Schedule a Demo

Specialty Pharmacy Intake: Is AI Adding Real Value or Just Pretending To?

AI Value
Prasad Karanam
Prasad Karanam

VP, Product

If you layer AI on top of a broken intake process, you will still get broken results
That observation, shared by a senior operations leader at a specialty pharmacy describing her daily intake workload, captures a growing disconnect in the market. AI models may look impressive on paper, but too often they fail to improve real-world specialty pharmacy operations.
AI models may look impressive on paper

Nowhere is this disconnect more evident than in specialty pharmacy intake. Rising therapy complexity, increased payer scrutiny, persistent staffing constraints, and fragmented systems have turned what should be a foundational workflow into a chronic bottleneck. Routine intake delays therapy starts, increases downstream rework, and quietly erodes operating margins. Leaders consistently describe the same pain points: manual handoffs, incomplete referrals, limited visibility into prior authorization risk, and teams overwhelmed by repetitive, non-clinical work.

Intake is not just the first step in the patient journey; it is one of the largest sources of operational friction, financial exposure, and staff burnout in specialty pharmacy today.
AI Presence ≠ Intake Operational Value
Against this backdrop, many software vendors now position “AI-powered intake” as the solution. But the question specialty pharmacy leaders are asking is no longer whether AI exists in a product. It is whether that AI is properly evaluated, embedded into real workflows, and measurably improves outcomes. If it does not make intake faster, more accurate, and more predictable, then it is selling the appearance of innovation without delivering its value.

When Intake Breaks, Patients Feel It First

Specialty drug patients represent a relatively small portion of the U.S. population, yet they account for a disproportionate share of total drug spending. National projections estimate total U.S. prescription drug spend at approximately $800+ billion annually6, with specialty medications accounting for nearly half of that spend, driven by biologics, oncology, rare disease therapies, and the rapid expansion of cell and gene therapies.

At the same time, access breakdowns are worsening1,2,3,5:
• More than one-quarter of branded prescriptions go unfilled
• Over half of newly launched specialty prescriptions never reach patients
• Fewer than one-third of patients remain on therapy after the first year
Over 50% of novel prescriptions go unfilled
These are not abstract statistics. In specialty pharmacy, intake is where these failures often begin.
Delayed Care, Disengaged Patient
Intake is where clinical documentation, benefit verification, payer requirements, and authorization logic collide. When intake is slow, incomplete, or inaccurate, the consequences ripple across both patient care and financial performance. Therapy starts are delayed, prior authorizations are denied or resubmitted, patients disengage, and pharmacies absorb unreimbursed labor and revenue risk.

As specialty spend continues to grow, pressure on intake, prior authorization, and revenue cycle workflows intensify. This reality has been repeatedly highlighted in NHIA educational sessions and Asembia roundtables, where operators emphasize that intake breakdowns are rarely caused by a lack of data but by a lack of coordination, context, and workflow intelligence.

“Imagine being told by your doctor that the condition you’ve been diagnosed with requires a specialty medication, but you may not be able to start treatment for weeks due to outdated processes and paperwork required for prior authorization,” said Cecelia Byers, Pharm.D., Clinical Product Advisor for Specialty at Surescripts9.

Delayed Time to Therapy Has a Quantifiable Impact

Specialty medications, many costing $5,000–$10,000 per month or more, are designed to treat serious, chronic, or life-threatening conditions. Yet surveys of specialty prescribers and pharmacists consistently show that intake and prior authorization delays push time to therapy well beyond acceptable windows.

While more than 80% of clinicians believe patients should start therapy within two weeks, only 20–30% report that this actually happens. Most patients wait three to four weeks or longer to begin treatment. The most common causes are incomplete referrals, missing documentation, and improperly prepared authorization submissions — all intake-related failures.
Only 20-30% start therapy within 2 weeks
$1,750 in deferred or at-risk revenue per patient per week delayed
Even modest delays have meaningful financial consequences. A one-week delay for a specialty medication with an average monthly cost of $7,0004 represents approximately $1,750 in deferred or at-risk revenue per patient per week delayed. Across hundreds or thousands of referrals, intake inefficiency becomes a material financial exposure — not a marginal operational issue.

Yet intake remains largely manual, fragmented, and error prone. Referral documents arrive via fax, portal uploads, or EHR messages. Staff must reconcile clinical notes, benefits data, payer rules, and authorization requirements across disconnected systems. It is within this complexity that many vendors claim AI can help, but the way AI is evaluated often misses the point.

Intake Is a Workflow Problem, Not Just a Model Problem

The Metric Trap

Many AI solutions approach intake as a narrow modeling challenge: extract text, classify fields, and report high performance metrics. Vendors frequently highlight AUC scores, extraction accuracy, or token-level precision. AI models are trained against historical annotations or abstracted datasets that do not reflect ground truth aligned with experts’ gold standard7. These metrics are useful for training models that score well against outdated ground truth, but they are poor indicators of downstream operational impact.
Vendors frequently highlight AUC scores, extraction accuracy, or token-level precision
AI models tend to fall into the same Dunning–Kruger failure mode
Common categories of intake AI in the market include OCR-plus-NER pipelines, generic LLM document extraction workflows, and rules-augmented automation layers. Across the market, these AI models tend to fall into the same Dunning–Kruger failure mode: strong benchmark accuracy creates confidence, while misaligned loss functions, payer rule drift, and unmodeled workflow dependencies limit real-world impact on authorization success and time-to-therapy.
Intake workflows do not treat all errors equally, yet most AI metrics do.
Missing a diagnosis code, misclassifying therapy urgency, or failing to recognize payer-specific documentation requirements can delay therapy by weeks. By contrast, a formatting error or minor demographic inconsistency may have little downstream impact. Model-level metrics often weight these errors the same, producing “high-performing” models that still generate bad outcomes.
“High-performing” models that still generate bad outcomes

Consider a common intake scenario:

A referral arrives for a CAR-T patient. The high-cost oncology therapy is processed by an AI intake tool. The model accurately extracts patient demographics and medication name, achieving excellent extraction scores (99% accuracy). However, it fails to recognize a required staging note embedded in the clinical documentation — a payer-specific requirement. The prior authorization is submitted incomplete, denied days later, reworked manually, and resubmitted. Therapy is delayed by nearly 21 days.

Model Success, Operational Failure
From a model perspective, performance looks strong. From an operational perspective, intake failed. This gap explains why many AI intake tools speed up data entry but fail to reduce rework, denials, or time to therapy. Optimizing model performance without optimizing workflow outcomes simply shifts where the work happens; it does not eliminate it.

What Real AI Value in Intake Should Look Like

AI adds value in specialty pharmacy intake only when it is embedded into workflows and trained to reflect operational reality. True value looks like this:

Contextual understanding — not just capture

AI must go beyond text extraction to interpret clinical meaning, payer logic, and coverage nuance. It should identify missing or mismatched information before work enters prior authorization queues, not after denials occur.

Workflow intelligence and prioritization

Effective intake AI should:
• Instantly flag incomplete referrals
• Prioritize cases with the highest risk of delay or denial
• Apply payer-specific rules and definitions dynamically
This shifts intake from reactive cleanup to proactive risk management.

Deep integration with core systems

Workflow-embedded AI connects directly with pharmacy management systems, prior authorization platforms, benefits verification tools, and payer portals. Standalone AI modules that require manual uploads or handoffs often increase fragmentation rather than reduce it.

Workflow-Centric AI Intelligence
Evaluating AI for intake therefore requires shifting from model-centric metrics to workflow-centric outcomes: reductions in rework, denials, time to therapy, and staff burden.

Why This Matters Now

Specialty pharmacy spend continues to rise. Patient populations are expanding. Competition among specialty pharmacies has intensified. In this environment, intake efficiency is no longer a back-office concern, but it is a determinant of patient access, payer trust, and financial sustainability.

The American Society of Health-System Pharmacists has emphasized that while AI can reduce manual burden and improve clinician satisfaction, pharmacy teams must lead evaluation and implementation using the right metrics8.

High model performance alone does not guarantee operational value.

Specialty pharmacies that conflate the two, risk investing in technology that looks advanced but leaves intake outcomes unchanged.

The Bottom Line

AI that performs well in isolation is not enough. In specialty pharmacy intake, value is proven only when AI improves workflows, reduces errors, shortens time to therapy, and lowers downstream cost. In practical terms, value is created when the gap between a written prescription and a filled prescription meaningfully narrows.

That requires moving beyond model metrics and designing intelligence where the work actually happens. As a founding member of NASP and an active participant in NHIA and Asembia, Keycentrix purpose-builds intake technology informed by real operational challenges.

Schedule a demo today to explore how Newleaf delivers measurable operational impact.
Streamline Your Operations. Improve Patient Care.

References

  1. American Medical Association, Prior Authorization Survey, 2024
  2. American Medical Association, Prior Authorization Delays Care and Increases Healthcare costs, 2024
  3. Specialty Pharmacy Continuum, Half of Novel Specialty Prescriptions Go Unfilled, IQVIA Data Show, 2026
  4. AARP, Trends in Retail Prices of Prescription Drugs, 2024
  5. Journal of Managed Care and Specialty Pharmacy, The Association Between Cost Sharing, Prior Authorization, and Specialty Drug Utilization, 2023
  6. American Society of Health-System Pharmacists, National Trends in Prescription Drug Expenditures and Projections for 2025, 2025
  7. MIS Quarterly. Is AI Ground Truth Really True? The Dangers of Training and Evaluating AI Tools Based on Experts’ Know-what, 2021
  8. ASHP. AI Helps Pharmacists Streamline Routine Tasks, 2025
  9. Surescripts, Inaccurate and Incomplete Data Delays Specialty Treatment for Patients and is Top Stressor for Health Care Providers, 2022

Related Articles

Take a Tour

Contact us at 1-800-444-8486 or drop us a line to request a tour of our Newleaf pharmacy management software.

Schedule a Demo