Abstract Perforation intervals selection is an issue in heavy oil reservoirs since low water saturation is not directly related to low water cuts in production, thus, making it necessary to use fractional flow (fw) curves; nevertheless, for highly heterogeneous reservoirs it is useless to average an fw curve by stratigraphic or hydraulic flow units. In this paper we present continuous vertically variable fw curves, each half feet (logs resolution), as a novel approach to honor heterogeneity. Heterogeneity in the tertiary rocks of Llanos Basin, Colombia, is exhibited by dramatic changes in properties, mainly driven by diagenesis rather than mineralogy; and even when rock types were defined via FZI, there is a wide range of potential clay volume-effective porosity-permeability (VCL-PHIE-K) combinations within a single rock type. Following a series of multiple regressions, we could get unique values of fw, every half feet, for any VCL-PHIE-K combination at a given water saturation (Sw) and then, by simply averaging the fw values over the perforating intervals, it was possible to predict the expected water cut (BSW) during the first production stages. Then, a comparison of fw with BSW, in some random wells, yielded an average estimation difference of ±9 percent points, which validates the methodology and offers a much more reliable alternative to SW curves when predicting oil production. Among other results, since strong aquifers and an active water injection process influence these fields, the estimation of a water saturation before water breakthrough (Sw_BT) curve helped us understand how the highest permeability sands (RT1) were irrupted by high BSW slower compared to lower permeability sands (RT3). Consequently, the Sw cutoff (Sw_CUTOFF), estimated for a maximum fractional flow (fw_MAX) of 95%, showed averaged values between 32% and 45%, depending on rock types (RTs). We conclusively found that this methodology can help avoid undesirable high BSW, since we can first diagnose if the water is about to irrupt by using an Sw_BT curve, then we can dodge water saturated intervals above a variable Sw_CUTOFF (instead of using a constant 50% cutoff), and finally we can predict, with more than a 90% of accuracy, the BSW to manage at surface by using fw curves, rather than simple Sw curves.