Only 25% of enterprises have deployed AI, and fewer have reaped the rewards — here’s why

The deployment of artificial intelligence in enterprises remains limited, with only 25% of companies having implemented AI in production and a quarter of those seeing measurable results, according to a new State of AI Development Report from Vellum.

Current state of enterprise AI adoption: The majority of companies are still in early stages of their AI implementation journey, with most either evaluating strategies or running proofs of concept.

• 53% of companies are building out strategies and proofs of concept
• 14% are in beta testing phases
• 7.9% are still in the initial requirements gathering stage
• Document parsing, analysis tools, and customer service chatbots are the most common AI applications being developed

Early impact metrics: Organizations that have deployed AI are experiencing varied levels of success, though many have yet to see substantial returns on their investments.

• 31.6% report gaining competitive advantages
• 27.1% cite cost and time savings
• 12.6% note higher user adoption rates
• 24.2% haven’t observed any meaningful impact from their AI investments

Model preferences and trends: While OpenAI maintains market leadership, enterprises are increasingly adopting a multi-model approach to meet specific needs.

• OpenAI’s GPT-4o and GPT-4o-mini remain the most popular choices
• Open-source models like Llama 3.2 70B are gaining traction through providers such as Groq and Fireworks AI
• Text processing leads use cases, followed by file creation, images, audio, and video
• Retrieval-augmented generation (RAG) is becoming a standard approach for information retrieval

Cross-functional involvement: AI development is expanding beyond IT departments to include various stakeholders across organizations.

• Engineering teams lead 82.3% of AI projects
• Leadership and executives are involved in 60.8% of initiatives
• Subject matter experts participate in 57.5% of projects
• Product teams contribute to 55.4% of AI developments
• Design departments are engaged in 38.2% of AI efforts

Implementation challenges: Organizations face several obstacles in their AI deployment journey.

• AI hallucinations and prompt engineering difficulties persist
• Model speed and performance issues affect deployment
• Data access and security concerns remain prominent
• Many organizations lack in-house technical expertise
• 18% of developers work without any AI development tools

Quality assurance practices: Current testing methods show room for improvement in ensuring AI system reliability.

• Over three-quarters of organizations rely on manual testing and reviews
• Some teams utilize automated evaluation tools and A/B testing
• Evaluation processes are particularly crucial for advanced agentic systems
• Automated testing frameworks are recommended for reliable production deployment

Looking ahead to 2025: The focus is shifting toward practical implementation and tooling solutions to address current challenges and drive meaningful business impact.

• Development platforms and frameworks are expected to play a crucial role
• Organizations need to identify specific, viable use cases for successful AI implementation
• A balanced approach combining various tools and technologies will be essential for success
• The emphasis is moving from experimentation to practical, results-driven applications

Practical implications: While AI adoption continues to grow, success depends on organizations taking a measured, strategic approach rather than implementing AI solely for its own sake. The key to meaningful implementation lies in identifying specific use cases that deliver clear ROI, supported by proper tooling and evaluation frameworks.