipate it will have a substantial societal impact during the same period. However, 60% indicate they are a year or two away from implementing their first generative AI solution, due to barriers like talent, cost, and data privacy.
  2. Goldman Sachs is actively investing in generative AI. They are currently testing multiple use cases, particularly in areas such as automating code and document categorization.
  3. Preliminary results from these experiments are encouraging. For instance, document classification has demonstrated accuracy equivalent to human performance, while initial coding experiments show potential efficiency gains in the double digits.

Deep Dive: Generative AI at Goldman Sachs


The company’s initial approach involved a pilot to make developers more productive using AI co-pilot tools. This initiative led to quick efficiency gains and allowed developers to concentrate on crucial tasks instead of repetitive ones. The company is now experimenting with various use cases, including document classification and categorization.


Goldman Sachs has several proof-of-concept implementations underway, though none have reached the production stage. They are experimenting with LLMs for tasks such as summarizing earnings calls and creating daily digests. Additionally, generative AI is being explored to categorize and extract information from millions of documents received by the company.


Early results are promising. Document classification has achieved accuracy as good as human performance. Initial experiments in code generation suggest that the AI-produced code could potentially be accepted by developers up to 40% of the time, leading to considerable efficiency gains.

Challenges and Barriers

Despite the promising potential of generative AI, several barriers to full implementation exist:

  1. Knowledge gap: The rapidly changing technology landscape and the need to adapt control frameworks for security necessitate continuous learning and adaptation.
  2. Talent scarcity: The limited number of professionals with advanced AI and LLM knowledge present a challenge in implementing these technologies.
  3. Resource constraints: High demand for graphical processing units (GPUs) from large tech companies may lead to resource scarcity for training AI models.

Future Outlook

Argenti envisions a rapid pace of implementation, estimating that AI integration at Goldman Sachs may take only months rather than years. However, he acknowledges the exact timeline is uncertain due to the novelty of the technology.


Despite the challenges, the potential benefits of generative AI make it an area of priority for Goldman Sachs. While the timeline for full implementation is still uncertain, the promising results from initial experiments and the high anticipated impact underscore the potential of this emerging technology. As such, Goldman Sachs’ journey offers a valuable case study in embracing AI and LLMs in the corporate sector.

Goldman Sachs CIO Tests Generative AI

Get in touch

Whether you’re looking for expert guidance on an AI initiative or want to share your AI knowledge with others, our network is the place for you. Let’s work together to build a brighter future powered by AI.