ETR recently conducted its latest survey on generative AI and large language models (LLMs), focusing on budget plans, evaluation and production rates, anticipated ROI, and generative AI vendors. The survey received responses from nearly 900 IT decision makers, including representatives from nearly a quarter of Global 2000 organizations. 

In this study, ETR polled nearly 900 IT decision makers about their budgets and priorities for generative AI, with almost a quarter of respondents representing Global 2000 organizations. Large and North American enterprises were the most common size and regional subsamples, and the IT/TelCo, Services/Consulting, and Financials/Insurance industries collectively represented 54% of respondents. Less than a quarter of respondents say their organization is NOT evaluating generative AI / LLMs, a number that has consistently fallen each survey period. Customer support, text and data summarization, and code generation and documentation are the most common business use cases for which organizations are evaluating generative AI / LLMs. Writing copy has risen considerably as a use case, whereas image editing remains stagnant.

Evaluation rates for each use case have risen since six months ago for every business use case identified in the survey. However, about a third of organizations have yet to move from evaluation to production with generative AI / LLMs. Production rates are rising similarly across most business use cases, with customer support the most common, followed by code generation and documentation. Production rates lag evaluation rates by around 10+ percentage points in each business use case. Continuing evaluation is the top reason why more organizations have not moved to production with generative AI / LLMs. However, around a third of respondents also cited data privacy or security concerns and legal, compliance, or regulatory concerns as reasons for not having moved into production.

More than a quarter of respondents remain unsure when they will see a return on investment (ROI) with their generative AI projects, which may signal a lack of clarity around metrics or business use cases for the technology. Nearly a quarter of those who could estimate an ROI timeline said between 4 to 6 months after initial deployment, and 15% said sooner. An equal 15% expects more than a year's duration before ROI is realized.

In this survey period, nearly 50% of respondents said generative AI investments were newly added to the budget, while 40% said generative AI dollars were reallocated from elsewhere. Of those who said generative AI funds were reallocated from elsewhere, business applications were the most common source, followed by non-IT departments and productivity applications. Only 7% said their organization's gen AI budget was sourced from existing RPA allocation.