How Vantra could extend Penda Health's AI-first care across the whole clinic, not just the front desk.

Penda Health operates a network of primary-care Medical Centres across Nairobi, built around a simple promise: high-quality, affordable care delivered consistently at scale. In 2025, Penda did something few health systems anywhere have managed — it ran a serious, governed study of AI in real consultations.

Working with OpenAI, Penda built a clinical copilot called AI Consult, embedded in its medical record. Across 39,849 patient visits at 15 clinics, clinicians using the tool saw a 16% relative reduction in diagnostic errors and a 13% reduction in treatment errors, with the work reviewed by independent physicians and approved by AMREF Health Africa, the Kenyan Ministry of Health and Nairobi County. It is one of the most credible real-world demonstrations of AI in primary care to date.

AI Consult addresses the clinical layer — helping the doctor inside the consultation. But a fast-growing network loses time and patients in two other places that the same AI-first thinking can fix.

The first is documentation: clinicians spending hours each day writing up notes instead of seeing patients, often into the evening. The second is the front door: calls ringing out while reception serves the person at the desk, and WhatsApp enquiries — 'Are you open today?', 'Do you take SHA?', 'How much is a consultation?' — sitting unanswered until the patient books with whichever clinic replied first. Neither is a clinical problem. Both are productivity and capacity problems, and at network scale they quietly cap growth across every branch.

Vantra would bring three complementary products to the network. Vantra Scribe would capture consultations ambiently and produce structured draft notes for clinicians to review, edit and sign; it would not file them automatically. Vantra Concierge would provide configured phone and WhatsApp coverage, collect non-clinical intake, recover eligible missed calls, send approved reminders, and pass booking requests through a validated calendar connection or staff handoff. Vantra Central Command would give staff a governed knowledge interface over approved Penda material, presenting sources separately from model inference.

All three would follow an assistive principle. The clinician controls every Scribe note; Concierge hands anything urgent, sensitive or clinical to a human under configured rules; and Central Command shows the basis for an answer rather than presenting inference as source material. None of the products gives medical advice.

The scenario would measure whether clinician-reviewed drafts reduce documentation time, whether configured communication coverage improves answered-contact and booking rates, and whether governed knowledge access reduces time spent searching for approved information. It would not assume an effect on no-shows or guarantee a response to every patient.

The through-line matches Penda's evidence-led approach: assist the team, preserve clinical judgement, and evaluate results against agreed operational measures.

A sensible first deployment would cover missed-call recovery and after-hours WhatsApp coverage — the moments when enquiries leak straight to competitors. Success would be measured in answered-contact rate, time-to-first-response, bookings per 100 enquiries, and no-show rate, tracked consistently across branches.

Penda has already proven it will adopt AI when it is designed to assist rather than override, and measured honestly rather than assumed. The patient communication layer is the natural next place to apply that same discipline — and the one where responsiveness translates most directly into more patients reaching great care.