Within countries, poorer populations have greater health needs and less access to good medical care than better-off populations. This is particularly true for tuberculosis (TB), the archetypal disease of poverty. Innovations also tend to become available to better-off populations well before they become available to those who need them the most. In a new era of innovations for TB diagnosis and treatment, it is increasingly important not only to be sure that these innovations can work in terms of accuracy and efficacy, but also that they will work, especially for the poor. We argue that after an innovation or a group of innovations has been endorsed, based on demonstrated accuracy and/or efficacy, introduction into routine practice should proceed through implementation by research. Cluster-randomised pragmatic trials are suited to this approach, and permit the prospective collection of evidence needed for full impact assessment according to a previously published framework. The novel approach of linking transmission modelling with operational modelling provides a methodology for expanding and enhancing the range of evidence, and can be used alongside evidence from pragmatic implementation trials. This evidence from routine practice should then be used to ensure that innovations in TB control are used for positive action for all, and particularly the poor.