Many agricultural extension services in developing countries continue to provide blanket recommendations to farmers despite large heterogeneity in local agronomic needs. More fine-tuned agricultural management could increase input profitability and limit environmental impacts. Mobile phone extension has the potential to reach farmers at scale at a low-cost, while also providing them with personalized information. We conduct a randomized control trial of a phone-based extension program with a sample of 6,000 smallholder Kenyan farmers recruited through input shops. We construct predictive maps using soil data and provide farmers with area-specific information about soil acidity and use of agricultural lime, an input used to increase soil pH. Using survey data and administrative records collected from the redemption of discount coupons distributed to farmers in treatment and control arms, we find that the intervention increased the probability of following the local recommendations by 3-7 percentage points (corresponding to a 10%-23% increase). We find that the program was effective at increasing the use of agricultural lime in high-acidity areas and reducing it in non- acidic areas.