I would recommend thinking in dollars, like the company does.
If:
- the cost of intervening pre-emptively is c_{pre}, and
- intervening preemptively has a 100% success rate
then expected cost of intervening preemptively at period t is c_{pre}.
If:
- the cost of intervening after infestation is c_{post}, and
- the probability of infestation in period t+1 is p
then the expected cost of not intervening preemptively at t is p*c_{post} + (1-p)*0
So the company would intervene at t whenever c_{pre} < p*c_{post}
I would apply this logic in a cross-validation setting. Fit your model on 80% of the dataset. Use the fitted model to predict p for the holdout. Plug in placeholder values for c_{pre} and c_{post} (e.g. $20 and $100, or ask the company for guidance on this ratio), then simulate what the company would have done for the holdout crops. Did they save money by intervening preemptively on the right crops?