Cases and evidence

Arkangel AI customer cases and studies

This page brings together published Arkangel AI use cases and studies on healthcare, clinical research, and pharmaceutical operations. Each study names its journal or conference and year, and each case links to its record with methods and limitations. Several study authors are part of the Arkangel AI team.

Customer cases

Pharma and provider · Colombia · Clinical pathway

ProPacífico, with BMS support: gastric cancer clinical-pathway adherence

Arkangel AI's Pandora audited a cohort against the clinical pathway to surface high-risk patients missing expected care, with medical review.

clinical records reviewed
6,530clinical records reviewed
patients classified as high risk
436 (7%)patients classified as high risk
of high-risk patients with no endoscopy ordered or performed
38%of high-risk patients with no endoscopy ordered or performed

Reported pilot figures; the case publishes its method and limits.

Read the ProPacífico case

Pharma · LATAM · Confidential

Confidential pharma case: from field visits to brand decisions

How a regional pharma operation connected field signals, physician questions and adoption to decide where to act.

Spanish-language document, restricted to corporate teams.

Request the case

Published studies

Evaluations of Arkangel AI's models and agents. Several authors are on the Arkangel AI team; we say so where it applies.

Mayo Clinic Proceedings: Digital Health, 2026

Arkangel AI, OpenEvidence, ChatGPT and Medisearch against medical standards

128 question-answer pairs and 1,024 blinded external evaluations. 100% reference authenticity for Arkangel AI-Deep, as for ChatGPT-Deep and OpenEvidence. The authors are Arkangel AI staff.

Read the study

Intelligence-Based Medicine, 2026

Randomized trial with medical students

Double-blind randomized trial with 83 students: ~55% less time per case (median 2:35 vs 5:44) and higher answer-validity scores.

Read the study

Intelligence-Based Medicine, 2025

Conversational agent for evidence-based medical questions

90.26% accuracy on MedQA (vs 87.51% for GPT-4o).

Read the study

IEEE Conference on Artificial Intelligence (CAI), 2025

PANDORA: unstructured clinical data extraction and COPD risk

Near-perfect extraction and 94% accuracy on the PUMA score.

Read the study

Talk to us about your case

Every conversation starts from your batch or your question, not a slide deck.

Payers and insurers

Audit pilot on your anonymized claims.

Plan a pilot

Pharma

Physician questions, doubts and barriers by molecule, specialty, disease area and country.

Talk to the pharma team

Hospitals and providers

Denials review on a sample batch.

Plan a review

Clinicians

MedSearch: clinical answers with cited sources. Free plan.

Try MedSearch