Why national EMR procurement keeps failing in African health systems

I have watched three national or state-scale electronic medical record programmes from close range, and the pattern does not vary much. A requirements document is assembled from other countries' requirements documents. Vendors respond with a feature matrix. The evaluation scores the matrix. A contract is signed against a fixed go-live date set by a political calendar rather than a facility readiness assessment. Eighteen months later, the dashboard is green and the consulting room has gone back to paper.

The procurement buys the wrong thing

A feature matrix measures what software can do. It does not measure what a nurse in a busy primary healthcare centre can do in the ninety seconds she has between patients. Those are different questions, and only one of them determines whether records get captured.

The most useful hour I have spent on an EMR selection was not in a vendor demo. It was standing in a general outpatient clinic with a stopwatch, timing the existing paper workflow: how long the folder retrieval took, where the queue formed, which fields the clinician actually wrote in and which she skipped every single time. That hour produces requirements no questionnaire produces, and it produces the only number that matters in the business case — how much clinical time the system must not cost.

Adoption is measured at the wrong level

Programme reporting aggregates. "78% of facilities onboarded" is a number that can be true while the programme is failing, because onboarding is an installation event and adoption is a habit. The number I want is per facility, per week: encounters captured in the system as a proportion of encounters that happened. That figure is uncomfortable to publish and it is the only one that predicts the next twelve months.

When we deployed EHR across Lagos State primary healthcare centres, the sites that succeeded were not the ones with the best connectivity. They were the ones where the facility manager personally checked the capture rate every week. Sponsorship at the facility level beat infrastructure, consistently.

The three failure modes, in order of frequency

• Clinical time debt. The system adds thirty to ninety seconds per encounter and nothing is removed to pay for it. Clinicians are rational; they stop.
• Data with no consumer. Staff are asked to capture fields that never come back to them as anything useful. Capture quality degrades to whatever passes validation.
• No migration story. Historic paper records stay paper, so the system holds a partial history and clinicians must consult both. A partial record is worse than either complete alternative.

What to do instead

Write requirements from observed workflow. Score vendors on time-to-complete a real encounter, performed by a real clinician on a real device on the facility's real connection — not on the count of features. Sequence deployment by readiness, not geography or politics, and let the first cohort be small enough that you can fix what it teaches you.

Then instrument adoption per facility per week and publish it internally, uncomfortably. Fund a data-quality function from day one, because it is not a project phase, it is a permanent operating cost.

And budget for the boring half: training refreshers after staff rotation, device replacement, the person who answers the phone when a printer stops working. Programmes that fund the software and not the operations are not cheaper. They just fail later, having spent more.

The uncomfortable conclusion is that EMR procurement is an operations problem dressed as a technology problem, and it is usually run by people whose job is procurement. Until the operating model is bought alongside the licence, the failure rate will not move.