Why distribution operators buy too much standby power and pay the price

Understanding the cost of excess standby power does not, by itself, change operator behaviour. The harder question is why the pattern persists despite that awareness. Three structural reasons recur, and recognising them is essential before choosing a technical remedy.

Reason one: unequal consequences

Under-procurement produces a visible, immediate record. If an operator reserves too little, comes up short at peak demand, and must call emergency reserves or buy in the intraday balancing market at elevated prices, the event enters the operational log. It may prompt regulatory review and will certainly feature in the post-incident analysis.

Over-procurement has a quieter, dispersed cost. Contracted standby capacity that never dispatches sits in the monthly operations budget, hidden inside total reserve procurement costs. It is not linked to the particular forecast that proved too pessimistic on a particular day. No "over-procurement incident report" exists, so the financial effect quietly builds over hundreds of operating days each year.

That imbalance reasonably favours buying too much. A reserve shortfall is concentrated and conspicuous, while excess reserve is spread out and hard to see. Operators judged mainly by security of supply will naturally select the option that avoids the obvious failure, even when accounting for both outcomes shows it costs more overall.

Reason two: unmeasured forecast uncertainty

A single-number load forecast can express uncertainty only through its reserve margin. When the duty engineer is unsure about tomorrow's demand, they enlarge the buffer around the point forecast. If the forecasting tool also gives no estimate of uncertainty, they use a cautious rule: choose a margin historically broad enough to cover most surprises.

That cautious rule reflects the worst situations the engineer has encountered or heard about, rather than the average day. When demand follows its expected pattern, the margin is excessive. Yet the engineer cannot know beforehand whether a day merely looks typical or will actually be typical, so the full conservative buffer remains in place.

This produces continuing structural over-procurement, although most days' actual demand variation could support a smaller reserve. The issue is not irrational decision-making but limited information about that day's particular uncertainty. Probabilistic forecasting makes this visible: a narrow P10-to-P90 range shows confidence, allowing reserves to sit nearer P50, while a wide range on a more uncertain day indicates that a larger margin is appropriate.

Reason three: misaligned organisational incentives

At many distribution operators, security of supply belongs to operations while commercial cost control sits with finance or commercial operations. Operations is accountable for preventing network capacity shortfalls. The commercial function must keep procurement spending within budget.

If those functions are not closely connected during day-ahead planning, operations treats over-procurement as the safer option because its cost is not part of that team's direct metric. Commercial staff see the budget effect but cannot trace it to individual forecast choices. Neither side sees operational context and financial feedback together.

This pattern has appeared among operators we have worked with. The duty engineer deciding the morning dispatch is concentrating on network coverage, not the EUR cost of the standby contract. Better tools can help, provided they display procurement cost beside operational risk at the decision point. Showing P90 demand together with the estimated EUR cost of procuring to P90 rather than P70 gives the engineer practical information that links supply security with operating cost, unlike a demand forecast alone.

What changing the behaviour really requires

A better load forecast is needed, but it is not enough. Users must also trust what it produces. That confidence comes from sustained accuracy on the feeder topology they manage. Someone who has relied on the same persistence model for 8 years and empirically calibrated reserve margins against it will not instantly accept a new model's confidence estimates, even if validation results are better.

Adoption is gradual. In our pilots, operators generally run the new forecast beside their current process for four to six weeks, compare results, learn the probabilistic band format, and confirm that P90 behaves as expected, with demand below it most of the time. They then try narrower margins on lower-risk days, such as stable forecasts with narrow bands, mid-week periods, and no major weather events. Savings build as confidence grows.

We are not saying that any operator who deploys better forecasting will immediately eliminate over-procurement. Structural incentives need time to shift. Still, showing quantified day-ahead uncertainty clearly at the decision point gives operators the information required for better choices. What follows is change management, not a technical problem.