Ecological communities used in biological pest control are usually represented as three-trophic level food chains with top-down control. However, at least two factors complicate this simple way of characterizing agricultural communities. First, agro-ecosystems are composed of several interacting species forming complicated food webs. Second, the structure of agricultural communities may vary in time. Efficient pest management approaches need to integrate these two factors to generate better predictions for pest control. In this work, we identified the food web components of an avocado agro-ecosystem, and unravelled patterns of co-occurrence and interactions between these components through field and laboratory experiments. This allowed us to predict community changes that would improve the performance of the naturally occurring predators and to test these predictions in field population experiments. Field surveys revealed that the food-web structure and species composition of the avocado community changed in time. In spring, the community was characterized by a linear food chain of Euseius stipulatus, an omnivorous mite, feeding on pollen. In the summer, E. stipulatus and a predatory mite, Neoseiulus californicus, shared a herbivorous mite prey. Laboratory experiments confirmed these trophic interactions and revealed that N. californicus can feed inside the prey nests, whereas E. stipulatus cannot, which may further reduce competition among predators. Finally, we artificially increased the coexistence of the two communities via addition of the non-herbivore food source (pollen) for the omnivore. This led to an increase in predator numbers and reduced populations of the herbivore. Therefore, the presence of pollen is expected to improve pest control in this system.