WHO AFRO uses a Claude skill to cut Ebola sitreps to under an hour

Anthropic published a feature page on its site describing how teams fighting a Bundibugyo ebolavirus outbreak in eastern Democratic Republic of the Congo use Claude. The most reusable detail is a skill built by three data staff at WHO AFRO, the World Health Organization's Africa regional office: it extracts figures from district slide decks and drafts the daily situation report. Anthropic says the report used to take all day and now takes under an hour. This article reports the page; I have not seen the skill or WHO's own account.

The workflow it replaces

According to the page, health facilities send case information to their districts, often over WhatsApp. Each district puts it onto PowerPoint slides and forwards them to the outbreak coordination centre. At the provincial health ministry, staff compile those decks into the published situation report, often working late into the night.

The page credits Tendai Muza, Tamayi Mlanda and Gianni-Ferrari Donkor Muza at WHO AFRO with a Claude skill that does four things. It pulls case and lab numbers out of each health zone's deck, checks them against the previous day's report, flags any change in the trend with a possible explanation, and summarises the districts' reports.

What else the page reports

• Modelling: Paul Ouma of WHO AFRO is quoted saying the team could previously run only one disease model because of time. The page says they can now run several at once and build forecasts, for example to help logistics staff decide where to build treatment centres.
• Vaccine work: the Coalition for Epidemic Preparedness Innovations (CEPI) used Claude to build a dashboard tracking the actions teams need to take on vaccine development. Polina Brangel, CEPI's R&D data innovation lead, is quoted saying Claude organises complex data so experts can compare proposals faster, and does not decide which serological cohorts to analyse.
• Sequencing: the page says INRB, the national biomedical research institute, uses Claude Science, an AI workbench, to assemble virus genomes and build family trees from plain-language prompts instead of terminal commands.
• Next steps: WHO AFRO is using Claude to clean and validate line lists for a chikungunya outbreak, and wants to link historical disease databases so the next response can draw on earlier ones.

What the page leaves out

It is a vendor-published feature, and it reads as one. It states the time saved but not how accuracy was measured. It does not say how often Claude misread a slide, who reviews the draft before publication, or what the work costs. It does not say whether the 'under an hour' figure includes that review. Those are the questions an operations lead would need answered before copying the pattern.

The setting is also unusual: an emergency, a partnership convened by CEPI with WHO AFRO and INRB, and direct support from Anthropic's Beneficial Deployments and Applied AI teams. A team without that support may see different results.

What a team can take from it

The shape is general: many contributors send structured numbers in an inconsistent format, and one person spends hours reconciling them. A skill that extracts, compares with yesterday and flags changes is a bounded task you can test. Before a pilot, replay a few weeks of past inputs, compare the extracted figures with the published ones, and record every discrepancy. Keep a named person accountable for the final report, as the CEPI quote implies.