KPI & Metric Reality Auditor

The problem

Management can end up optimizing for numbers that have drifted away from the reality they were meant to represent. When teams chase targets like "lower handle time" or "higher deployment frequency," they can create hidden technical debt, customer friction, or quality erosion that dashboards fail to capture.

What it does

• Identifies specific ways a rational actor could improve a metric without improving the intended business outcome.
• Audits metric definitions for denominator blindness and average-based distortions that can hide tail risks, segment failures, or cohort differences.
• Surfaces hidden measurement burden, including manual tagging, engineering instrumentation, reporting overhead, and employee monitoring costs.
• Evaluates whether a metric functions as a real control signal or whether crossing a threshold triggers no meaningful operational change.
• Maps conflicting incentives where speed, quality, cost, and volume targets push against each other.

Why this beats prompting it yourself

General LLMs often accept KPI frameworks at face value or respond with generic advice to "track more metrics." This skill applies a structured adversarial audit focused on proxy failure, incentive distortion, hidden burden, and whether the metric can actually change a decision.

Use cases

• Audit engineering DORA metrics to check whether faster deployment is hiding a rising change failure rate.
• Review sales commission structures for incentives that may increase short-term revenue while worsening long-term churn.
• Test support SLAs to see whether time-to-resolution targets could encourage premature ticket closure.
• Validate product activation metrics to check whether they reflect retained value rather than shallow clicks.

Known limitations

This skill does not verify underlying database integrity, certify analytics accuracy, prove intentional metric gaming, or replace formal statistical analysis. Strong conclusions depend on the quality of the metric definitions, data, and operating context provided.