Background Since the scale-up of the HIV "Treat All" recommendation, evidence on its real-world effect on predictors of attrition (either death or lost to follow-up) is lacking. We conducted a retrospective study using Zimbabwe ART program data to assess the association between "Treat All" and, patient-mix, programmatic characteristics, retention and predictors of attrition. Methods We used patient-level data from the electronic patient monitoring system (ePMS) from the nine districts, which piloted the "Treat All" recommendation. We compared patient-mix, programme characteristics, retention and predictors of attrition (lost to follow-up, death or stopping ART) in two cohorts; before (April/May 2016) and after (January/February 2017) "Treat All". Retention was estimated using survival analysis. Predictors of attrition were determined using a multivariable Cox regression model. Interactions were used to assess the change in predictors of attrition before and after "Treat All". Results We analysed 3787 patients, 1738 (45.9%) and 2049 (54.1%) started ART before and after "Treat All", respectively. The proportion of men was higher after "Treat All" (39.4.% vs 36.2%, p = 0.044). Same-day ART initiation was more frequent after "Treat All" (43.2% vs 16.4%; p<0.001) than before. Retention on ART was higher before "Treat All" (p<0.001).