National-scale carbon footprints of livestock production are commonly computed from a set of production system characteristics that serve as inputs for greenhouse gas (GHG) emission models. We evaluated the feasibility of using such equations at a finer scale to derive a simple farm-scale indicator of emission intensity (milk yield per head). Using probabilistic simulations, we quantified the impact of input variable uncertainty on emission estimates for smallholder dairy farms in Kenya. We simulated emissions for farm-scale scenarios generated from a survey of 414 households and published or expert-estimated uncertainty bounds. We simulated the impacts of five interventions: changing breeds, retiring unproductive males, keeping fewer replacement males, feeding forage supplements, and balancing animal diets. Impacts were assessed against a true counterfactual and against a more realistic scenario affected by random effects. We estimated errors incurred in classifying farms into adopters and non-adopters of the innovations based on changes in milk yield per animal. Given the current uncertainty, such classification would either miss a large percentage of adopters or misclassify many non-adopters as adopters. As a critical uncertainty, we identified the milk yield of dairy cows. Added precision on this metric reduced but did not eliminate classification errors. We remain cautiously optimistic about using milk yield per head to proxy emission intensity, but its effective use will require further reduction of critical uncertainties. Replacing generic recommendations of parameter uncertainties with context-specific error estimates might lead to a more efficient quantification of the carbon footprint of milk production on smallholder farms.