This study examines complementarities between experiential and social learning in health technology adoption. Our setting allows us to test the potential of within-treatment spillovers to confound the way in which we interpret estimates from cluster-randomized controlled trials. We engage 1800 households in peri-urban Pakistan in a field experiment on water chlorination. Our experiment has four arms: control households, who receive no intervention; households who receive free chlorine tablets; households who receive tablets and small daily financial incentives for chlorine use; and households who receive tablets and an experiential learning intervention. In the learning intervention, community health workers help caregivers observe salient signals about the efficacy of chlorine tablets by visually tracking child diarrhea rates relative to control households before and after chlorine distribution. While monetary incentives generate higher chlorination than experiential learning in the short run, these effects quickly dissipate while those in the learning arm remain. These sustained effects among learning households are driven entirely by those who also have a neighbor in the learning arm. Households not in the learning arm exhibit no difference in behavior by whether they have a neighbor in the learning arm. Our results suggest that experiential and social learning not only complement, but rely, on each other. This powerful interaction between individual treatment effects and within-treatment spillovers is obfuscated by cluster-randomized designs. Comparing our results with a cluster-randomized trial testing the experiential learning arm, we find that interpreting results from that intervention as individual treatment effects would lead to a six-fold overestimation of individual treatment effects. These results have direct implications for how public health information campaigns should be administered and scaled.