For decades, enterprises have poured billions into Enterprise Resource Planning (ERP) systems to optimise operational processes. Yet, in 2025, the efficiency crisis still festers across production-heavy industries.
Despite the ERP market having surpassed $80 billion, plants continue to run on fragmented data, manual repairs, and siloed workflows.
A quiet rebellion is brewing among AI-native engineering firms like ShellKode and master-data specialists like Verdantis, whose common belief is that ERP didn’t fail, enterprise data discipline did.
The Billion-Dollar Efficiency Paradox
Verdantis’ MRO survey of global asset-intensive enterprises revealed that over 51% struggle with data quality issues, particularly fragmented master data across ERP and procurement systems.
While interest in AI adoption is rising, most organisations lack operational data readiness, resulting in continued maintenance delays, reactive decision-making, and inconsistent operational truth, despite heavy ERP investment.
Rohan Salvi, associate director at Verdantis, points out, this failure is not solely technological.
“ERP systems were built on rigid frameworks decades ago. They maintain data, but they don’t evolve with it. Enterprises still treat ERP as the single source of truth, even when that truth is inconsistent,” he says.
Enterprises run AI pilots at the edge, while core operations remain stuck in reactive mode. AI dashboards forecast supply risk, but a missing part number stalls production for weeks.
Enter ShellKode, a four-year-old, bootstrapped firm from Coimbatore that recently signed a multi-year strategic collaboration agreement with AWS. Unlike large SaaS vendors, ShellKode doesn’t build assistants or copilots. It builds agents capable of making decisions inside live enterprise workflows.
Founder and CEO of Shellkode, Arun Kumar, elaborates on the distinction: “People talk about retrieval agents. But an enterprise needs decision-making agents, ones that change databases, update systems, and execute actions. That’s the real frontier.”
ShellKode has quietly deployed over 200 enterprise agents across BFSI, retail, logistics, and healthcare. They aren’t replacing ERP, but layering autonomy over it.
Why ERP Isn’t Enough Anymore
Both ShellKode and Verdantis agree: ERP systems were never designed for a world of dynamic, unstructured, continuously shifting data.
Salvi observes that every prospect they meet is using copilots somewhere. But copilots don’t fix bad data, they interpret it, he quips, “enterprises now want specialised agents with governance and accuracy, not generic chat interfaces.”
ShellKode echoes the sentiment from the builder’s perspective. Kumar explains how ERPs break under real-world velocity. “Large platforms like SAP or Oracle can’t be customised fast. We don’t replace them. We extract the data, build an autonomous layer, and return decisions. That’s where enterprises see ROI.”
This fundamental shift, from static systems to live intelligent orchestration, marks a new era. ERP is no longer the brain. It becomes the body. The brain, more often than not, is an agentic AI.
Metadata, Not Models
While AI fever grows, Verdantis warns that the root crisis lies in something unfashionable: metadata. Thousands of manufacturing plants still lack basic uniformity in part numbers, vendor records or asset lifecycles.
“In many enterprises, metadata itself doesn’t exist. They load ERP with legacy codes and expect AI to read intent. Without structure, even the best models hallucinate,” Kumar warns.
Salvi agrees, citing recurring operational breakdowns. “Users shorten entries, skip fields, enter data enough for their department, but unusable for others. That’s why maintenance stays reactive. You can’t predict with fractured truth.”
While AI promises autonomy, it requires trust. Trust, in turn, demands consistency, which leads to effective governance.
Agentic AI: From Insight to Intervention
Unlike traditional AI models that suggest, agentic AI acts. In one insurance deployment, ShellKode is building a team of 32 agents capable of orchestrating claims, from document ingestion to customer callback.
The real test, Kumar says, isn’t technical, it’s ethical. “Customers ask: if I cut 80% of manpower using agents, what happens when something fails? So we build supervisor agents, audit logs, and fallbacks. It’s not just code, it’s accountability.”
This is where Verdantis sees rapid adoption of specialised governance models. Rather than replacing ERP, enterprises are retrofitting AI layers around it to enforce auditability, lineage and version control, without compromising compliance.
Maintenance, repair, and operations (MRO) may be the most complex battlefield for AI adoption across enterprises. Maintenance and spare parts don’t follow dashboards; they follow downtime. Predictive promises crumble when a discontinued component halts a line.
Salvi illustrates the challenge. “Assets evolve. Parts phase out. If updated information isn’t captured continuously, maintenance stays reactive. AI is powerful, but without fresh data, it predicts yesterday’s failure, not tomorrow’s.”
Yet the rewards are vast. Verdantis reports up to 60% reduction in procurement turnaround time post data governance. ShellKode reports double-digit ROIs on decision agents in logistics and retail.
Salvi believes the next frontier lies in agent trust. “Hallucination is a serious deterrent. When you alter master data, everything downstream is affected. So every AI conversation now starts with one question: how do you guarantee reliability?”
The question isn’t whether enterprises will adopt AI; they already are. The question is whether they will allow it to alter operational truth.
Consultants or Clouds
While hyperscalers push monolithic platforms, both experts believe the future belongs to engineering-first builders.
“You don’t need huge teams, you need focus. Agent AI will be led by those who understand process pain, not platform menus,” Salvi asserts.
Kumar adds, “We are not an applied tech company. We are industry solution builders. That’s why enterprises trust us over tools, they want agents that speak factory, not English.”
As ERP giants race to embed AI features, an uncomfortable truth surfaces: enterprises don’t need smarter reports. They need systems that fix themselves. Systems that ask no permission to act. Systems that don’t wait for month-end consolidation to correct a fault created six weeks earlier.
This is where the story truly shifts, from software adoption to operational autonomy. ERP gave enterprises structure. Agentic AI may finally give them motion.
The post Can Agentic AI Fix Enterprise Efficiency Where ERP Failed? appeared first on Analytics India Magazine.
