Leverage AI in Your ERP Transformation – Just Don’t Rely on Agents

Why It Matters

Billions invested in generative AI are yielding minimal ROI, exposing a critical gap between AI hype and real enterprise value. Without memory‑enabled AI, organizations cannot eliminate costly shadow processes or improve operational efficiency.

Key Takeaways

• ERP systems handle ~30% of transactions, 70% need manual workarounds
• Agentic AI struggles with undocumented shadow processes and lacks persistent memory
• 95% of firms see no measurable AI impact; only 5% reach production
• Enterprises need AI that retains contextual memory to improve ERP outcomes

Pulse Analysis

ERP platforms have long promised end‑to‑end visibility, yet most enterprises still rely on spreadsheets, email chains, and informal workarounds to handle the 70% of transactions that fall outside the ideal path. These shadow workflows are invisible to management, create audit risks, and inflate operating costs. As businesses scale, the divergence between recorded data and real‑world activity widens, making it harder for finance and supply‑chain leaders to maintain compliance and accurate forecasting.

Agentic AI entered the scene in 2026 with the allure of autonomous agents that can navigate interfaces, interpret ambiguous data, and adapt to regulatory shifts without explicit reprogramming. In practice, these agents are limited to the data and documented processes they can observe. They cannot surface the informal approval loops or ad‑hoc reconciliations that have become the de‑facto operating system for many firms. Moreover, the prevailing stateless architecture means each task is executed in isolation, erasing the contextual insights that seasoned employees accumulate over months or years. The result is a cycle of re‑reasoning that hampers scalability and erodes trust.

The market response has been sobering: despite $30‑40 billion poured into enterprise generative AI, a recent MIT study found 95% of companies report no measurable impact on the bottom line, and only 5% have moved beyond pilot phases. Executives are therefore calling for a new class of AI that embeds persistent memory, learns from each interaction, and can bridge the gap between formal ERP logic and the reality of shadow processes. Such memory‑enabled solutions could finally unlock the promised efficiencies of ERP transformations, turning AI from a flashy add‑on into a strategic asset that drives measurable profit and risk reduction.