Capability, Adaptability, Performance, Reliability — the four-axis scoring framework we use to evaluate every AI voice vendor before recommending one to an ANZ healthcare network.
The CAPR framework: how to evaluate AI voice vendors for healthcare
Most healthcare networks pick an AI voice vendor on demo polish, sales chemistry and a single peer reference. Twelve months later, ~60% are mid-replatform. CAPR is the framework we developed to make vendor selection deterministic and defensible.
Four axes. Twenty-two sub-criteria. Weighted by deployment stage.
Capability (the headline)
Does the platform actually do what your clinic needs — not what a generic demo shows?
- Multi-turn conversation depth past 6 turns.
- PMS write-back to your PMS, in production, not on a roadmap.
- AHPRA-aligned consent prompts as a first-class primitive.
- Te reo Māori or AU-accent edge case coverage.
- Outbound + inbound parity.
Failure mode: demo dazzles, production stalls on turn-4 booking flows.
Adaptability (the silent killer)
How fast can you change a flow when a clinical lead asks?
- Prompt-edit-to-production cycle time (target: <24h).
- Self-serve flow editor for ops, not just engineering.
- Versioning, rollback, A/B testing as native features.
- Custom triage keywords without vendor PS engagement.
Failure mode: every change is a 3-week vendor ticket. Compounds for years.
Performance (the boring axis)
- P50 and P95 latency under load (target: <700ms / <1.2s).
- ASR word-error-rate on AU/NZ accents on your recordings.
- Booking-completion rate on cold calls (target: >70%).
- Handoff accuracy to live agent when escalation triggers fire.
We insist on benchmarking against the network's own call recordings, not vendor-curated samples.
Reliability (where everyone lies)
- AU-region hosting + DR posture.
- Uptime SLA in writing (target: 99.9%+).
- Incident response time (target: <15 min).
- Compliance posture against APP / Privacy Act / HISO 10029 — documented, not asserted.
- Vendor financial health and 24-month runway.
Failure mode: small vendor goes bust mid-contract. We've seen it twice.
How to weight the axes
- Discovery / shortlist: 35 / 25 / 20 / 20.
- Pilot: 25 / 20 / 35 / 20.
- Network rollout: 20 / 25 / 25 / 30.
Reliability weight increases as you scale. Adaptability weight increases the longer you operate. Capability weight decreases as your flow library matures.
What CAPR is not
It is not a referral-fee disguise. We take none. The framework exists because the alternative — picking on demo polish — is what created the replatforming wave we're cleaning up across ANZ healthcare in 2026.
If you want the full 22-criterion scorecard with our 2026 vendor scores, request the shortlist tool — it's the same scoring engine, just productised.