'Creaming' and 'parking' are endemic concerns within quasi-marketised welfare-to-work (WTW) systems internationally, and the UK's flagship Work Programme for the long-term unemployed is something of an international pioneer of WTW delivery, based on outsourcing, payment by results and provider flexibility. In the Work Programme design, providers' incentives to 'cream' and 'park' differently positioned claimants are intended to be mitigated through the existence of nine payment groups (based on claimants' prior benefit type) into which different claimants are allocated and across which job outcome payments for providers differ. Evaluation evidence suggests however that 'creaming' and 'parking' practices remain common. This paper offers original quantitative insights into the extent of claimant variation within these payment groups, which, contrary to the government's intention, seem more likely to design in rather than design out 'creaming' and 'parking'. In response, a statistical approach to differential payment setting is explored and is shown to be a viable and more effective way to design a set of alternative and empirically grounded payment groups, offering greater predictive power and value-for-money than is the case in the current Work Programme design.