How Enterprise AI Can Move Beyond Randomness to Accountable Decisions

The AI revolution has arrived in the enterprise, but it's exposing a fundamental gap: language models are inherently inconsistent, and inconsistency makes them unsuitable for high-stakes decisions. Alex Porcelli, CEO of Aletyx and longtime contributor to open-source business rules engines, argues that the real challenge isn't model capability—it's accountability.

The Accountability Gap#

When Mark Cuban recently observed that AI cannot reliably produce the same answer to the same question twice, he identified what many practitioners see as a feature masquerading as a bug. The deeper issue, Porcelli contends, is not inconsistency itself, but the lack of governance and explainability that mission-critical decisions demand.

In regulated industries—banking, insurance, healthcare—inconsistency is only the symptom. The disease is the inability to:

• Reproduce a decision and trace exactly why input A and input B produced output C
• Version and audit a decision-making process in real time
• Ground decisions in policy rather than opaque model outputs
• Review and approve changes before deployment

Without these properties, enterprise teams cannot satisfy auditors, comply with regulations, or delegate genuinely important decisions to AI systems alone.

Why Domain Knowledge Alone Isn't Enough#

The technologists thriving with AI share one trait: deep domain expertise. But domain knowledge and individual judgment cannot scale to enterprises with thousands of employees making decisions that carry legal and financial stakes. That's where decision models become essential.

Decision models are a proven technology for encoding business rules into executable, auditable forms—the same structures that have governed enterprise decision-making for decades. The challenge is bringing that rigor into the agentic AI era.

Decision Models in Practice#

Porcelli describes decision models as business rules expressed in both human-readable (spreadsheet-like decision tables) and executable formats. A decision table lists conditions in columns on the left and outcomes on the right. Each row is a rule: "If customer is foreign, classify as BBC; if U.S. resident earning over $80,000 and aged X–Y, classify as AAA."

Simple rules become powerful when chained. Complex decisions layer multiple decision tables to assess risk, eligibility, or classification. One financial services client processes 18,000 transactions per second using decision models—all 20 million daily transactions reclassified and re-evaluated in under 20 minutes. The speed matters because customer tier determines product access, pricing, and the entire experience with the institution.