For years, if you had asked “how much does it actually cost to host the Community Health Toolkit (CHT),” Medic’s answer would have been…
“Well, it depends.”
For as long as I’ve worked at Medic, it’s a question we’ve wrestled with. It is a real-world problem for the ministries of health, NGOs, and implementing partners who rely on the CHT: they need to budget properly and realistically, targeting a sustainably sized program for the population they serve.
While drawing up a hosting budget might seem straightforward, things can quickly get difficult to nail down on the technical side, especially when you factor in the necessary hardware. There’s a ton of variables when it comes to how much hardware you might need, starting with considering how the CHT is going to be used and what the program might look like in two, five, or ten years from now.
Four years ago, the most accurate figure we could reach based on the best data we had at the time, was 10 cents (USD) per community health worker per month. Yet we knew this was still approximate, and therefore hard to trust and harder still to guarantee for any one country’s national-scale community health system.
A solution three years in the making
Things began to change when we created the CHT Watchdog system. For years, it has collected granular data on elements like how many users are connecting to the system, how much disk space is being used, and the number of documents in a database. By comparing months of data across 30 instances from different contexts, we started building a much more accurate picture of what it looks like to host a production CHT instance.
The hard part was done. Now we needed an easy way to present that data for a non-technical audience. That was where the seed of the cost calculator idea first started growing: an accessible way for a non-technical person to get a high-level grasp of what their costs might be and how certain changes to their configurations, workflows, and population served might affect those costs.
The last mile can be the hardest to reach
Now, I’ll admit, I have much more experience with back-end programming, and when it comes to making software and tools look user-friendly and intuitive, it takes a lot of effort and hours spent tinkering with design parameters that I often don’t fully understand.
I’m also not an expert on querying CHT Watchdog historical metric data. It takes a lot of time and effort to produce the complex queries needed for measuring cost impact.
That’s where AI came in. It built and iterated with me on creating a user interface, testing which parameters mattered, what values users would actually want to include in their cost measurement, and what should surface at the end. It also helped me write the complex queries needed to slice years of production data, and check the output against my own understanding rather than taking it on faith.
Suddenly, a month of developer time became something I did in a week. That being said, the calculator is not a “side-project” of questionable quality. Every design decision, data model, and calculation was reviewed by myself and another engineer. We carefully confirmed the quality of the code, the correctness of the assumptions and the reproducibility of the results before publishing.
This is where AI is already proving its value across the digital health sector right now: helping close the gap between good, queryable data and sustainable systems for communities that need them most.
Numbers people can always trust
Cost claims are easy to make, but often hard to verify. Ours had to be defensible and as closest to reality as possible, as national programs depend on them. We built the calculator on values chosen not to make the CHT look cost-effective, but to hold up to that scrutiny.
When we evaluated a typical deployment supporting 1,000 health workers delivering care for 100,000 people through the calculator, the result was nine cents (USD) per CHW per month. Close to the old estimate, but this time built from years of production evidence. And, critically, that number adjusts to a partner’s actual context.
One caveat: the calculator determines cloud hosting costs, a critical budget line for any digital health system, but not the whole picture. A well-functioning digital community health system also needs devices, connectivity, and support staff, and those must be budgeted separately.
Partners in Nepal are already using the calculator to plan new and expanding deployments. And the underlying costs are already moving: a recent shift to newer AWS server cores is expected to cut processing costs by around 20%.
At Medic, we want that number to keep going down, with a leaner CHT platform that gets more efficient as it scales, and we are working hard to achieve it. AI-powered development is a huge part of this process, already showing results like MCP servers giving agents precise contextual data about the CHT, automated code reviews, and much more!
Building the future in the open
Right now, the best practices and most reliable procedures for integrating AI into the process of writing and maintaining software are being actively developed, debated, challenged, and refined.
As open source maintainers, we believe the most effective way to build software that authentically addresses the needs of humans is to build in the open with a diverse community. Everyone’s unique perspectives and contributions are leveraged for the benefit of the whole ecosystem. With emerging AI technologies, we want to empower people, regardless of their technical expertise, to rapidly advance the platform while continuing to honor the weight and responsibility of deciding every spec that’s safe to ship, for better health in the communities we serve.