An application on a mobile device, in response to received partial queries from users, displays suggested results based upon the received partial query, allowing the user to select a suggested result without having to input the complete query. In order to ensure that suggested results can be provided to the user in a timely manner, the application determines an expected latency of a network connection of the mobile device by periodically measuring the latency of network requests and predicting an expected future latency based upon the measured latency values. Based upon the expected latency, the application may retrieve the suggested results from a server, or from a local cache storing results of previous queries by the users as well as popular results associated with a geographic area of the user, or some combination thereof.