Healthcare AI Should Start in the Store Room

Almost every AI conversation in healthcare starts with diagnosis. Image recognition, clinical decision support, the model that spots what a doctor missed.

It is the most interesting version of the subject and, for most African health organisations right now, the least useful place to begin. The problems that cost money now are further back, in stock, scheduling, records and supply.

I am not a clinician and this is not clinical advice. What I see from outside is the same pattern as every other sector: a lot of attention on the sophisticated application, very little on the administrative one that would pay next quarter.

## The problems that are already solvable

**Supply chain and stock.** Medicine that expires in one facility while another runs out is a routine occurrence across the region. It is a forecasting and distribution problem of exactly the kind these tools handle well, and the data mostly exists in procurement and dispensing records already.

**Appointment and capacity scheduling.** Clinics running at capacity with long waits, while theatre time or specialist slots go unused elsewhere in the same system. Matching demand to capacity is unglamorous and enormously valuable.

**Records and documentation.** Clinicians spend a large part of their day on paperwork. Structured capture of consultation notes returns clinical time directly, and clinical time is the scarcest resource in African healthcare.

**Claims and billing.** For private providers and HMOs, claim processing and fraud detection are pattern problems with clean historical data.

None of these require a model to make a clinical judgement. All of them free up people who do.

## The data question is different here

Patient data is the most sensitive category in the Nigeria Data Protection Act, and healthcare organisations hold it in volume, often across systems that were never designed to talk to each other.

Which makes BROAI, people using their own AI tools because nobody gave them an approved one, more serious here than almost anywhere. A hospital administrator pasting a patient list into a free chatbot to reformat it has done something ordinary and created a genuine breach. A pharma commercial team dropping prescriber data into a tool has done the same.

The people doing it are not careless. They are working in under-resourced environments with systems that make simple tasks hard. That is precisely why prohibition without provision does not work. Give them an approved route or they will keep finding their own.

## For pharma specifically

The commercial side of pharmaceutical businesses in the region looks much like FMCG: distribution, demand forecasting, field force effectiveness, trade spend. The same opportunities apply, with one additional constraint. Promotional claims are regulated, and generated marketing copy that has not been through review is a compliance exposure rather than a style problem.

## What good looks like

Start administrative. Prove the method where nobody's care depends on it. Write down which tools are approved and what patient data may never go into them, and make that rule known to the people actually handling records rather than filing it with the policy set.

Start with an honest look at the data. In healthcare it is usually more fragmented than the organisation chart implies.

A question worth asking at the next management meeting. If a member of your administrative staff pasted a patient list into a free AI tool this morning to save an hour, what in your current arrangements would prevent it, and what would tell you?

## Frequently asked questions

**Where should an African hospital or health system start with AI?** Supply chain and scheduling. The data exists, the value is measurable, and no clinical decision is involved.

**Is it safe to use AI with patient data?** Only within systems the organisation controls and has assessed against the Nigeria Data Protection Act. Patient data in a free consumer tool is a breach regardless of intent.

**Can AI help with medicine stock-outs?** Yes, and it is one of the clearest opportunities in the region. Expiry in one facility alongside shortage in another is a forecasting and distribution problem.

**Does AI replace clinicians?** No. The applications that matter most in this region return clinical time by reducing documentation and administration.

**What is different for pharmaceutical companies?** The commercial operation resembles FMCG and the same use cases apply, with the added constraint that promotional claims are regulated and generated copy must go through review.