NEWMindMap Digital has acquired Bluetide.co— deepening our data & agentic-AI stack.Read more →
Home · Customer Stories · US Health Insurer
Healthcare · North America

Prior Authorisation Acceleration at a Top-25 US Health Insurer — From 3-Day Wait to 4-Hour Turnaround

Prior Auth Accelerator + DocGenie collapsing the payer submission cycle by 18x, with 70% of authorisations approved end-to-end with no human review.

4 hrs
From 3-day average
22w
Delivery duration
Private Cloud
Deployment
4
Accelerators used
Private CloudUS Health Insurer — 4 hrs From 3-day average
4 hrs
Avg turnaround (was 3 days)
70%
Auto-approved end-to-end
180K
Auth requests / month
31%
Provider abrasion score reduction
In this storyHealthcarePrior AuthHIPAAOn-Prem LLMAgentic AI
01
The challenge

The challenge

The insurer — a regional health plan serving more than nine million members across six US states — was facing a prior-authorisation crisis. Prior auth is the process by which a provider asks the payer to confirm coverage for a specific procedure, drug, or service before it is rendered. It exists to protect against unnecessary care; in practice, it is one of the most-hated processes in US healthcare, blamed for delays, denials and clinician burnout.

The insurer's average prior-auth turnaround was 3.2 calendar days. State regulators in two of the six markets had imposed twenty-four-hour decision requirements for non-urgent requests and were threatening fines. Provider abrasion — the technical term for how badly providers hate dealing with a particular payer — was at the worst level the insurer's measurement system had ever recorded. The Medical Group Management Association's annual survey had specifically called out the insurer's prior-auth process as a leading complaint.

The plan was receiving roughly 180,000 prior-auth requests per month, of which 84% arrived as faxes or PDF uploads to the provider portal. The clinical-review team — 240 nurses and physicians — was spending the majority of their time on data entry, document classification and policy lookup, rather than on the clinical judgement work they were trained for. Forty per cent of requests were administratively denied for missing documentation, only to be resubmitted within seventy-two hours, creating a re-work loop that consumed roughly 22% of the team's capacity.

02
The approach

The approach

We led with two accelerators — Prior Auth Accelerator (Pr), our healthcare-specific submission and adjudication pipeline, and DocGenie, our medical-document processing engine that we positioned as the ingestion layer for the entire prior-auth flow.

The first phase was triage. We worked with the insurer's clinical leadership to classify all 180,000 monthly requests into four bands: Band A (clear policy match, no clinical judgement required, e.g. routine imaging for in-network providers), Band B (clear policy match but requiring clinical-criteria confirmation, e.g. specialty drugs with step-therapy requirements), Band C (clinical judgement required, e.g. atypical surgical requests), and Band D (urgent, member-in-care). Band A was 38% of volume, Band B was 36%, Band C was 24%, and Band D was 2%.

We then built the automation in priority order. Band A — the largest volume, lowest clinical risk, and the most policy-clear — was tackled first. We mapped the insurer's clinical-policy library (roughly 1,400 active policies, each ten to forty pages long) into a structured policy knowledge base that the Prior Auth Accelerator could query. Each policy was decomposed into a set of machine-readable criteria — required documentation, clinical thresholds, member-eligibility conditions and excluded conditions. The Prior Auth Accelerator then matched each inbound request against the relevant policy, extracted the necessary fields from the submitted documents using DocGenie, and either auto-approved, auto-denied (with clear rationale), or routed to clinical review.

Band B was tackled in phase two — the clinical-criteria layer required a more sophisticated reasoning step, with the LLM generating a draft recommendation and a senior clinician providing one-click approval or rejection. Band C requests remained human-led, but we deployed a clinical drafting assistant that pre-populated the case summary, surfaced the relevant policy excerpts, and flagged any missing documentation up-front — reducing time-per-case by roughly 60% even where no automation was possible.

Accelerators in this engagement

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.

Pr

Prior Auth Accelerator

Policy adjudication, automated submission and decision routing

Mp

Medical Records Parser

Clinical field extraction from EHR exports, notes and labs

Cd

Coding Assistant

ICD-10 / CPT validation and gap detection

Hl

Human-in-Loop Manager

Clinician approval workflow with one-click adjudication

03
The architecture

The architecture

The system runs in the insurer's existing AWS HIPAA-eligible environment, with all PHI processing happening inside a dedicated VPC. The insurer's compliance team required that no PHI be processed by any third-party LLM API; we satisfied this by deploying a private inference layer using Llama 3 70B-Instruct served via Amazon Bedrock's private model deployment, with all inference traffic remaining inside the insurer's VPC.

Document ingestion happens through multiple channels: the existing provider portal (where providers upload PDFs directly), an inbound fax gateway (which converts faxes into PDFs in near-real-time), an EHR integration layer (which receives prior-auth requests from major EHR systems via FHIR PA endpoints), and a payer-portal API for providers using third-party prior-auth submission tools.

DocGenie ingests every document, performs OCR on faxed and scanned material, classifies the document type (clinical notes, lab results, imaging report, member ID card, etc.), and extracts structured clinical fields — diagnosis codes, procedure codes, clinical history, lab values, prior-treatment history. The extracted fields feed into the Prior Auth Accelerator's policy engine, which combines deterministic rules (eligibility, network status, prior auth requirement existence) with LLM-grounded clinical-criteria evaluation (does the documented clinical picture meet the medical-necessity criteria in the applicable policy?).

Every automated decision is captured in a full audit log with the policy citations, document excerpts and reasoning steps that led to the decision. This audit log is the evidence the insurer presents during state regulatory audits and is also the data feed for the insurer's quality-management programme. All adverse decisions (denials and approvals at less than the requested level) are routed for human review before going back to the provider — automation generates the draft, but a human signs the response.

The outcomes

The numbers behind the story

4 hrs
Avg turnaround (was 3 days)
70%
Auto-approved end-to-end
180K
Auth requests / month
31%
Provider abrasion score reduction

Average prior-auth turnaround has dropped from 3.2 calendar days to 4.1 hours. 70% of requests are now resolved end-to-end without human review — these are predominantly Band A requests with clean documentation. A further 22% receive automated drafts that a clinician reviews and approves or modifies in under three minutes. Only 8% require full clinical workup, and those receive the clinician's full attention because everything else has been removed from their queue.

The clinical-review team has not been reduced. Instead, the reclaimed capacity has been redirected to two priorities: complex case management (Band C cases, which now get significantly more clinician time per case) and member-facing care coordination (clinicians proactively reaching out to members whose prior auths suggest complex care journeys ahead).

Provider abrasion has dropped 31% on the insurer's measurement scale. State regulator turnaround requirements are being met with significant headroom. The administrative-denial rate has fallen from 40% to 6% — because DocGenie identifies missing documentation up-front and the provider portal now prompts for it before submission, the re-work loop has largely been eliminated.

Member-experience scores have improved meaningfully. The insurer's care-experience team reports that 'time-to-care' — the time from a provider requesting prior auth to the member actually receiving the approved care — has dropped by an average of 4.7 days, primarily because the auth itself completes in hours rather than days.

We tried to do this twice before — once with a major consulting firm, once with an in-house build. Neither got past the policy-knowledge-base problem. MindMap walked in with the accelerator already built, the LLM stack already running inside our VPC, and an informaticist on the delivery team who understood our policy library better than some of our own analysts. Four months later, prior auth is no longer the thing providers and members complain about.
Chief Medical Officer· US Regional Health Insurer
04
Why MindMap was chosen

Why MindMap was chosen

The insurer ran a formal vendor selection involving two well-known healthcare AI vendors and one of the major consulting firms. MindMap was chosen for three reasons.

First, the Prior Auth Accelerator is purpose-built for this problem. The competing vendors were proposing custom builds; the consulting firm was proposing a strategy engagement followed by a custom build. We could demonstrate, on the insurer's own redacted documents, that the accelerator could parse them, match them against the insurer's policy library, and generate accurate adjudications. The proof-of-concept took three weeks; the competing vendor's POC was scheduled for sixteen weeks.

Second, our willingness to deploy the LLM inference layer inside the insurer's own AWS VPC — and our experience doing so at other US healthcare clients — solved their HIPAA concerns up front. The other vendors required at least some PHI to be processed in shared cloud environments.

Third, our delivery model included embedded clinical informaticists, not just engineers. Three of our delivery team members had clinical backgrounds and could speak the language of medical-necessity criteria and policy interpretation with the insurer's clinical leadership. This dramatically accelerated the policy-knowledge-base build, which is the work that breaks most prior-auth automation programmes.

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.

Book a walkthrough →Explore Healthcare
Talk to the product team