Specialty-Pharmacy Prior Authorisation at a US PBM — From 6-Day Cycle to Same-Day on Cell-and-Gene Therapies
Prior Auth Accelerator + Medical Records Parser + Rx Compliance Stocker reshaping cell-and-gene-therapy prior auth at one of the largest US specialty PBMs.
The challenge
The client — a US specialty pharmacy benefits manager managing the prior-auth workflow for a large network of payers across cell-and-gene therapies, infusibles, oncology specialty drugs and orphan-disease therapeutics — was facing a turnaround-time crisis on the highest-stakes prior-authorisation cases in US healthcare. The PBM's specialty-pharmacy book covered approximately 240 high-cost specialty drugs with average per-course costs in the hundreds of thousands of dollars and patients whose clinical condition typically did not tolerate the prior-auth delay.
Average prior-auth turnaround across the specialty book was 6 calendar days, with the longest cycles reaching 14 days on the cell-and-gene-therapy authorisations that required the deepest clinical evidence review. State-regulator pressure on specialty-pharmacy prior-auth timelines was tightening, and the PBM's payer clients were threatening contract renegotiation if turnaround did not improve.
The constraints were specific. The specialty-pharmacy clinical-criteria library was complex (each drug having a multi-page evidence-of-medical-necessity criteria document, often updated as new clinical evidence emerged), the clinical documentation submitted by prescribers was variable in quality and structure, and the clinical-review team — board-certified pharmacists and physicians — was already at capacity with the existing volume. The HIPAA compliance posture meant any AI processing of patient clinical data had to happen inside the PBM's HIPAA-compliant environment.
The approach
MindMap deployed Prior Auth Accelerator (Pr) as the adjudication engine, Medical Records Parser (Mp) as the clinical-evidence extraction layer, Rx Compliance Stocker (Rx) as the formulary-and-compliance layer, and Human-in-Loop Manager (Hl) as the clinician-review workflow.
Phase one was the criteria-library digitisation. The PBM's 240-drug clinical-criteria library — historically maintained as a set of PDF documents and Excel-based decision tables — was migrated to Prior Auth Accelerator's structured criteria store. Each drug's medical-necessity criteria, prior-treatment requirements, contraindications and continuation-of-therapy criteria were decomposed into machine-readable assertions that the adjudication engine could evaluate.
Phase two was the clinical-evidence extraction layer. Medical Records Parser was trained on the PBM's historical clinical-documentation corpus (with HIPAA-compliant access controls) to extract the clinical facts the criteria evaluation required: the patient's diagnosis with ICD-10 codes, the relevant clinical-history elements, the prior-therapy history with dates and outcomes, the relevant lab values and imaging findings, and the prescriber's specific medical-necessity rationale.
Phase three was the adjudication-and-review workflow. For each new prior-auth submission, the platform performs the clinical-evidence extraction, evaluates the extracted evidence against the drug's criteria, and produces an adjudication outcome with the specific criteria-by-criteria assessment and the supporting evidence citations. Cases where the criteria are clearly satisfied (the evidence-of-medical-necessity, prior-therapy history and absence-of-contraindications all check out) are auto-approved. Cases where the criteria are clearly not satisfied are auto-denied with the specific deficiency cited. Cases that fall in the middle — most cell-and-gene-therapy authorisations — are routed to the clinician-review queue with the platform's structured assessment as a starting point.
The pre-built building blocks
Rather than commission a ground-up build, the engagement leaned on MindMap's pre-built accelerator library — production-tested components that compress what would otherwise be a six-to-nine-month build into weeks.
Prior Auth Accelerator
Specialty-pharmacy criteria-evaluation and adjudication engine
Medical Records Parser
Targeted clinical-evidence extraction against criteria-required facts
Rx Compliance Stocker
Formulary, dispensing-compliance and DSCSA layer
Human-in-Loop Manager
Clinician-review workflow with one-click adjudication
The architecture
The platform runs in the PBM's HIPAA-eligible AWS environment, with all PHI processing inside a dedicated VPC and the PBM's HIPAA-compliant encryption-at-rest and in-transit posture maintained throughout. No PHI leaves the VPC.
The LLM inference layer uses Llama 3.1 70B served via Amazon Bedrock's private deployment model, with all inference traffic remaining inside the PBM's VPC. The model is fine-tuned on the PBM's clinical-documentation corpus and on the criteria-library content, with the fine-tuning corpus PHI-redacted via Redacto before training.
Prior Auth Accelerator's criteria-store is versioned. As new clinical evidence emerges and the PBM's medical leadership updates a drug's criteria, the new version is deployed alongside the previous version, with the version applied to each authorisation recorded for audit. The criteria-evaluation engine produces a deterministic structured outcome — criterion-by-criterion satisfied/not-satisfied/insufficient-evidence — that the adjudication layer uses.
Medical Records Parser handles the clinical-evidence extraction. Input documents include the prescriber's prior-auth submission form, the patient's relevant chart notes, prior-therapy records from the patient's EHR, lab and imaging reports, and any supporting evidence the prescriber has included. The extraction is structured around the criteria-library's required clinical facts — meaning the extraction is targeted rather than generic.
Human-in-Loop Manager handles the clinician-review workflow. Clinical reviewers see the platform's structured assessment, the supporting evidence citations, the criteria-by-criteria evaluation and the recommended decision; they confirm, modify or override the recommendation with the modification logged for audit and model retraining.
Full HIPAA audit trail on every authorisation — every PHI access, every model inference, every clinical-reviewer action — is persisted with HIPAA-required retention and the PBM's internal HIPAA-compliance team has direct query access.
The numbers behind the story
Average prior-auth turnaround has dropped from 6 calendar days to same-day for the specialty-pharmacy portfolio. The cell-and-gene-therapy authorisations — historically the longest-cycle cases — are now decisioned within 24 hours in the majority of cases, with the small minority of complex multi-disciplinary cases reaching 48-72 hours.
Auto-adjudication rate is 68% — meaning the platform produces the final authorisation decision end-to-end without clinician review on 68% of submissions, primarily on the cases where the clinical evidence cleanly meets or cleanly does not meet the criteria. The remaining 32% receive clinician review with the platform's structured assessment as a starting point.
Clinician productivity on the cases that do require review has improved by approximately 2.5x — the structured assessment, the evidence citations and the recommended decision eliminate most of the chart-review and criteria-lookup work the clinician previously performed.
Patient-experience outcomes have followed. The PBM's payer clients have reported materially improved patient-experience scores on the specialty-pharmacy journey, with the turnaround improvement being the most-cited driver. Several payer clients have specifically referenced the PBM's specialty-pharmacy prior-auth as a contract-renewal positive.
An unexpected outcome: the criteria-library digitisation has accelerated the PBM's medical-leadership process for incorporating new clinical evidence into the criteria. The previous PDF-and-Excel-based criteria management had created a slow review-and-update cycle; the structured criteria store has reduced the time-to-update for clinically significant evidence changes from several months to several weeks.
“Cell-and-gene-therapy prior-auth in six days was hurting patients and threatening payer contracts. MindMap delivered same-day in twenty-six weeks, with our clinical reviewers doing genuinely judgement-intensive work on the cases that need it. Several of our payer clients have specifically cited the turnaround improvement in contract-renewal conversations.”— Chief Pharmacy Officer· US Specialty PBM
Why MindMap was chosen
The PBM had evaluated two healthcare-AI vendors and one major consulting firm. The healthcare-AI vendors had general prior-auth capabilities but had not previously tackled the specialty-pharmacy cell-and-gene-therapy depth that the PBM's portfolio required. The consulting firm proposed a custom-build approach with a multi-year timeline.
MindMap's Prior Auth Accelerator was already deployed at another US payer for general prior-auth work, with the depth of clinical-criteria handling that the PBM required. We could extend the accelerator to the specialty-pharmacy criteria depth and demonstrate it on the PBM's own clinical-criteria library during the bid.
The willingness to deploy entirely inside the PBM's HIPAA-eligible AWS environment — including the LLM fine-tuning, which involved PHI-redacted clinical data — was the regulatory differentiator. Our embedded clinical-informatics expertise on the delivery team (three clinically-trained delivery team members, including a former specialty-pharmacy clinical-reviewer) was the third factor.
Related deployments
Sovereign WhatsApp Banking
ChatNext-powered WhatsApp bot deployed inside the bank's air-gapped data centre, handling balance, transfers, statements and loan applications in English and Swahili.
Cheque OCR at 99.1% Accuracy
DocuMage replaced a legacy template OCR for cheque clearing, processing 10,000 cheques per day at 99.1% field accuracy with 94% straight-through processing.
UK Challenger Bank KYC
OnboardX rebuilt the KYC pipeline with liveness, sanctions and PEP enrichment in a single STP flow, collapsing onboarding from 5 days to 4 hours.
Want an outcome like this?
Start with a 2-week AI Readiness Sprint. We deliver a prioritised use-case backlog and business case grounded in what's actually buildable with our accelerator library.