Text this: Designing Randomized Experiments to Predict Unit-Specific Treatment Effects