Definition
Ratio of network performance outcomes when RF engineers override SON decisions compared to outcomes when SON decisions are accepted, measuring whether human intervention adds or destroys value.
The Measurement Gap
Decision Supported: Should the organization trust SON more and override less, or is RF engineer judgment providing genuine value that the AI is missing?
Measurement Gap:
Legacy override tracking in telecom is binary: the override happened or it didn't. Nobody measures whether the override improved anything. The SON platform logs the parameter change, the change management system records the override, but no system connects the override to the subsequent performance outcome in the affected cells. This means RF engineering teams can override SON decisions thousands of times per quarter with zero accountability for whether those overrides helped. Organizations either trust SON blindly (and may be missing valuable human pattern recognition) or override SON constantly (and may be degrading network performance). Without SOVR, neither the trust nor the override can be evaluated against evidence (Chen & Zhang, 2023; GSMA Intelligence, 2024).
Formula
Formula (Plain Language): For every RF engineer override of a SON parameter change, measure the network performance outcome in the affected cell(s) over the following N hours. Compare to the average outcome when similar SON changes are accepted without override. Ratio above 1.0 means overrides improve outcomes. Below 1.0 means overrides make things worse.
Notation:
SOVR = Mean_outcome(overridden_changes) / Mean_outcome(accepted_changes)
Variable Definitions:
- Mean_outcome(overridden_changes) = average performance outcome metric in affected cells for the N hours following RF engineer overrides, using the same metric the SON was optimizing (e.g., spectral efficiency, call drop rate, throughput) (metric units match the SON objective)
- Mean_outcome(accepted_changes) = average performance outcome metric in affected cells for the N hours following accepted SON changes of comparable type (same metric units)
- N = outcome observation window (proposed starting point: 24 hours)
Units:
Dimensionless ratio. Values greater than 1.0 indicate overrides add value. Values less than 1.0 indicate overrides destroy value. A value of 1.0 indicates overrides are neutral.
Sign Convention:
Higher is better from the perspective of override value, but the organizational goal is not to maximize SOVR, a SOVR near 1.0 with few overrides may be healthier than a SOVR of 1.3 requiring constant human intervention.
Edge Cases:
- Override during a known event (outage, mass gathering): exclude from SOVR calculation. The override may be operationally correct regardless of measurable outcome in the standard metric.
- Confounding from concurrent changes: use only cells where the override was the sole change in the observation window, or apply statistical matching techniques to control for concurrent changes.
- Small override volume: if fewer than 20 overrides per quarter, confidence intervals will be wide. Report SOVR with uncertainty bands and do not draw conclusions from single-quarter results.
Proposed formula, pending validation pilots.