In 2004 the Government in England and Wales published a new policy on responding to women who offend. The aims were to reduce women's involvement in crime and to divert them from prison. The 'Together Women' project was funded under this policy initiative to demonstrate how services for women offenders should be provided in the community. The first stage of the associated evaluation included interviews with Together Women's clients as their feedback was seen as important in helping to develop effective services and as an early indicator of impact. However, the final assessment of impact relies on a quantitative assessment based on project files and criminal records data. The only interviews to be conducted will focus on asking sentencers about whether they use Together Women to divert women from custody. This article draws on interviews conducted with Together Women clients in the project's development phase to argue that outcome evaluations which rely exclusively or mainly on information in project databases and criminal records may not capture key elements which make an intervention 'work'. Neglecting service users' insights may lead to under-estimating resource needs, unrealistic target setting, and the eventual abandonment of promising ideas in favour of the next 'new' magic bullet. (Published Abstract) Notes and references