This paper studies the problem of demand driven employee scheduling for direct-sales retail outlets. The demand for sales personnel during each half hour is predicted a week in advance based on previous recorded data and the problem of computing an optimal schedule to meet this demand is posed as a mixed-integer linear programming model. This model is incorporated into a commercially deployed online scheduling package, which has been in use at over 38 retail outlets in Switzerland over the last year. The key difference in the proposed modeling approach is that no complex heuristics or decompositions are used; the full model is formulated and solved. This is possible due to the relatively small scale of the target stores and leads to significant improvements in schedule quality. This quality is assessed through comparison with previous manually gener- ated methods in these 38 stores and an average reduction of 10% is seen in over- and under-staffing rates. Furthermore, a comparison is made against a large-scale scheduling system used in the UK  and a 4.4% improvement is demonstrated.