The phrase "last mile" in public health refers to the gap between a health intervention and the people who need it. You can have an effective vaccine sitting in a national cold store. That vaccine does not protect children until it reaches the community health worker who administers it to children in a village with no road.
Early in my career, I thought of last-mile delivery as a logistics problem. Get the supplies to the right place at the right time. Manage the cold chain. Ensure transport is available. These are real and important challenges.
But I have come to think of the last mile primarily as a design problem.
When I have seen last-mile delivery fail, the failure has rarely been because a truck broke down. It has been because the intervention was designed without genuine input from the communities and health workers closest to those communities. Standard dosing schedules that do not account for seasonal agricultural patterns. Community mapping exercises done at the district level that miss community-level geography. Training materials that are technically accurate and practically unusable by health workers with limited literacy.
The last mile is where assumptions that were made during programme design meet reality. And in my experience, those assumptions are often wrong in ways that are entirely predictable if you ask the right people early enough.
What works is designing with the last mile in mind from the beginning, not as an afterthought. This means involving community health workers in programme design, not just programme implementation. It means testing interventions in the field before scale-up, in the communities where they will actually be used, not in pilot sites that have been selected for their accessibility.
It means recognising that the most knowledgeable person about what will and will not work in a community is usually someone who works in that community, and building that knowledge into the programme rather than overriding it with expertise from outside.
The last mile is hard. But many of its difficulties are preventable, if we design for them rather than discovering them during implementation.