AI Trends The Enterprise Systems Group Cannot Ignore

Introduction

The landscape of enterprise artificial intelligence has reached an inflection point in late 2025. With 88% of organizations now regularly using AI in at least one business function, up from 78% a year ago, and the enterprise AI market projected to grow from $24 billion in 2024 to between $150 and $200 billion by 2030, the question is no longer whether to adopt AI but how rapidly and effectively to scale these capabilities. This analysis examines the critical AI trends that Enterprise Systems Groups must address to maintain competitive advantage and operational excellence.

1. The Emergence of Agentic AI as the Next Operating Paradigm

Perhaps the most transformative trend reshaping enterprise systems is the shift from assistive AI tools to autonomous agentic AI systems.

Unlike the copilots and chatbots that characterized the first wave of enterprise generative AI, agentic systems are designed to perceive, reason, plan, and act autonomously across enterprise workflows. According to McKinsey’s 2025 State of AI survey, 62% of organizations are already experimenting with AI agents, with 23% actively scaling agentic deployments within at least one business function. The fundamental difference between traditional AI and agentic AI lies in operational behavior. Traditional AI reacts to prompts and waits for commands, while agentic systems anticipate needs and initiate actions across interconnected systems including ERP, CRM, and ITSM platforms. This architectural shift enables what analysts are calling “autonomous orchestration,” where AI becomes the connective layer that coordinates between systems, across teams, and ahead of events. Boston Consulting Group notes that agentic AI is installing intelligent virtual assistants capable of analyzing data and making decisions without constant human intervention, representing a fundamental redefinition of how businesses operate. For Enterprise Systems Groups, the implications are significant. Research indicates that enterprises adopting agentic architectures have reduced repetitive resolution cycles by more than 60% because agents handle entire workflows rather than passing tasks back to humans. However, 78% of executives now agree that digital ecosystems will need to be built for AI agents as much as for humans over the next three to five years. This means that enterprise architecture must evolve to accommodate both human users and AI agents operating as autonomous participants within business processes.

2. Multimodal AI Capabilities Are Transforming Enterprise Data Utilization

The second major trend requiring attention is the mainstream adoption of multimodal AI systems that process and integrate text, images, video, audio, and other data types within unified models. The multimodal AI market is experiencing explosive growth, projected to surge from $1.4 billion in 2023 to $15.7 billion by 2030, reflecting a compound annual growth rate of 41.2%. Gartner predicts that by 2027, 40% of generative AI solutions will be multimodal, a substantial increase from just 1% in 2023. The enterprise implications of multimodal AI extend across virtually every business function. In customer support, multimodal systems can now interpret not only written queries but also voice tone nuances, facial expressions during video calls, and accompanying images or screenshots to deliver more contextually relevant responses. Finance and security teams are deploying multimodal AI for advanced fraud detection by analyzing transaction records alongside voice stress patterns and user intent in real time. Manufacturing and supply chain operations leverage multimodal analysis that combines visual inspection data with textual documentation and sensor readings for more comprehensive quality control and predictive maintenance. Enterprise Systems Groups must recognize that most business work involves more than text alone, encompassing screenshots, invoices, call recordings, specification sheets, and product images. Traditional text-only models cannot process these diverse inputs, creating gaps in analytical coverage.

Multimodal capabilities unlock entire workflow segments that were previously inaccessible to AI automation, enabling more complete process digitization and intelligence extraction.

3. AI Governance and Regulatory Compliance

AI governance has transitioned from an optional best practice to a regulatory requirement and competitive necessity. The European Union’s AI Act, which entered into force on August 1, 2024, represents the world’s first comprehensive AI regulation and adopts a risk-based approach that categorizes AI systems into four tiers with corresponding compliance obligations. Organizations deploying prohibited AI systems now face fines of up to €35 million or 7% of global annual turnover, while high-risk AI violations carry penalties of €15 million or 3% of global turnover. Despite these mounting pressures, a significant governance gap persists. Research indicates that while 64% of companies now use generative AI in core business functions, only 19% have established formal AI governance frameworks. This disparity represents both a compliance risk and a strategic vulnerability. According to Gartner, by 2025, 75% of organizations implementing AI governance tools will reduce compliance-related incidents by 40%. Additionally, by 2026, 80% of large enterprises are expected to formalize internal AI governance policies to mitigate risks and establish accountability frameworks.

For Enterprise Systems Groups, building robust governance infrastructure has become essential

For Enterprise Systems Groups, building robust governance infrastructure has become essential. This includes implementing comprehensive monitoring for AI model behavior, establishing audit trails for AI-driven decisions, enforcing data privacy controls, and ensuring compliance with sector-specific regulations beyond the AI Act such as GDPR, HIPAA, and financial services requirements. The governance challenge is compounded by the emergence of “shadow AI” deployments where employees use AI tools without organizational oversight, creating uncontrolled risk exposures.

4. Domain-Specific and Small Language Models as Strategic Assets

The enterprise AI landscape is witnessing a decisive shift from reliance on general-purpose foundation models toward domain-specific and smaller, more efficient language models optimized for particular industries and use cases. Research shows that specialized AI models consistently outperform general-purpose alternatives in business-critical applications, delivering higher accuracy and efficiency while requiring fewer computational resources. Organizations are now deploying three or more foundation models in their AI stacks, routing tasks to different models depending on requirements Notable examples of domain-specific models include BloombergGPT for financial forecasting and analysis, Med-PaLM 2 for healthcare applications, ChatLAW for legal research, and FinGPT for real-time financial analysis. Healthcare is now leading generative AI adoption with $500 million in enterprise investment, driven by precision requirements that make domain-specific AI essential for regulatory compliance and patient safety. Complementing this specialization trend, small language models such as Mistral 7B, LLaMA 3, and IBM’s Granite series are gaining enterprise traction. These models offer several advantages over their larger counterparts. They require fewer computational resources, enabling deployment in constrained environments including on-premises installations and edge devices. They can be fine-tuned with minimal data for specific enterprise applications while maintaining data privacy since processing can occur locally rather than in third-party cloud environments.

Enterprise Systems Groups should evaluate where smaller, task-focused models might deliver superior performance-to-cost ratios compared to large general-purpose models.

5. RAG Becomes the Enterprise Standard

Retrieval-augmented generation has emerged as a foundational architecture pattern for enterprise AI deployments, with the RAG market reaching $1.85 billion in 2024 and growing at 49% annually. This approach connects large language models to enterprise knowledge bases, grounding outputs in verified organizational data rather than relying solely on what models learned during training. The value proposition is compelling: 86% of enterprises now augment their AI models with RAG to improve accuracy and reduce hallucinations.The RAG architecture operates through two core phases:

• Enterprise content is encoded into vector representations and indexed for efficient retrieval.
• When users submit queries, the system retrieves the most relevant document snippets and includes them in the prompt sent to the language model, enabling source-attributed responses. Advanced implementations now incorporate hybrid retrieval combining keyword and semantic search, re-ranking algorithms for improved relevance, and multimodal embeddings that unify text and images in the same search space

Enterprise use cases with demonstrated ROI include employee policy copilots that answer HR and benefits queries with citations, customer support systems that ground responses in product documentation and known issues, legal and financial research tools that extract obligations and generate audit trails, and operations assistants that retrieve procedures from maintenance logs and safety documentation. For Enterprise Systems Groups, implementing RAG infrastructure represents a practical path to deploying AI that delivers accurate, traceable, and enterprise-specific intelligence.

6. The Data Foundation Crisis

A recurring finding across enterprise AI research is that AI systems are only as effective as the data foundations underlying them. As organizations increasingly deploy agentic AI that acts autonomously on information, this foundation becomes non-negotiable. Agents that act on flawed, outdated, or conflicting data sources risk undermining both performance and organizational trust. The challenges are substantial. Generative AI makes use of structured and unstructured data including audio, images, and video, yet most organizations have not historically governed unstructured data. Information retrieval systems in complex enterprise environments often encounter outdated or conflicting sources for the same queries, resulting in inaccurate AI responses. Data fragmentation across departments, complexity in legacy systems, and misalignment between business and technology ownership create persistent barriers to AI value realization

The challenges are substantial.

Leading organizations are responding by treating data as a strategic asset, prioritizing high-value data initiatives, establishing clear ownership and accountability for data domains, and building data products as curated datasets for specific purposes. One North American utility company that strengthened its data foundations achieved 20% to 25% efficiency gains in the first year and recovered approximately $10 million from billing discrepancies. Enterprise Systems Groups must recognize that successful AI deployment depends fundamentally on robust data strategy, governance, and quality management.

7. Explainable AI

As AI systems assume greater roles in high-stakes enterprise decisions, the demand for explainability and appropriate human oversight has intensified. The explainable AI market is projected to reach $9.77 billion in 2025 and grow to $20.74 billion by 2029. Explainability refers to the ability to understand and interpret why AI systems produce specific outputs, a capability essential for regulatory compliance, stakeholder trust, and operational accountability. The National Institute of Standards and Technology has articulated four principles driving explainable AI: systems must deliver accompanying evidence for outputs, provide explanations understandable to individual users, ensure explanations accurately reflect the system’s actual reasoning process, and operate only under conditions for which they were designed or have achieved sufficient confidence. For regulated industries including finance, healthcare, and legal services, the ability to explain AI decisions is not merely preferable but often legally required. Human-in-the-loop automation represents the practical implementation of appropriate oversight. Rather than allowing AI to execute tasks end-to-end without intervention, HITL approaches add approval, rejection, or feedback checkpoints at critical decision points. This is particularly important for agentic AI systems that take autonomous actions with potential real-world consequences. The goal is to achieve automation efficiency while maintaining the precision, nuance, and ethical reasoning that human judgment provides. Enterprise Systems Groups should design AI deployments with clear policies on when human intervention is required, who is responsible for reviews, and how feedback is captured to improve future performance.

8. AI Security Threats

AI security risks have evolved from theoretical concerns to active enterprise threats that fundamentally reshape cybersecurity requirements. Unlike traditional attack vectors targeting static infrastructure, AI security risks exploit the dynamic, learning nature of machine learning models. Adversarial machine learning attacks involve carefully crafted inputs designed to fool AI models into making incorrect decisions while appearing normal to human observers. Data poisoning attacks target the training phase by injecting malicious samples into training datasets, embedding corruption into the model’s learned behavior that becomes extremely difficult to detect.

Unlike traditional attack vectors targeting static infrastructure, AI security risks exploit the dynamic, learning nature of machine learning models

The threat landscape is intensifying rapidly. Security researchers have documented a 1,265% surge in phishing attacks linked to generative AI trends, with AI-generated phishing now considered the top enterprise email threat of 2025. The FBI has explicitly warned that AI greatly increases the speed, scale, and automation of phishing schemes by helping fraudsters craft highly convincing messages tailored to specific recipients. Beyond phishing, AI-powered malware can now operate autonomously, copying its behavior across networks and timing attacks strategically to avoid detection. Enterprises face unique vulnerabilities from inadequate visibility into AI model behavior, insufficient logging of AI decision-making processes, and weak identity and access management for AI systems. Only 14% of European IT and cybersecurity professionals feel their organizations are “very prepared” to manage the risks associated with generative AI, while 51% identify AI-driven cyber threats as their biggest concern for the coming year. Enterprise Systems Groups must integrate AI-specific security monitoring, implement zero-trust principles for AI agent interactions, and establish adversarial testing programs to identify vulnerabilities proactively

9. Workforce Transformation

The AI talent crisis has reached critical proportions, with skills shortages potentially costing the global economy up to $5.5 trillion by 2026. Over 90% of global enterprises are projected to face critical skills shortages by 2026, while AI demand exceeds supply by a ratio of 3.2:1 across key roles. The mismatch is stark: 94% of CEOs and CHROs identify AI as their top in-demand skill for 2025, yet only 35% of leaders feel they have prepared employees effectively for AI roles. The skills gap manifests in multiple dimensions. Technical skills including machine learning engineering, data engineering, and MLOps remain scarce, but soft skills gaps are equally concerning, with 73% of AI roles requiring business context understanding and 68% of projects failing due to poor AI-business alignment. Only 22% of employees receive sufficient AI training support today, even as 48% of workers express desire for formal generative AI instruction. A related trend is the rise of citizen developers and business technologists. Gartner predicts that by end of 2025, citizen developers will outnumber professional software developers by a ratio of 4:1 at large enterprises, with 41% of employees performing technology work now residing outside traditional IT departments. These business technologists leverage low-code platforms and AI tools to create applications without extensive programming knowledge. Forrester research confirms that AI-infused applications now top the list of projects citizen developers are building. Enterprise Systems Groups must balance investment in specialized AI talent with programs to develop and govern the growing citizen developer community.

10. Low-Code AI Platforms

The democratization of AI development through low-code and no-code platforms represents a fundamental shift in how enterprises build and deploy AI capabilities. Research indicates that 70% of organizations are planning adoption of low-code/no-code platforms by 2025, with these platforms enabling application development 50% faster than traditional coding approaches. Platforms such as OutSystems, Mendix, n8n, and Appian now incorporate AI capabilities that allow business users to build intelligent applications without deep technical expertise. Simultaneously, AI model orchestration has emerged as an enterprise imperative. As organizations deploy multiple AI models for different purposes, orchestrating these models into coherent workflows becomes essential. AI orchestration platforms coordinate, integrate, and manage multiple models, agents, data pipelines, and workflows across the organization. McKinsey finds that organizations redesigning processes around AI agents and integrating orchestration into their architecture unlock substantially higher ROI compared with fragmented deployments. The orchestration layer handles operational complexity including automated deployment and scaling, trigger management, data exchange between models, lifecycle management, and governance enforcement. Advanced capabilities include federated orchestration across partner ecosystems, continuous learning loops where models automatically retrain on production data, and seamless integration with existing enterprise systems.

Enterprise Systems Groups should evaluate their need for unified orchestration platforms as AI deployments proliferate across business functions.

11. Edge AI

Edge computing combined with AI is creating opportunities for real-time intelligence at the point of data generation rather than relying solely on centralized cloud processing

According to Gartner research, over 50% of enterprise data will be processed outside traditional data centers by 2025. The edge AI market is projected to grow at 28% annually through 2030, reflecting enterprise demand for low-latency, locally processed intelligence. The advantages of edge AI include reduced latency for time-critical decisions, lower bandwidth costs by processing data locally, improved data security through local processing, and better scalability as billions of IoT devices come online. Manufacturing environments use edge AI for predictive maintenance and real-time quality inspection. Retail operations deploy edge-based customer behavior analysis. Healthcare applications enable continuous patient monitoring without cloud round-trips.Digital twin technology represents a particularly powerful convergence of edge computing, AI, and enterprise systems. Digital twins are virtual replicas of physical assets, processes, or entire facilities that are continuously updated with real-time sensor data. AI transforms these from passive simulations into active decision-support engines, with manufacturers reporting 30-60% productivity improvements, 20% reduction in material waste, and 25% decrease in production quality issues. As these technologies mature, Enterprise Systems Groups should evaluate where edge-based intelligence could deliver operational advantages.

12. Sustainability Considerations

AI’s energy consumption presents a classic Jevons Paradox: while individual AI tasks become more energy-efficient through hardware and software optimization, aggregate energy consumption is exploding because efficiency gains make AI more accessible and affordable, fueling a surge in overall demand.

The environmental footprint of AI has become an enterprise governance concern that cannot be ignored. Data centers now consume approximately 4.4% of all electricity in the United States, with carbon intensity 48% higher than the national average. By 2028, researchers estimate that energy allocated specifically to AI functions will reach 165 terawatt-hours annually, surpassing the total electricity currently consumed by all US data centers for all purposes. AI’s energy consumption presents a classic Jevons Paradox: while individual AI tasks become more energy-efficient through hardware and software optimization, aggregate energy consumption is exploding because efficiency gains make AI more accessible and affordable, fueling a surge in overall demand. Organizations are responding with multiple strategies. Google has reported achieving a 33-fold decrease in energy consumption per AI query over 12 months, while carbon emissions per query dropped 44-fold. Techniques such as model quantization, pruning, and the use of smaller specialized models can dramatically reduce energy requirements for individual AI tasks. Data center operators are transitioning to renewable energy through long-term power purchase agreements and implementing advanced cooling technologies and waste heat reuse. Sustainable AI frameworks are emerging as governance priorities, encompassing energy efficiency, resource optimization, and electronic waste reduction. Small language models align with sustainability objectives by requiring fewer computational resources and enabling on-premises or edge deployment that reduces data transmission energy. Enterprise Systems Groups should incorporate sustainability metrics into AI deployment decisions and vendor evaluations

13. Quality Assurance

The challenge of AI hallucinations, where systems generate factually incorrect or fabricated outputs that appear confident and credible, has emerged as a critical operational and governance concern. Benchmark measurements reveal hallucination rates ranging from 31% to 82% across different domains, presenting stark contrast to the single-digit error rates often claimed on public leaderboards. This gap creates uncertainty for enterprises attempting to assess AI reliability. The business risks are substantial. Hallucinated outputs in regulatory reporting, medical advice, financial analysis, or contract negotiations can create legal liability, reputational damage, and operational failures. A notable case involved fabricated legal citations surfacing in a New York court matter, underscoring the need for source grounding and review processes. Mitigation approaches include implementing retrieval-augmented generation to ground outputs in verified knowledge bases, employing careful prompt engineering that explicitly requests uncertainty acknowledgment, leveraging multi-model ensemble approaches that compare outputs from independent systems, and maintaining human oversight especially in high-stakes applications. Organizations should establish graduated trust levels based on use case criticality, where creative content generation may tolerate higher hallucination rates than factual reporting or analytical outputs informing strategic decisions.

Enterprise Systems Groups must develop hallucination risk frameworks as part of broader AI governance.

Strategic Recommendations for Enterprise Systems Groups

The AI trends outlined in this analysis converge on several strategic imperatives.

1. Enterprise Systems Groups must architect for agent-first operations, designing systems that accommodate both human users and autonomous AI agents as first-class participants in business processes. This requires rethinking APIs, access controls, workflow engines, and audit mechanisms.
2. Data infrastructure demands immediate attention. The recurring finding that AI effectiveness depends on data foundations means that investments in data quality, governance, ownership, and accessibility are prerequisites for AI value realization. Organizations should prioritize data product development that creates curated, discoverable, interoperable datasets built for specific high-value purposes.
3. Governance infrastructure must mature rapidly. With regulatory requirements intensifying and risks from ungoverned AI proliferating, enterprises need comprehensive AI management systems covering model inventory, risk assessment, compliance monitoring, and incident response. The EU AI Act timeline requires documented compliance roadmaps.
4. Hybrid talent strategies combining specialized AI expertise with citizen developer enablement offer the most practical path forward given the severe skills shortage. This means establishing proper governance frameworks for citizen development while investing in up-skilling programs that prepare existing employees for AI-augmented roles.
5. Enterprise Systems Groups should adopt portfolio approaches to AI, deploying multiple specialized models orchestrated through unified platforms rather than seeking single general-purpose solutions.

Domain-specific models, small language models, and RAG architectures should be evaluated alongside large foundation models based on use case requirements for accuracy, latency, cost, and explainability. The organizations that treat AI as a catalyst for enterprise transformation rather than an incremental efficiency tool, that redesign workflows rather than merely automating existing processes, and that build the governance and data foundations required for responsible scaling will establish sustainable competitive advantages in the years ahead.

References: