What Bad Demand Forecasting Accuracy Actually Costs an Electric Utility

Demand forecasting is one of the most consequential functions a utility performs — and one of the least scrutinized until something goes wrong. When a forecast misses, the consequences rarely stop at an operational inconvenience. They compound: inflated power purchasing costs, failed demand response events, and, increasingly, five- and six-figure capacity market penalties. For utilities navigating an era of rising electrification, extreme weather volatility, and distributed energy resource (DER) proliferation, forecast error is no longer a rounding problem. It’s a balance sheet problem.

The Real Cost of Getting It Wrong

What’s in this article

• How demand forecast errors drive excess energy procurement costs
• The operational and financial impact of missed demand response event targets
• How capacity market structures amplify the penalty for under-performance
• What modern, AI-driven demand forecasting looks like in practice

1. Procurement Dollars Left on the Table — or Spent on the Wrong Power

Every day-ahead energy purchase a utility makes is downstream of a demand forecast. When that forecast is inflated, the utility over-procures — locking in power at day-ahead prices that it will not need. When it underestimates, it scrambles for real-time market energy, often at a substantial premium.

An investigation by the Institute for Local Self-Reliance found that seven of the ten largest U.S. electric utilities overestimated demand by an average of 17% over a three-year horizon — and that the error widened over longer planning windows. As RMI has noted, forecasts that overshoot reality “lead to misallocated investment and higher costs” — and in vertically integrated utility territories, those costs flow directly to ratepayers.

In competitive wholesale markets, the stakes are comparably high on the short end. Locational marginal prices (LMPs) at real-time can spike sharply when supply is tight and demand has been under-forecasted, meaning the difference between a 2% forecast miss and a 5% miss is not linear — it can translate to orders-of-magnitude cost differences during peak periods.

Accurate short-term demand forecasting, informed by hyper-granular weather variables and real-time load data, is the only reliable way to optimize energy arbitrage in this environment. Without it, energy procurement becomes expensive guesswork.

2. Missed Event Targets: When Your VPP Doesn’t Deliver

For utilities running demand response (DR) programs or virtual power plants (VPPs), the cost of bad demand forecasting shows up differently: in megawatts promised that never materialize.

The operational workflow is straightforward in theory. A grid operator sees an upcoming peak, calls a DR event, and expects (or hopes) enrolled DERs — smart thermostats, water heaters, battery storage systems — to deliver a pre-committed load reduction. But if the system load forecast that triggered the event was off, one of two things happens: the event is called too late and the peak has already been missed, or the event is called unnecessarily, burning customer goodwill and program resources for a curtailment that wasn’t needed.

The latter problem is underappreciated. Demand response programs depend on customer trust and participation rates. Unnecessary events erode both. Utilities that repeatedly over-call events due to poor forecasting often see enrollment attrition, which compounds the capacity deficiency problem downstream.

The solution is not just a better system load forecast in isolation — it’s the integration of system load forecasting with DER event forecasting. Knowing the expected system load and the expected flexible capacity from enrolled DERs, simultaneously, is what allows program managers to call the right event at the right time. With the right forecasting suite, DER Event Forecasting can predict available flexible capacity up to 48 hours before an event, giving operators enough lead time to plan — not react.

3. Capacity Market Penalties: The Bill Arrives Years Later

Capacity market exposure may be the most structurally underestimated risk of demand forecast inaccuracy. Here is why: in markets operated by PJM, MISO, ISO-NE, and NYISO, utilities and load-serving entities commit capacity resources — including demand response — years in advance of a delivery year. Those commitments are based on forecast models. When actual load behavior diverges from those models, or when committed DR resources fail to perform during a performance assessment interval, non-performance penalties apply.

The numbers are not hypothetical. During Winter Storm Elliott in December 2022, PJM assessed nearly $2 billion in non-performance penalties on 81 member organizations — a figure that was ultimately negotiated to approximately $1.25 billion. The storm exposed what market analysts had long warned: that poor resource modeling and underperformance against committed capacity levels creates systemic financial exposure.

MISO has compounded this risk environment with a significant policy shift: the penalty for non-performing load modifying resources (LMRs) has been restructured to reference the value of lost load (VOLL), which MISO recently increased from $3,500/MWh to $10,000/MWh. This means a shortfall of even a few megawatts during an emergency event can translate to tens or hundreds of thousands of dollars in penalties per incident.

Meanwhile, PJM capacity prices themselves have become dramatically more volatile. Between 2024 and 2025, PJM capacity prices rose 830% in one auction cycle, and MISO’s summer auction rate surged from approximately $30/MW-day to $666/MW-day. In this environment, accurately forecasting load — and matching capacity commitments to realistic, achievable targets — is not just a planning best practice. It is a financial obligation.

4. What Modern Demand Forecasting Must Do Differently

The traditional utility load model — built on static historical averages, seasonal coefficients, and infrequent manual updates — was designed for a grid with predictable demand patterns. That grid no longer exists.

The modern grid is defined by EV charging load, behind-the-meter solar generation, distributed battery storage, and behavioral demand flexibility. Each of these introduces non-linear, non-stationary load behaviors that legacy models cannot adequately capture. Weather volatility — an increasing feature of climate change — adds another layer of uncertainty that static models are structurally incapable of addressing.

Effective demand forecasting today requires:

• Real-time machine learning that continuously retrains on current load and weather data, rather than relying on static seasonal models
• Premium, hyper-granular weather inputs that go beyond basic temperature to account for humidity, cloud cover, wind speed, and localized microclimate variation
• DER-aware modeling that incorporates the behavior of enrolled distributed resources, not just aggregate load signals
• Human-in-the-loop oversight — because models alone, however sophisticated, require expert review to navigate unusual events, data anomalies, and market shifts
• Performance transparency — dashboards that surface forecast vs. actual load so accuracy can be continuously monitored and improved

As such, platforms that employ an advanced AI operations framework to automate model training while incorporating expert analyst oversight — what the company calls an “expert-verified” approach, are increasingly valuable to utility operations. With the ideal forecasting solution, system load forecasts are available at hourly granularity across 1-day to 1-year horizons, and integrate directly with a distributed energy resource management system (DERMS), allowing operators to schedule DER dispatch events within the same forecasting interface where they monitor system load. The goal is a single pane of glass that closes the gap between forecasting teams and program operations teams — because that gap is where missed peaks live.

What Bad Demand Forecasting Accuracy Actually Cost an Electric Utility Conclusion

Demand forecasting is not a back-office function. It’s the operational foundation on which procurement strategy, demand response program design, and capacity market obligations are built. When the forecast is inaccurate — whether due to over-reliance on legacy models, failure to account for DER penetration, or insufficient weather data granularity — the financial consequences are real, measurable, and increasingly difficult to absorb in a capacity market environment defined by historic price volatility and strengthening non-performance penalties.

The utilities best positioned to manage grid complexity and cost are those that treat demand forecasting as a continuous, AI-supported operational discipline — not a periodic planning exercise.

Glossary of Terms

• Capacity market — A wholesale electricity market mechanism through which utilities and load-serving entities procure future resource adequacy by committing generation or demand response capacity years in advance of a delivery period.
• Day-Ahead market — An electricity market in which energy is bought and sold for delivery the following day, with prices set through a forward auction process.
• Demand forecasting — The process of predicting future electricity consumption, typically using historical load data, weather variables, and behavioral models. Accurate demand forecasting informs energy procurement, grid operations, and DER dispatch decisions.
• Demand response (DR) — A utility program that compensates customers for temporarily reducing electricity consumption during periods of peak grid stress, contributing to load reduction without the need for additional generation.
• Distributed energy resources (DERs) — Customer-sited energy assets — including smart thermostats, rooftop solar, battery storage, and EV chargers — that can be aggregated and dispatched to provide grid services.
• DER event forecasting — A predictive capability that estimates the expected load reduction available from enrolled distributed energy resources ahead of a scheduled demand response event.
• Grid-Edge DERMS — A Distributed Energy Resource Management System deployed at the distribution grid edge, enabling utilities to monitor, control, and optimize behind-the-meter devices at scale.
• Locational marginal price (LMP) — The real-time or day-ahead price of electricity at a specific node on the transmission grid, reflecting the marginal cost of serving one additional unit of load at that location.
• Non-Performance penalty — A financial charge assessed against a capacity market participant that fails to deliver on its committed capacity obligation during a Performance Assessment Interval.
• Value of lost load (VOLL) — A measure used by grid operators to represent the economic cost of one megawatt-hour of unserved electricity demand; used in MISO and other markets as the basis for calculating non-performance penalties.
• Virtual power plant (VPP) — An aggregation of distributed energy resources coordinated to operate as a dispatchable, grid-scale resource, providing services such as peak shaving, frequency regulation, and capacity.