Rajiv Gupta
June 24, 2026
The bust: True, and that's exactly where most organizations stop. The real question isn't whether noisy data is a problem. It's whether your system is designed to surface and help clear the noise or silently absorb it.
Every large transformation program runs on imperfect data. Requirements that were never fully documented. Decisions that happened in a hallway and never made it into a system. Organizational dynamics that shaped a scope call but don't appear in any artifact. That's the baseline, not the exception.
A system that waits for clean data before it can be useful isn't a program intelligence tool. It's shelfware.
The more dangerous failure is subtler: automated analysis that processes incomplete or noisy inputs and returns an answer that looks authoritative. No flag. No caveat. Just a clean output built on a distorted foundation. By the time someone traces the problem back to the source, the misinformed decision has already been made.
Axiamatic is built around a different assumption: that data will always be incomplete and noisy but that should neither lead us astray nor shut us down. The way out? Human judgement as continuous input in a form similar to what powers LLMs: RLHF (Reinforced Learning with Human Feedback). When the platform detects conflicting signals, incomplete artifact chains, or patterns that don't resolve cleanly, it doesn't paper over them. It brings them forward to the appropriate human — business analysts, the PMO, the workstream lead — with enough context to make a call. The system flags what it can't resolve. The human closes the loop. The system learns.
Designing for that gap is the point, not a concession to it.
The programs that get this right aren't the ones with the cleanest data. They're the ones where the gap between what the system knows and what actually happened never gets wide enough to matter.
