Definition
Percentage gap between AI-predicted demand response capacity at dispatch and actual measured load reduction delivered: measuring whether the DR program delivers the MW it promises when the grid actually needs them.
The Measurement Gap
Traditional DR program management tracks enrolled capacity (MW), event count, and aggregate seasonal performance. These metrics report what the program contains, not what it delivers at the moment of dispatch. A program can show 300 MW enrolled and 85% average seasonal performance while systematically failing during the highest-stress events. When customers override thermostats, when heat indexes exceed comfort thresholds, or late in a season when repeated dispatches have exhausted customer tolerance.
The specific tracking failure is temporal: program reports aggregate across all events and smooth over the variance. A single high-performance early-season event can mask three underperforming late-season events in the seasonal average. An AI dispatch model trained on historical participation rates will not detect that participation is declining in real time unless DRDG is computed per event and trended.
The exact operational decision this KPI enables is dispatch crediting: how many MW of DR can the control room actually count as available in real-time reserve calculations? Without DRDG, the answer is whatever the enrollment database says. With it, the answer is the enrollment figure discounted by the current-season gap rate, a fundamentally different number.
What goes wrong without this information: reliability coordinators commit DR capacity to cover a system peak, the event fires, 30% fewer MW materialize than planned, and the margin violation occurs. The gap was visible in the trend data. Nobody connected it to planning.
AI and AMI together make this measurement possible now. Every enrolled customer's interval meter generates baseline and event reads. The gap is computable event by event, customer by customer, in near-real-time. The data already exists; the metric just hasn't been built.
Formula
DRDG = (MW_predicted − MW_delivered) / MW_predicted × 100
Companion: Deliverability Trend:
DRDG_trend = (DRDG_current_season − DRDG_prior_season) / DRDG_prior_season
Variables:
- MW_predicted: AI-forecasted load reduction at time of dispatch decision (MW). Source: DR management platform / demand forecast model
- MW_delivered: Actual measured load reduction during the event window using the established M&V baseline methodology (MW). Source: AMI interval data, verified against M&V protocol
- DRDG_current_season: Average DRDG across all events in the current program year (%). Source: DR platform, derived
- DRDG_prior_season: Same metric for the prior program year (%). Source: DR platform, historical
Calculation Frequency: Per event; seasonal average updated after each dispatch
Output: Percentage gap, 0–100%. Zero percent indicates full delivery of predicted reduction. Values above 30% indicate the program is under-delivering against its planning basis. A rising DRDG_trend is the critical leading indicator of participation fatigue; a year-over-year increase exceeding 10 percentage points warrants immediate program review before the next reliability planning cycle.
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
When the AI says "we can call 200 MW of demand response," how much actually shows up, and is the gap growing over time as customers fatigue?