AI in Healthcare: What 269 Healthcare Leaders Told Us

Respondents are mostly medical schemes and insurers (46%) and providers, clinicians and managed care (19%). Nine in ten are based in South Africa, the rest in regional countries.

Survey respondent roles

A large sample spanning the industry. 269 delegates completed the survey. Medical schemes, insurers and payers account for 46% of respondents. Providers, clinicians and managed care add a further 19%. Regulators, legal and compliance, IT and digital, actuarial and benefits functions are all represented. 90% based in South Africa. Respondents from Namibia (17), Botswana (3), Zimbabwe (2), and “other” countries (4) make up the rest.

The AI Maturity Gap

Half have not yet run a pilot. 51% are not yet In the pilot stage: no active use (8%), informal individual use only (16%) or early exploration (27%). 23% are piloting or in limited operational use. 23% have multiple use cases in production. The share with AI in production is higher than the 19% recorded in our CMS conference survey 2 months earlier. The distribution is bimodal i.e. a large exploring majority alongside a group already operating at scale. A range of adoption, consistent with historic adoption patterns for other innovations and technologies, is seen.

Where does AI add most value?

Asked where AI adds most value, respondents point first to administrative efficiency (49%) and member or patient communication (27%). Claims-related uses rank high: fraud, waste and abuse detection (21%) and claims or funding decisions (18%). Specialised clinical uses rank lower: clinical decision support (11%), population health and risk stratification (6%).

Barriers to AI adoption

Capability rates as the highest single barrier but data runs a close second. The biggest single barrier is a lack of internal capability: 40% of respondents named this. Governance and legal uncertainty (23% of respondents) and poor data quality or fragmented systems (23%) are tied for second. The top four barriers account for 73% of all mentions. Data and infrastructure are more than footnotes here: data quality, privacy and weak infrastructure together make up 39% of barriers mentioned.

Barriers change with adoption maturity

Splitting respondents by maturity shows constraints shifting as organisations progress with AI adoption. Among pre-pilot organisations a lack of capability dominates (48%), with data quality secondary (22%). Among those already running multiple use cases in production the order reverses: data quality and fragmented systems become the top barrier (37%) while capability falls to 25%. Before the first pilot, the problem is skills. Once AI is in production, the problem is data quality. Support programmes must therefore differ.

Readiness is not just skills

Asked which infrastructure or readiness issue is most limiting, respondents named data availability and quality (24%) along with cybersecurity and privacy safeguards (23%) most often. Internet connectivity (18%) and electronic health record or claims system maturity (16%) follow. Only 7% say infrastructure is not a major limitation. Data and security issues rank highest in terms of readiness for AI implementation; connectivity and core systems come next.

Programmatic support should address specific barriers

Asked what support would be most useful over the next 12 months, respondents choose the foundational over the technical. AI basics for leaders and managers (38%), case studies from similar organisations (32%) and AI governance and responsible-use guidance (29%). Short webinars (38%) lead the preferences for support format but the appetite for face-to-face is strong: half-day workshops (26%) and full-day in-person training (25%). Four offerings cover 70% of the expressed need and may need to be delivered in person at least in part

In their own words

142 respondents wrote in a specific need. The most frequent requests are for support with data, reporting and integration (31 mentions) and for training and capability building (23). Member engagement and access (20) and efficiency, workflow and automation (17) are mentioned often. Fraud, waste and abuse detection appear 11 times – the most frequently named single use case. These responses mirror the closed questions: they say “help us use our data”, and “help our people learn”.

The Takeaway

Almost a quarter of respondents already run multiple AI use cases in production. But half have not reached the pilot phase, and the binding constraint shifts as organisations progress: capability and governance are an issue early on, data quality and integration increase in importance later. Support may need to be sequenced: AI basics and responsible-use guidance for the majority still in the exploratory phase, data readiness and peer case studies for those already in production, delivered in short, practical formats, often in person.