Why Agentic AI Isn't Delivering on Its Promise
As shared services organizations increasingly invest in generative and agentic AI, expectations are rising around its ability to transform finance operations and unlock efficiency at scale.
Despite the momentum, many teams struggle to translate AI pilots into enterprise-wide outcomes. The challenge is not the technology itself, but that many organizations are deploying isolated tools, rather than embedding agents into end-to-end processes.
The gap between agentic ambition and execution was a key theme at the 2026 Shared Services & Outsourcing Week in Orlando. IBM shared practical insights to agentic finance transformation in the workshop Stuck in Pilots? Practical Approaches to Shift Agentic AI Aspirations into Business Outcomes.
The discussion brought together:
- Khalid Siddiqui, Global Offering Leader, IBM
- Anish Jain, OTC Offering Lead, IBM Consulting, IBM
- Marisela Cadenas, Finance Operations, IBM
As they discussed how organizations can turn experimentation into sustainable results. Below are five key takeaways from the session, and what they mean for organizations looking to scale agentic AI across shared services.
1. Data is the foundation of scalable agentic AI.
Agentic AI only delivers value at scale when it is grounded in robust enterprise data. Modern AI tools rely on an interconnected ecosystem where agents, models, and databases work together.
For example, a general query assistant may be unable to accurately answer how many vacation days a user has, as it cannot access employee records, policies, or other context. But an AI agent connected to HR systems could retrieve insights accurately and instantly.
"The system does not recognise the prompt because the system does not know anything about me."
Without direct integration into enterprise systems, AI remains constrained by context gaps and fragmented information.
2. Orchestration is what makes agents truly agentic.
Orchestration was highlighted as the single most important word of the SSOW conference, and for good reason. Agentic AI success comes from moving beyond a single model or assistant, and organizations should orchestrate ecosystems of agents that seamlessly interact.
"The magic happens when you're creating a network of agents, and each agent has a key skill."
Without a strong orchestration layer, teams will remain stuck in fragmented pilots that cannot scale.
"The agentic ecosystem is simple if you know what end-to-end process you want to achieve."
3. Agentic AI metrics should focus on business outcomes.
Too often, success is defined narrowly through efficiency metrics or task automation, rather than the business outcomes those improvements are meant to enable, such as:
- Reduced task completion times
- Improved cash flow
- Reduced delays
- Enhanced customer experience
"We are not creating an agent just to create a solution, but to execute an experience or outcome for that business."
These metrics highlight how organizations need to align AI projects with broader business metrics; efficiency is only one part of the ROI.
4. Scaling agentic tools depends on people, not just technology.
Amongst the agentic hype, it is important to remember that implementation does not guarantee success. Sustainable agentic initiatives rely on workforce readiness and leadership alignment.
"We have to be extremely transparent on what these programmes mean for people and for the business."
As AI automates tasks, roles are evolving as employees shift towards strategic, data-driven work. Shared services are beginning to manage hybrid teams of humans and digital agents. As such, investing in skills, communicating clearly, and embedding an AI mindset is key.
"AI is a racecar, and the humans are the professional drivers."
5. Robust security and governance are key to sustainable AI implementation.
Without clearly defined controls, organizations risk limiting adoption and undermining trust. Best practices include:
- Role-based access: tailoring access to agents based on operational roles to prevent misuse.
- Embed data governance into AI strategy: explicit control over how agents interact with enterprise data and systems.
- Flexible deployment: organizations operate across diverse IT landscapes, so platforms need to support on-premises cloud-based, and hybrid environments.
This highlights the importance of enterprise readiness when scaling agentic AI initiatives. Robust security and governance allow agents to be embedded into core finance processes without heightened risk.
"We have already taken steps… to ensure that we have the right infrastructure, security and controls."
Where Agentic AI is Already Delivering Scalable Impact
While agentic AI has the potential to transform finance end‑to‑end, one of the strongest examples is Order‑to‑Cash (O2C), particularly in the resolution of blocked orders.
Traditionally, resolving a blocked order requires manual investigation across multiple systems, which can take 15–30 minutes depending on complexity. When orchestrated agents are applied, the same tasks can be completed in seconds. Agentic tools reduce manual intervention by:
- Retrieving order details
- Identifying causes of blocks
- Aggregating customer financial data
- Providing risk insights
- Triggering approvals
- Ordering release workflows
Beyond O2C, customer query management is another strong use case. IBM presented a case study, noting that across channels such as WhatsApp, SMS, Slack, and email, 50–60% of a representative's time is spent managing inbound queries. By introducing agentic assistants, organizations can reduce query handling time from 30–40 minutes to just four minutes, while supporting between 1.5 and 2 million queries annually.
These examples illustrate how agentic AI delivers scale when it is embedded into end‑to‑end processes and measured by business impact.
How to Move from Agentic Pilots to Scale
Agentic AI projects fail when organizations treat the technology as an experimental tool, instead of an enterprise capability. Organizations that successfully scale operate on:
- Data as the foundation: integrating agents with governed enterprise systems.
- Orchestration over isolation: building networks of specialized agents across processes
- Outcome‑led metrics: KPIs focused on business impact rather than efficiency.
- People at the centre: talent supported as part of hybrid human‑agent teams.
- Strong security and governance: controls enable trust and enterprise-wide deployment.