A new report from Deloitte explores the evolving landscape of generative AI in enterprises, highlighting significant advancements and ongoing challenges. Titled “The State of Generative AI in the Enterprise: Now Decides Next,” the survey gathered insights from 2,770 business and technology leaders across 14 countries and six industries.
As part of Deloitte's quarterly series on generative AI, this report builds on the findings released in January, which indicated concerns among business leaders regarding societal impacts and the need for technical talent.
The latest survey reveals that organizations are eager to harness the potential of generative AI while facing hurdles related to scalability, data management, risk mitigation, and measuring value. Although early successes are boosting investment, widespread implementation still faces considerable challenges.
Key Findings:
- 67% of organizations are boosting investments in generative AI due to strong initial returns.
- 68% have moved 30% or fewer of their generative AI experiments into production.
- 75% are increasing investments in data lifecycle management for generative AI.
- Only 23% feel well-prepared for risks related to generative AI governance.
- 41% struggle to define and measure the impacts of their generative AI initiatives.
- 55% have avoided certain generative AI applications due to data-related issues.
Kieran Norton, a principal at Deloitte, noted that many clients are in the prototyping and piloting phase without progressing to production. This caution is often linked to concerns about data quality and potential biases in AI models.
Impact of Risk Concerns on AI Deployments
The Deloitte survey aligns with recent reports emphasizing the current state of enterprise AI adoption. For instance, a PwC report indicated high interest in generative AI but also highlighted gaps in understanding AI risks.
Norton pointed out that executives are hesitant to proceed with enterprise AI deployments until their concerns—such as data quality, bias, security, trust, privacy, and regulatory compliance—are addressed. While these risks aren’t new, generative AI introduces unique complexities. Norton believes that organizations can utilize their existing risk management practices to navigate these challenges, though certain areas, particularly data quality management, require enhancement to specifically address generative AI risks.
Norton expressed the concern over "hallucination," where generative AI models produce inaccurate outputs, often stemming from a misunderstanding of the input data. To mitigate this, organizations may benefit from using smaller, targeted language models and tailored training.
Demonstrating the Value of Generative AI Initiatives
The report reveals that 41% of organizations find it difficult to measure the impact of their generative AI efforts, with only 16% regularly reporting value creation to their CFOs. Norton attributes this challenge to the diverse use cases and emphasizes a need for a detailed, use-case-specific approach.
He recommends that organizations establish key performance indicators (KPIs) tailored to each initiative, focusing on specific business problems such as productivity or user experience. By addressing issues at the use case level rather than viewing them as a collective portfolio, organizations can more effectively evaluate and demonstrate the value of their generative AI investments.
In summary, while generative AI presents promising opportunities for enterprises, addressing data quality, risk management, and establishing clear metrics are essential for successful implementation and measurement of value.