Peak shaving reduces a utility's maximum demand charge exposure and defers the capacity additions that would otherwise be triggered by peak load growth. The toolkit for achieving it includes battery storage dispatch, flexible load management, demand response activation, and switching operations. AI feeder forecasting changes the calculus on each of these strategies, though not in the same way or to the same degree.
Battery storage dispatch
Utility-owned or utility-contracted battery storage is the most direct peak shaving tool available today. A battery that can discharge for 2 to 4 hours at a rate sufficient to shift a feeder's peak below a target threshold is a near-certain peak shaving asset, assuming the dispatch timing is correct. The forecasting requirement is modest: you need to know with reasonable confidence that the peak event is coming, approximately when it will start, and how long it will last. A 4-hour ahead forecast with reasonable accuracy handles this.
Where forecasting accuracy matters more is in avoiding unnecessary discharge. A battery that discharges in preparation for a peak that does not materialize at the expected level is a battery that is unavailable for the next opportunity. Better forecasting reduces false-positive dispatch events and extends effective battery utilization across a planning period.
Flexible load management
Direct load control programs, where the utility can interrupt or modulate specific customer loads (water heaters, EV chargers, HVAC units) via smart meters or control hardware, can provide several megawatts of dispatchable load reduction on a feeder with high enough program enrollment. The timing challenge is knowing when to activate this resource without over-activating it in ways that reduce customer participation.
A 4-hour ahead feeder forecast allows direct load control activation to be staged: pre-conditioning actions (shifting water heater cycles forward to build thermal storage) can begin 3 to 4 hours before the projected peak, with harder interruptions reserved for confirmed peak conditions. This staging approach gets better peak reduction with less customer impact per megawatt-hour of load shed.
Demand response
Large-customer demand response programs typically require 1 to 4 hours of activation notice. At 4-hour ahead accuracy levels, the forecast is reliable enough to trigger demand response activation with reasonable confidence that the event conditions will materialize. At 24-hour ahead accuracy levels, demand response is often pre-scheduled against forecast peaks rather than dynamically activated, which means it fires on some events that do not develop to the projected level.
The financial implication: demand response events that do not develop to projected levels still consume the utility's contracted event budget. Most demand response contracts limit the number of events per season. Using that budget more precisely means reserving it for events that actually require it.
Switching operations
Load shifting through feeder switching and tie-line operations is the fastest-acting peak shaving tool available at the distribution level and requires no contracted assets. It does require accurate knowledge of which feeders have available capacity headroom at the time of the switching event. A real-time view of all feeders' current and projected load levels is what enables the dispatcher to identify and execute beneficial switching operations without inadvertently overloading the receiving feeder.
This is the peak shaving application where Ampgrove's dispatch recommendation engine is most directly useful: monitoring all feeders simultaneously, identifying developing peak conditions, and surfacing specific switching recommendations with projected outcomes before the operator would typically see the developing condition in their normal monitoring workflow.
The coordination challenge across multiple strategies
Most utilities with developed peak shaving programs use multiple strategies simultaneously, and coordinating them across a single event window is more complex than deploying each individually. Battery storage discharge timing needs to align with demand response activation to avoid double-stacking responses on a peak that does not reach projected levels. Switching operations to shift load between feeders need to account for the load that demand response will remove from the sending feeder, otherwise the switching recommendation may create unnecessary load imbalance.
Ampgrove's dispatch engine tracks all available response resources alongside feeder load projections specifically to support this kind of coordinated response. A utility that has battery storage on feeder A, demand response enrollment on feeders A and B, and available tie-switch capacity between feeders B and C can see a recommendation that stages these resources in sequence as a projected event develops, rather than activating each independently based on separate threshold triggers.
Measuring peak shaving effectiveness
Measuring whether peak shaving programs are delivering value requires clear baselines and consistent measurement methodology. The fundamental challenge is counterfactual: what would the peak have been without the intervention? The standard approach is weather normalization, comparing the actual peak against a modeled peak for the same weather conditions using a baseline period before peak shaving measures were active.
Ampgrove's platform generates pre- and post-intervention load forecasts for every event window where a dispatch recommendation was acted on. The difference between the forecast load without intervention and the actual load after intervention is our estimate of the peak reduction achieved. This measure is not perfect -- it inherits the forecast's accuracy limitations -- but it is more defensible than simple before-after peak comparisons that do not control for weather differences.
The financial case for peak shaving at different utility sizes
The economic case for investing in sophisticated peak shaving infrastructure scales with feeder count and peak period exposure. A small municipal utility with 30 feeders and relatively stable load gets less total financial benefit from peak shaving optimization than a larger investor-owned utility with 200 feeders and significant EV-driven load growth. The investment threshold for automated forecasting and dispatch support needs to be calibrated to the scale of the financial benefit.
For utilities in the 25 to 150 feeder range, which is the core of Ampgrove's current market, the capacity reservation cost reduction from better forecasting typically exceeds the subscription cost by a margin of 3 to 8 times in our pilot data. The threshold for positive ROI is lower than utilities expect, partly because the comparison is against the total cost of over-provisioning, not just the avoided capacity reservation contract cost. When you include capital deferral on infrastructure that does not need to be built for 2 to 3 additional years, the economics for mid-sized utilities are substantially better than a simple operating cost calculation shows.