There is growing demand for disaggregated estimates for business statistics at small areas or domains. Small area estimation models however break down when applied to business statistics due to skewed distributions and outliers. Since many countries have business registers where covariate information is available, the nested-error unit-level small area model under the log-transformation will be applied. Generally, small area predictors do not add up to the design-based direct estimator of the total computed for a large (and planned) domain. Several methods for calibrating small area predictors have been proposed for the linear model and we review and propose methods for calibration under the log-transformation. Empirical results based on simulated data from the unit-level model are presented. This is joint work with Rodolphe Priam.