Despite decades of investment, enterprise resource planning (ERP) software platforms frequently fall short of delivering the process efficiencies and business intelligence expected. More than 80% of ERP transformations continue to miss budget, timeline, and value goals, according to Bain & Company’s recent benchmarking survey.

One reason for underperformance is that implementations often leave platforms too customized and costly to maintain, as well as delivering low return on investment from a business intelligence standpoint. The vision of an agile and harmonized backbone running standardized processes across the enterprise too often yields sprawling, bespoke systems that are complex and difficult to upgrade.

Generative AI’s automation potential offers a way forward. While some early use cases have focused on accelerating ERP migrations by automating testing, code remediation, or documentation, the real potential lies in creating “touchless platforms” powered by agentic AI.

The agentic opportunity

Agentic AI systems are intelligent, event-driven digital components that act automatically—for example, by rerouting workflows or initiating decisions without human input. They can unlock big opportunities in efficiency and agility, so the ERP becomes not just a passive source of data but a dynamic decision-and-execution engine. In the future, this might mean employees can work directly with AI agents, rather than through a software user interface.

Platform providers are racing to shape this new landscape, offering mostly off-the-shelf tools with some studio customization that converge to a large degree. However, despite these products still being at an early stage, and often addressing only narrow use cases, the trend toward more advanced agents is clear.

CIOs need to plan how they’re going to get there based on their business needs.

Stuck in the pilot phase

Agentic tools give organizations the ability to deploy packaged agents and develop custom ones. Yet in practice, five main roadblocks mean many companies remain stuck at the pilot stage. First, organizational challenges include unclear operating models for human-agent interaction and limited internal skills. On the technical side, most agentic tooling is still immature, with orchestration frameworks and customization options only recently emerging.

Data quality is a significant concern, especially where information is siloed or lacks governance. Strategic concerns, such as fear of vendor lock-in or lack of sponsorship from senior executives, are further slowing adoption. Finally, ROI is hard to quantify, with pricing, productivity gains, and outcomes difficult to predict.

Still, the direction of travel is clear. According to Bain’s study, 78% of IT leaders expect at least some ERP functionality to be replaced or augmented by agentic AI over the next three years.

The impact is expected to be most visible in core finance and planning processes. Survey respondents highlighted procure to pay, record to report, and forecast to plan as the ERP areas most likely to see early gains from agentic automation.

Moving forward

To scale pilot projects effectively, CIOs need to answer four interlinked questions.

1. What’s required? The first step is laying the foundations. That starts with prioritizing the best use cases. A focused approach brings faster results, so organizations should begin by prioritizing the three to five use cases with the highest business value.

New operating models are often needed, too, whether to define the human-agent interaction framework, adjust workforce roles and responsibilities, or ensure transparency into agent performance. Robust governance is essential to manage compliance and data risks, as well as to enable scale.

Organizations must also redesign core processes to be agent-first. This goes beyond bolting agents onto existing workflows. It means defining clear triggers and data structures, creating guardrails and auditability, and rethinking decision making and execution. An agentic transformation doesn’t stop at one or two pilots; it requires a system that can support dozens or hundreds of agents across the organization.

Many CIOs are already wrestling with these challenges. But without securing this groundwork, agentic AI is unlikely to scale.

2. Build, buy, or partner? With foundational readiness in place, the next question is how to deliver. Should companies build custom agents? Buy from existing vendors? Or partner to codevelop solutions? The answers vary per use case and depend on seven considerations:

• the strategic relevance of what is being contemplated and whether it is core to differentiation;
• vendor application fit, and how much of the desired functionality they can provide;
• how easily an external solution could be integrated with existing systems;
• total cost of ownership, including development and maintenance;
• whether there is sufficient internal talent and capacity to proceed internally;
• regulatory risk (for example, relating to niche standards); and
• how any decision to buy or build would affect future vendor lock-in.

One instance in which CIOs may choose to partner with platform publishers is when use cases involve standard, noncritical workflows confined within a single system—enabling speed and efficiency at a lower cost.

On the other hand, they may buy from a third-party specialist when the capability is a commodity and not overly complex, allowing the organization to gain traction quickly without heavy investment.

Alternatively, building in-house is likely to make the most sense for cross-system, high-impact processes that enable competitive differentiation and where maintaining full control justifies the longer timelines and higher cost involved.

In the end, the right delivery model will rarely be one-size-fits-all, and CIOs need to choose the best path based on their use cases and company priorities. Regardless of the initial decision, it is important to remain dynamic and vigilant, as innovation in this space is moving fast and tools will need to be refreshed frequently to stay best in class.

3. How can partners help? Partnerships are critical to moving from pilot to scale. But the right type depends on where the organization is in its journey. Early on, when companies are exploring agentic possibilities, working with their platform provider is essential. They supply the tools, templates, and technical support needed for safe experimentation. Many offer low-code studios and prepackaged agents that can jump-start deployment.

As organizations mature, business integrators are increasingly important. Their role is to support strategic thinking: helping CIOs prioritize use cases, define the operating model, and redesign core processes. They can also help to develop build-buy-partner blueprints and establish frameworks for measuring ROI.

System integrators, meanwhile, are beginning to build deep agentic capabilities. Today, their main contribution is helping clients customize and deploy agents across complex system landscapes. As tooling matures and knowledge diffuses, their role is expected to expand.

These partnerships will evolve as the organization achieves greater maturity in its deployment of agents. As projects advance, the organization will learn from partner experience and be able to carry out more of these roles internally.

4. What are the risks of inaction? The final question is, What happens if companies don’t scale use of agentic AI? While many remain at the pilot stage, a growing number are making progress, scaling agents, and realizing benefits—meaning those that don’t act risk falling further behind. Bain research finds that leaders who have scaled AI across workflows are already banking EBITDA gains of 10% to 25%. Those who wait are in danger of being outpaced not just by rivals but by their own vendors’ roadmaps, with less opportunity to influence the development of features that meet their needs.

Second, the longer an organization delays, the greater the possibility of vendor lock-in. As platforms evolve into orchestration hubs controlling both the agents and their interconnections, switching becomes harder. Enterprises that fail to build their own “AI muscle” risk becoming passive consumers of someone else’s automation.

Third, transformation costs may rise. Organizations that separate their agentic journey from broader IT modernization may end up reworking processes twice: once during ERP migration and again when embedding AI. To optimize costs, organizations should ensure their agentic transformation is part of their current enterprise technology overhaul.

Finally, there are people implications. Agentic AI is becoming a factor in workforce retention and recruitment. Talented professionals increasingly want to work on future-facing technologies. Organizations that delay risk losing their edge—not just in systems but in talent.

The case for action

Agentic AI is now embedded in the leading enterprise platforms. The tools exist and the potential is proven, yet many CIOs remain in “wait and see” mode, unsure of how and when to scale. Breaking this deadlock hinges on focusing on the four questions outlined above.

One of the biggest questions still to be decided is who will control the orchestration layer. Platform providers are positioning themselves as the central hub, offering interoperability on paper but nudging customers toward proprietary ecosystems.

CIOs should be vigilant. Agentic AI success depends on tools and talent, as well as on architectural decisions that preserve flexibility. As multi-agent systems become more complex, the ability to direct and monitor them may become a defining source of enterprise advantage.

The winners will be those that successfully make the move from experimentation to execution, unlocking the next wave of value from their enterprise platforms.