State of AI: Widely used for planning — drives the business at just 25% of firms

For CIOs and transformation leaders, the most important signal here is not the adoption rate; it is the split between strategic intent and operational readiness. When AI is influencing planning but not yet shaping the core operating model, the real management question becomes whether the organisation is funding a set of isolated use cases or building a repeatable decision system. That distinction affects who owns value, who absorbs risk, and whether AI remains a series of pilots with limited enterprise leverage.

The article points to a practical governance issue: budgets are moving faster than production control. That creates a portfolio trade-off. Leaders may need to choose between sponsoring visible productivity wins in business functions and investing in the less visible work of data quality, policy design, model oversight and workflow integration. Those investments are harder to defend quarter to quarter, but they determine whether AI can be extended beyond individual teams without creating compliance, resilience or accountability gaps.

Next questions for executives:

  • Which executive owns AI benefits once a use case crosses from experiment to production?
  • What decision rights govern tool approval, data access, human review and exception handling?
  • Which KPIs will prove AI is changing cycle time, cost-to-serve or revenue quality rather than simply increasing activity?

The operating-model implication is that AI maturity is now as much a leadership and change-management challenge as a technical one. Organisations that want scale will need to align business sponsorship, data stewardship and process ownership around a small number of priority workflows. The trade-off is speed versus control: moving quickly may show early value, but without explicit governance and workforce adoption plans, the organisation risks locking in fragmented tools, inconsistent usage and weak auditability. A more durable path is to treat AI as a managed transformation portfolio, not a collection of experiments.

AI adoption remains uneven. While many organizations are experimenting with AI — including data science and machine learning (DSML), generative AI and agentic AI — enterprise-wide deployment remains below 50%, according to recent research by Dresner Advisory Services. Reflecting that uneven maturity, only about a quarter of the 500 respondents to the “Special Report: Agentic and Generative AI” said AI was a primary driver of business strategy at the end of 2025. Still, that figure more than doubled compared with the first half of 2025 — a reminder of how quickly expectations are shifting. A larger share — 55% — report that AI influences strategic planning but is not yet central to it. Only 16% of organizations say they remain primarily focused on learning what AI can do, suggesting most have moved beyond experimentation — even if they have not yet scaled AI across the enterprise. As for why they are investing in AI, organizations cite tackling long-standing business challenges (49%), the risk of industry disruption (26%), and maintaining competitive parity (8%) as their primary motives. “Despite the hype, a majority of organizations are still early in their AI journeys, experimenting selectively rather than deploying AI at scale throughout their core business processes,” said Brian Lett, vice president at Dresner Advisory Services. “However, for those that are ready, AI has become an integral part of strategy that is worthy of investment,” Lett added. “For these organizations, AI is no longer a skunkworks initiative or speculative technology. Their AI adoption occurs in operations and processes, and links directly to concrete business outcomes.”

The strategic divide

Taken together, the data suggest a market in transition. The emerging divide is no longer between organizations experimenting with AI and those that are not. It’s between those that are strategically embedding AI into governed, production-grade processes and those using AI tactically to augment work. In my conversations with vendors over the past year, data maturity consistently came up as the primary bottleneck to scaling AI. Roughly half say they are building tools aimed at accelerating what has been a slow, multi-year process of data preparation and governance. Dresner’s findings reinforce that constraint: Without production-grade data and governance, AI initiatives stall at the pilot stage rather than moving into full production. What organizations put into production shows how far their AI efforts have progressed beyond experimentation. It also highlights the distinct roles different forms of AI play now.

Where AI is being applied

Data science and machine learning remain the most mature forms of enterprise AI, focused on optimizing decisions and generating operational insight. Common applications include churn modeling, forecasting, A/B testing, personalization, anomaly detection and resource allocation. Generative AI has gained traction primarily through use cases focused on workforce productivity. Its value lies primarily in empowering employees to augment their daily work. While valuable, these gains alone don’t necessarily translate into business transformation. Improving individual output is not the same as redesigning how work gets done. Agentic AI combines analytical models, generative capabilities and workflow automation to execute multi-step tasks across systems. Rather than stopping at insight or content generation, these systems act. They trigger workflows, update records and resolve issues guided by defined policies. Unlike DSML models that optimize decisions or generate predictions, agentic systems carry those decisions forward. Where DSML informs generative AI assists, agentic systems operate.
In the second half of last year, a majority of respondents reported being excited or cautiously optimistic about generative and agentic AI. n=500 Source: Dresner Advisory Services

Generative and agentic AI adoption

At the end of 2025, slightly more than half of organizations reported actively experimenting with generative and agentic AI. However, production deployment remains more limited: — 34% for generative AI and 15% for agentic AI — though both rates have more than doubled since 2024. Budget alignment is also accelerating, with 72% allocating money to generative AI initiatives and 66% to agentic AI. The gap between budget allocation and production deployment suggests that some of this spending is not yet translating directly into scaled applications. And while generative AI attracts significant investment, enterprise leaders say a portion of that funding is directed toward foundational data work required to support advanced cases. In other words, AI budgets are quietly underwriting data modernization. As one university technology leader noted to me, teams may begin with simpler use cases, but developing an AI application that delivers a single view of the student or identifies at-risk students depends on unified, well-governed data environments.

Data maturity as a constraint

Dresner research on agentic AI shows a consistent pattern: Organizations that have moved agentic systems into production typically report previous success with BI, and data modeling and machine learning. They are also more likely to have a clearly defined data leader. In other words, AI adoption correlates with established data discipline. Organizations that have already invested in modernizing analytical data infrastructure, improving data quality, strengthening governance and reducing data silos are better positioned to operationalize AI at scale. Agentic capability tends to follow data maturity — not the other way around. Organizations that progress beyond experimentation tend to follow a structured path.

Steps to AI maturity: Experimentation to execution

For CIOs and data leaders, the priority is clear: move AI from experimentation to embedded execution. That shift requires discipline in use-case selection, governance and a commitment to data.
  1. Map DSML, generative and agentic AI to specific business problems. Define measurable outcomes and aligned funding accordingly.
  2. Prioritize use cases that can deliver measurable results using current systems and data. Avoid delaying value while waiting for ideal architectures.
  3. Embed generative AI into knowledge work and operational workflows, and measure productivity gains at the team and function level.
  4. Establish clear policies on approved tools, acceptable use, data handling and risk management.
  5. Audit AI capabilities already embedded in core enterprise applications (ERP, CRM, human capital management) and activate features before investing in new tools.
  6. Identify AI use cases that materially improve customer experiences or create new revenue streams.
  7. Start with priority business use cases, then define the minimum viable data capabilities required to scale them.
  8. Define a phased roadmap for delivering production-grade, governed data.
  9. Present executives with a clear investment choice: accelerate full data industrialization or pursue a staged capability model that incrementally advances data maturity — both require sustained funding and business ownership.
  10. In data-mature organizations, expand DSML to optimize end-to-end processes and reduce structural costs.
https://www.informationweek.com/machine-learning-ai/state-of-ai-widely-used-for-planning-drives-the-business-at-just-25-of-firms

Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.