# A forecast is not a number

> Forecast accuracy matters only where it changes an action. The useful question is not how good the forecast is, but which decision it moves.

Source: https://myto.energy/insights/forecast-is-a-decision

Accuracy only matters where it changes an action. The useful question is not how good the forecast is, but which decision it moves.

Forecasting in energy is often evaluated the way a school test is marked: one accuracy score, averaged across every interval, reported monthly. It is a comfortable metric. It is also close to useless for deciding whether the forecast is doing its job.

The reason is simple. Most intervals do not require a decision. Midday on a clear day in a plant with headroom, the forecast can be several percent off and nothing changes: the same energy is exported, the same battery sits idle, the same schedule holds. The forecast earns its keep in the intervals where a different number would have produced a different action.

> Forecast quality should be measured in decisions changed, not in error averaged.

## Where forecasts actually pay

- Ramp events, where a battery either has capacity ready or does not.

- Constrained periods, where knowing curtailment is coming allows storage to absorb rather than waste.

- Peak windows, where a forecast error translates directly into a demand charge that lasts a month.

- Commitment points, where a schedule or position has to be declared ahead of delivery.

These are a small fraction of intervals and a large fraction of the money. A forecast that is excellent on average and weak at ramps is a worse operational tool than one that is average overall and reliable when conditions turn.

## Uncertainty is part of the output

A single expected value hides the thing an operator most needs: how confident the system is. Two forecasts of the same quantity, one tight and one wide, justify completely different actions. The tight one supports committing capacity; the wide one argues for holding reserve.

This is why a forecast should be delivered as a distribution and consumed by an optimiser that knows what to do with it. Handing a point estimate to a control system and hoping for the best throws away the most decision-relevant information the model produced.

## The feedback loop is the model

Forecasting improves in the places it is used and measured. When every prediction is stored alongside the decision it informed and the outcome that followed, the error stops being an abstract score and becomes an attributable cost: this much energy curtailed, this peak missed, this cycle spent badly.

That loop is the difference between a forecasting feature and a forecasting system. The first ships a number. The second gets better every week it runs.

## What we build for

Myto treats forecasting as an input to a decision, never as a deliverable in itself. The horizon, the resolution and the uncertainty are chosen to match the decisions they serve — and the value of a better forecast is measured in what changed downstream.

[Next · Markets · Grid

### Flexibility is the asset

Read →](https://myto.energy/insights/flexibility-is-the-asset) [BESS

### The value of a cycle

Read →](https://myto.energy/insights/value-of-a-cycle) [Solutions

### Solar Intelligence

Explore →](https://myto.energy/solar-intelligence)