Avenga Reveals Why Broken Processes Doom Enterprise AI

According to an analysis by Zuzana Drotárová, who leads a team of approximately 100 business analysts at IT professional services firm Avenga, many enterprise artificial intelligence failures originate entirely outside the model. When an AI system delivers a flawed or operationally incorrect result, organizations routinely blame hallucinations or poor prompting. However, Drotárová argues that the technology is often executing its programming perfectly, only to be sabotaged by chaotic file versioning, undocumented verbal agreements, and sluggish approval pipelines that cannot keep pace with automated outputs.

To illustrate the issue, Drotárová highlights how an AI contract reviewer might select an outdated document simply because a user autosaved it recently, or block a valid transaction because a verbal agreement—such as a €2,000 payment exception—was never codified. This reliance on tacit knowledge is a major hurdle. McKinsey recently identified a similar bottleneck in agentic AI deployments, noting that building effective autonomous agents forces companies to explicitly document expert practices that previously existed only in employees' heads. Furthermore, a Forbes analysis from late 2025 warned that when technology accelerates workflows faster than an organization can absorb them, the result is employee overwork rather than true efficiency.

For AI practitioners and enterprise leaders, this shift means that preparation must focus heavily on operational readiness rather than just model selection. Drotárová recommends that organizations answer three critical questions before deploying any AI agent: identifying undocumented decisions, establishing clear rules for identifying authoritative data, and ensuring human approval chains can handle the increased volume. To manage these risks, practitioners can look to the NIST AI Risk Management Framework, which advises clearly defining roles and responsibilities for human oversight. Ultimately, preparing for AI requires companies to make their internal processes explicit, transforming undocumented tribal knowledge into structured, machine-readable rules.