The starting point
The carrier was running pre-authorisation through a hybrid of clinical reviewers and rules engines. Median SLA was 24 hours; 9% of cases breached. Network providers were vocal; member NPS on hospitalisation journeys was 18 points below the rest of the book.
What we deployed
An agentic pre-auth workflow that ingests the provider request, extracts the clinical context, matches against policy benefits and network terms, runs the carrier's existing rules, and issues a decision with the full reasoning chain. Cases under a confidence threshold are routed to clinical reviewers with the AI's analysis pre-attached.
- Clinical document extraction across Arabic and English
- Policy + network + benefits match against the core admin system
- Rule-based and model-based decisioning combined
- Append-only audit log on every decision, regulator-aligned
How we worked
Eight-week deployment. Senior insurance and AI team embedded with the carrier's clinical and operations leaders. Shadow mode in week 4, assisted in week 6, authoritative within confidence bounds in week 8. No juniors on the engagement.
Outcomes
Median pre-auth decision dropped from 24 hours to under two minutes on 82% of cases. SLA breach rate fell from 9% to under 0.5%. Member NPS on hospitalisation journeys recovered the 18-point gap within two quarters. Clinical reviewers redeployed to complex cases where their judgement actually matters.