Definition
Measures the percentage of AI-resolved billing disputes where the resolution pattern matches a known systematic error: indicating the AI is propagating a historical mistake rather than correctly adjudicating the dispute.
The Measurement Gap
Traditional billing dispute management tracks resolution volume, average handle time, customer escalation rate, and first-contact resolution rate. These operational metrics report process efficiency: they say nothing about whether the AI's resolution decisions are financially correct. A system that resolves 95% of disputes on first contact and satisfies customers is reported as high-performing even if 15% of those resolutions match a known error pattern and are systematically under-billing or over-crediting the utility.
The specific tracking failure is the absence of a correctness audit tied to the AI's resolution pattern. Human agents making systematic errors get caught through supervisor review, customer re-escalation, or internal audit sampling. AI systems making the same errors at 100× volume can run for months before the financial impact surfaces in a revenue variance report: by which time the cumulative leakage is significant and the tracing is difficult.
DREPR enables the operational decision of resolution model correction: when a periodic audit identifies a systematic error pattern, how quickly is that pattern eliminated from the AI's resolution logic? DREPR quantifies both the prevalence of error replication and the speed of correction after audit findings are implemented.
What goes wrong without it: the utility deploys automated dispute resolution, a billing system migration introduces an error in demand charge calculations for a specific commercial rate class, the AI learns the error as correct behavior, and over 18 months it incorrectly resolves 4,000 disputes, systematically crediting customers $280 each who weren't entitled to the credit, a $1.1M revenue leakage that the efficiency dashboard reported as a 97% first-contact resolution success. AI and periodic audit together make DREPR computable immediately.
Formula
DREPR = AI_resolutions_matching_error_patterns / Total_AI_resolutions_sampled
Companion: Revenue Impact:
DREPR_$ = DREPR × Avg_dispute_value × Total_AI_resolutions
Variables:
- AI_resolutions_matching_error_patterns: Auto-resolved disputes where the resolution action (credit applied, charge reversed, adjustment type) matches a documented systematic error pattern from the most recent periodic audit. Source: Revenue protection case management system, audit findings database
- Total_AI_resolutions_sampled: Sample of AI-resolved disputes reviewed in the measurement period (minimum 50 per audit cycle, random selection). Source: CIS / dispute management system
- Avg_dispute_value: Average dollar value of disputes in the sampled population. Source: CIS billing records
- Total_AI_resolutions: Full count of AI-resolved disputes in the measurement period. Source: Dispute management system
Calculation Frequency: Quarterly audit cycle; monthly trend monitoring via random sampling
Output: Percentage, 0–100%. Below 5% is healthy: error patterns are isolated incidents, not systemic. Above 15% means the AI is a material revenue leakage vector. DREPR should decline after each audit cycle as identified error patterns are corrected in the resolution model. A flat or rising DREPR after a correction cycle confirms the model update failed to eliminate the pattern.
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Is the automated dispute resolution engine learning from correct historical resolutions, or is it codifying and amplifying past mistakes?