Rate Recommendation Multi-Objective Alignment Score (RRMOAS)

GMI-EU-042Energy & UtilitiesInfrastructure-Dependent

Definition

Measures whether AI rate recommendations achieve balanced performance across four objectives (customer bill savings, utility revenue stability, grid signal alignment, and equity) or sacrifice some for others.

The Measurement Gap

Traditional rate design evaluation checks revenue sufficiency (does the proposed rate recover the utility's costs?) and cost-of-service allocation (does each rate class pay its proportional share?). These are necessary but insufficient for AI-designed rates. They answer "is this rate legal?" not "is this rate balanced across the objectives the commission, customers, and the grid all need it to serve?"

The specific tracking failure is dimensional collapse: AI rate optimizers typically maximize one objective (minimizing bill volatility, or maximizing TOU adoption, or minimizing revenue variance) and treat the others as soft constraints or post-hoc checks. The resulting rate may score 0.92 on its primary objective while scoring 0.31 on equity, a combination that looks like success in the optimizer output but constitutes a rate case liability in front of a commission with equity priorities.

RRMOAS enables the operational decision of objective rebalancing: when any dimension's score drops below an acceptable threshold, the optimizer must be constrained or the rate design must be revised before filing. This is a model governance decision. Without the multi-dimensional score, the rate design team has no structured way to catch single-dimension optimization failures before filing.

What goes wrong without it: the utility files a rate designed by an AI optimizer, the equity dimension was ignored during development, consumer advocates produce an analysis showing disproportionate burden on low-income customers, and the commission orders a remedial filing. RRMOAS would have flagged this before the initial filing. This metric requires billing simulation infrastructure and income-matched customer data, hence Infrastructure-Dependent, but is achievable for utilities with mature CIS and AMI analytics environments.

Formula

RRMOAS = (Score_savings × Score_stability × Score_signal × Score_equity)^(1/4)

Variables:

  • Score_savings: 0–1, percentage of eligible customers whose bills decrease under the recommended rate (normalized to 0–1 scale). Source: CIS billing simulation engine
  • Score_stability: 0–1, ratio of projected revenue variance under recommended rate to current-rate revenue variance; lower variance yields higher score. Source: Revenue accounting, billing simulation
  • Score_signal: 0–1, degree to which the rate preserves or strengthens time-varying price signals (peak/off-peak differential maintained relative to current rate). Source: AMI interval usage data, rate design model
  • Score_equity: 0–1, inverse of the Gini coefficient of bill-impact-as-percentage-of-income across customer income quintiles. Source: Census ACS data geocoded to service territory, CIS billing simulation

Calculation Frequency: Per rate design iteration (pre-filing evaluation); not a continuous operational metric
Output: RRMOAS, 0–1 geometric mean. Values above 0.6 indicate reasonable balance across all four objectives. Below 0.4 means at least one dimension is severely compromised. If any single dimension scores 0, RRMOAS equals 0 by construction. This is intentional; a rate that completely fails one dimension is not acceptable regardless of other scores.

Proposed formula, pending validation pilots.

At a Glance

Category
Customer AI Fairness and Revenue Integrity

Decision Supported

Is the AI rate optimizer solving the right multi-objective problem, or is it maximizing one dimension while degrading others that the utility also cares about?

← Rate Design Equity Impact Score (RDEIS)Revenue Protection Segment Bias Rate (RPSBR) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment