How to turn a forecast into an operational decision

At a shrimp farm, someone checks the rain forecast before finishing work for the day. There is a chance of a downpour the following night. The manager knows the ponds, knows which ones need more attention and has several preparation measures available. The question is how much it is worth spending today on something that may not happen.

The same question arises when a farm schedules an agricultural application or a company plans outdoor maintenance. There is a job to be done, conditions that could affect it and someone who has to decide before knowing what the weather will be.

A forecast can help. To use it, the company must connect what it predicts to an action it can take and to the consequences of making the wrong choice.

Start with a decision that already exists

Consider the shrimp farm as a hypothetical example. When heavy rain is possible, its technical manager can determine what preparation is appropriate for the current state of the ponds and how much time it will take. Checking the aeration equipment and organising staff are two possibilities; the specific measures depend on how that farm is managed.

Before looking for a probability in the forecast, the manager must define what kind of rain is a concern. Any rain? A certain accumulation overnight? An event that, given the current state of the ponds, would require different preparation than usual? The probability of any rain may answer a very different question from the one the manager needs to resolve.

The manager must also confirm that there is still time to act. If preparing the farm takes several hours, the forecast must be available that far in advance. Information that arrives later may help explain what happened, but it can no longer support that decision.

This defines a concrete problem: for a specified event, decide whether to carry out a known preparation before a set time.

How much preparation costs and how much loss it can prevent

Suppose the preparation costs $300. The farm incurs that cost whether or not it rains. Suppose also that the preparation can prevent a $10,000 loss if the event of concern occurs. These figures are invented to explain the calculation.

With a probability of 5%, the avoidable loss weighted by that probability would be $500: 0.05 × 10,000. In this simplified example, spending $300 on preparation makes economic sense.

The break-even point is found by dividing the cost of preparation by the loss it can prevent:

300 ÷ 10,000 = 0.03, or 3%.

Above 3%, the expected benefit exceeds the cost. At exactly 3%, the two are equal under the assumptions of the example. The 5% figure was a forecast probability used to evaluate the decision; the calculated threshold remains 3%.

The word “prevent” matters. If the preparation reduces only part of the damage, that portion must be used in the calculation. Dividing by the entire possible loss would make a measure appear more worthwhile than it really is. Nor is the price of materials the only cost: staff time and work left unfinished may also need to be included.

This calculation compares expected monetary costs. It is a starting point for a repeated decision under simple assumptions; other operational constraints may require a different analysis. The probability used must also have been evaluated for that event and location. A correct calculation cannot make up for an unsuitable forecast.

Two companies may act differently

A farm that can prepare at little cost has reason to act at relatively low probabilities. The same preparation may cost another farm much more, or prevent less damage. Even if they receive the same forecast, their thresholds will differ.

The person responsible for the operation should therefore be involved from the start. They can estimate what may be lost, what portion can be protected and what it takes to mobilise staff. The forecast provider contributes the weather information and explains its limits. The rule needs both kinds of knowledge.

The reasoning applies to other work, but the figures must change with it. For an agricultural application, rescheduling may carry an agronomic cost as well as the cost of the equipment. For maintenance, avoiding an unproductive day may justify changing the date, as long as the change does not create a larger problem elsewhere.

Review a rule after using it

Return to the farm. It prepared, but in the end it did not rain. The expense is real, and it is reasonable to ask whether it was worthwhile. That single outcome does not show that the decision was wrong: a probability of 5% left plenty of room for the event not to occur.

To evaluate the rule, keep the forecast that was available when the decision was made, the action taken and what happened afterwards. Once several events have been recorded, the company can review how much it spent on preparation, what losses occurred and whether its assumed costs were close to the actual costs. It can also check whether the forecast probabilities correspond to the observed frequency of the events.

There will be reason to change the rule if preparation costs more than expected, protects less than expected or is triggered by unreliable information. Keeping these records allows the company to assess the rule using more than the memory of the last time it did not rain.