This report is confidential

Enter the access code to view.

Invalid code. Please try again.

Intelligence factory opportunity map

Regeneron Pharmaceuticals · Outside-in analysis · August 2026

Confidential — prepared for internal review only

6
Departments mapped
27
Workflows inventoried
15
Stack candidates
$17–64M
Estimated annual value
  1. 78% of mapped decision points are automatable. Across 42 decision points in the four flagship workflows, 19 (45%) can run as deterministic edge functions at near-zero cost and 14 (33%) as generative API calls — leaving only 9 (22%) as irreducible human judgment nodes.
  2. Patient Services is the highest-value target. The PA/benefit verification workflow alone represents $13–56M in combined savings and revenue protection from reduced prescription abandonment across Dupixent MyWay, EYLEA4U, LIBTAYO Surround, and MyPRALUENT.
  3. Regeneron's AI is concentrated in R&D — commercial is the white space. The published AI policy confirms ML/NLP/LLM use in Global Development and the Genetics Center, but the six priority commercial/infrastructure departments have minimal disclosed AI deployment.
  4. Five permanent kill gates constrain the architecture. FDA promotional compliance (MLR), AKS/FCA (copay assistance — elevated by active DOJ litigation), SOX (financial attestation), hiring decisions (Regeneron's own AI policy), and government pricing certification cannot migrate to automation regardless of accuracy.
  5. The gap your current vendor misses is the task→decision→process hierarchy. Mapping roles and tasks is necessary but insufficient. The value unlock comes from connecting each task to the decision it serves, assigning that decision through the Jagged Edge Trilemma, and chaining decisions into agent assembly lines with guardian agents at every handoff.

What your current approach misses

Your vendor ingests role descriptions and proposes tasks with AI potential. This analysis goes three layers deeper: it maps every decision point within each workflow (not just the tasks), assigns each to the optimal intelligence layer through the Jagged Edge Trilemma (deterministic edge function vs generative API call vs human judgement), and designs the agent assembly line — the complete pipeline of specialized agents with guardian agents validating quality at every handoff. The result: you see not just what can be automated, but exactly how the automation chains together into end-to-end intelligent workflows.

Regeneron at a glance — Q2 2026

Financial and strategic signals that shape the automation opportunity

$14.34B
2025 revenue
$2.70B
2025 GAAP SG&A
$5.85B
2025 R&D spend
~15,410
Employees
$15.1B
Net cash

Friction signals from leadership

Q2 2026 earnings call surfaced: a temporary manufacturing disruption at Limerick that dented gross margin (78% vs 83%); FDA leadership changes creating regulatory uncertainty; the April 2026 MFN pricing agreement adding government-pricing complexity; and a DOJ False Claims Act case over copay-assistance donations. Each represents an information-processing burden that automation can address.

AI initiatives already underway

Regeneron's published Position Statement on Responsible Use of AI (July 2025) confirms company-wide use of ML, NLP, GPTs, LLMs, and RPA, governed by a cross-functional AI Advisory Committee. Disclosed deployments concentrate in Global Development and the Regeneron Genetics Center — not in the six priority commercial functions. Two hard internal kill gates already codified: AI is barred from final decisions on personnel hiring and financial decisions.

Peer benchmarks

Sanofi plai: 15,000+ daily users, 300+ AI models, predicted 80% of low inventory positions, ~$300M supply-chain savings, $800M reallocated per Fortune report. Eli Lilly/Yseop: eliminated 10,000 hours of clinical narrative writing, 53% third-party cost reduction. Veeva Falcon MLR: potential to "eliminate 70% or more of manual MLR labor within five years." Develop Health Calibrate: "80–85% reduction in PA volume." All figures are vendor/company claims and should be treated as directional.

Six priority departments — automation potential by function

Click any department to see its workflows. Star ratings indicate analysis depth available.

Patient services
~60–150 FTEs · 4 workflows
Info density
Automation
$15–60M
estimated annual value
PA & benefit verification★★★★
Copay & financial navigation★★★
Adherence monitoring★★
Hub operations management★★
Market access
~40–80 FTEs · 5 workflows
Info density
Automation
$12–27M
estimated annual value
Government pricing★★★
Formulary tracking★★★
Field reimbursement★★★
HEOR dossier development★★
Contract analytics★★
Marketing
~60–100 FTEs · 4 workflows
Info density
Automation
$4–10M
estimated annual value
MLR content review★★★★
Omnichannel HCP campaigns★★★
Congress & medical ed★★
Brand planning
Sales
~120–250 FTEs · 3 workflows
Info density
Automation
$5–10M
estimated annual value
Pre-call planning★★★
Sample & speaker programs★★
SFE & incentive comp★★
Finance
~80–150 FTEs · 7 workflows
Info density
Automation
$7–16M
estimated annual value
Accounts payable★★★★
Gross-to-net calculations★★★
Financial close★★
FP&A reporting★★
IR earnings prep★★
Internal audit / SOX★★
Procurement★★
HR
~50–80 FTEs · 4 workflows
Info density
Automation
$3–8M
estimated annual value
Talent acquisition★★★★
Comp & benefits admin★★
Learning & development
Workforce planning
Edge function API call Human operator ★★★★ = Full decision-point architecture available

Four flagship workflows mapped at full depth

Each workflow mapped with the complete methodology: every decision point documented, assigned through the Jagged Edge Trilemma, and chained into an agent assembly line with guardian agents and trust escalation.

MLR content review
PA & benefit verification
Accounts payable
Talent acquisition

MLR content review — 11 decision points, 36% edge / 36% API / 28% human

Edge 36% API 36% Human 28%

Assembly line: Brief validator → Reference checker → Claims-to-label mapper → Fair balance analyzer → Pre-review synthesizer → Human: Legal/Reg/Medical review → Revision checker → Approval recorder → Expiration monitor

The human operator today

On any given Tuesday morning, a Medical Reviewer at Regeneron opens Veeva Vault PromoMats and finds fourteen items in queue. Three are Dupixent HCP detail aids for the new COPD indication. Two are EYLEA HD patient brochures for the pre-filled syringe launch. One is a Libtayo congress poster submitted Friday at 4:47 PM with a note reading "urgent — booth printing deadline Thursday."

The reviewer doesn't read these documents linearly. She scans for specific failure modes: every efficacy claim must trace to the approved prescribing information. Every comparative statement must be supported by head-to-head data. Fair balance — risk information presented with comparable prominence to benefits — must be adequate. No off-label promotion. A single piece may cycle three to five times before approval, and the ASCO poster will consume four hours across two reviewers while the brand team works past midnight.

The bottleneck isn't competence — it's the volume-to-reviewer ratio and the cognitive load of context-switching across 8+ brands, dozens of indications, and hundreds of approved claims that may have been updated by a supplemental approval since the last review cycle.

Decision-point map

#Decision pointAssignmentInputsJudgmentOutputConsequence
1Content brief intake & completeness EdgeSubmitted asset, brief templateChecklist match — all fields present?Accept or return to brandLow — caught downstream
2Reference verification APIClaims text, cited sources, PIDoes each claim resolve to an approved source?Citation map with flagsHigh — FDA Warning Letter
3Claims-to-label mapping APIClaims, current PI, prior approvalsIs every claim within approved labeling?Mapped claims + off-label flagsCritical — off-label promotion, FCA
4Fair balance assessment Edge + APIAsset layout, risk/benefit textRisk prominence ≥ benefit prominence?Balance score + deficiency flagsHigh — OPDP letter
5Competitive claims review HumanComparative statements, competitor dataLegal risk of each comparative claimApprove / revise / escalateHigh — Lanham Act complaint
6Legal review HumanFull asset, IP terms, co-promo agreementsIP exposure, contractual complianceLegal sign-off or revision listMedium-high — IP, co-promo terms
7Regulatory review Human + APIFull asset, FDA guidance, OPDP historyRegulatory risk across all claimsRegulatory sign-off or holdCritical — regulatory non-compliance
8Medical review (final) Kill gatePre-screened asset + all prior reviewsClinical accuracy, patient safetyFinal approval or rejectionCritical — patient harm, FDA enforcement
9Revision compliance check APIRevised asset, revision commentsWere all requested changes implemented?Compliance confirmation or re-routeMedium — additional cycle delay
10Version control & approval recording EdgeApproved asset, metadataCorrect version, correct status in VaultRecorded approval + audit trailMedium — compliance finding
11Expiration & withdrawal monitoring Edge + APIApproved assets, PI update feed, calendarHas PI changed since approval? Is asset expiring?Withdrawal alerts + re-review triggersHigh — outdated content in circulation

Guardian agents

Claims guardian

Between Reference checker → Claims-to-label mapper

Validates every citation resolves to an approved PI section. Rejects any claim where the source is a poster, abstract, or non-peer-reviewed publication unless explicitly permitted by the approved claims matrix.

Fair balance guardian

Between Fair balance analyzer → Pre-review synthesizer

Validates risk/benefit prominence ratio meets FDA standards (OPDP guidance). Measures text area, font size, and placement of risk information relative to efficacy claims. Flags any asset where risk prominence falls below 80% of benefit prominence.

Approval guardian

Between Revision checker → Approval recorder

Validates all revision comments from Legal, Regulatory, and Medical reviewers are addressed before final recording. Cross-references comment log against revised asset. Blocks recording if any comment is unresolved.

Trust escalation ladder

1
Full human review
All items reviewed by Legal, Regulatory, and Medical reviewers. AI not involved.
Current baseline
2
AI pre-screens, humans review flagged issues + sample
AI runs reference verification, claims mapping, and fair balance checks. Produces a pre-review package highlighting issues. Humans review AI-flagged items plus a random 20% sample of cleared items.
Target: 0–3 months
3
AI clears routine items, humans review new claims + high-risk
Routine items (renewals, minor copy updates, previously approved claims) auto-clear. Humans focus on new claims, new indications, competitive materials, and congress submissions.
Target: 3–9 months · Gate: ≥95% accuracy on pre-review flags
4
AI handles full pre-review, humans do final sign-off only
AI produces complete review package with recommended disposition. Human reviewers approve or override. Cycle time drops from 2–3 weeks to 3–5 days.
Target: 9–18 months · Gate: ≥98% concordance with human reviewers over 6 months
5
Full autonomy — N/A
Medical review final sign-off is a permanent kill gate per FDA 21 CFR 202.1. Rung 5 is not a target state.

Economics

Current estimated cost
$3–5M/yr
Projected savings at Rung 3
$1.5–2.5M/yr
Plus: time-to-market
2–3 wk → 3–5 days

Permanent kill gates

Medical review final sign-off — FDA 21 CFR 202.1 requires human physician sign-off on all promotional materials. No AI accuracy level removes this requirement.

Competitive claims review — Lanham Act litigation risk on comparative claims requires human legal judgment on case-by-case basis.

Rate this deep-dive analysis
✓ Rating saved

PA & benefit verification — 13 decision points, 46% edge / 31% API / 23% human

Edge 46% API 31% Human 23%

Assembly line: Case intake → Insurance verifier → Benefit investigator → PA pathway router → Clinical evidence assembler → PA form completer → Submitter → Status tracker → [Denied: Analyzer → Human: appeal decision] → Copay/PAP router → Human: eligibility check → Enrollment → Adherence monitor

The human operator today

A Patient Access Coordinator at Dupixent MyWay starts each morning with forty to sixty new cases. Each represents a patient whose physician has prescribed a ~$36,000/year biologic and who needs help navigating insurance coverage. The coordinator logs into payer portals, determines PA requirements (which vary by payer, plan, and indication across atopic dermatitis, asthma, COPD, CSU, and eosinophilic esophagitis), assembles clinical evidence packages, completes payer-specific PA forms, submits, and tracks status — calling payer phone lines that average 12–25 minutes of hold time per call.

If denied, the coordinator initiates appeals: peer-to-peer scheduling, additional documentation, formal appeal letters. Each adds 2–4 weeks. Industry data shows 30–40% of specialty prescriptions are abandoned during PA. Each abandoned Dupixent prescription is both a patient who doesn't receive therapy and approximately $36,000 in lost annual revenue.

Multiply across four hub programs — Dupixent MyWay (1.5M+ patients), EYLEA4U, LIBTAYO Surround, MyPRALUENT — and the operation processes thousands of cases monthly, each with its own payer rules, clinical requirements, and urgency level.

Decision-point map

#Decision pointAssignmentInputsJudgmentOutputConsequence
1Case intake & triage EdgePrescription, patient demographicsCompleteness check, priority scoringTriaged case recordLow — incomplete data caught next step
2Insurance verification EdgeMember ID, payer infoEligibility lookup against payer DBCoverage status + plan detailsMedium — wrong payer = wasted submission
3Benefit investigation APIPlan details, drug formulary dataPA required? Step therapy? Tier?Coverage pathway mapHigh — missed PA = claim denial
4PA pathway routing EdgePayer rules, indication, prior therapyWhich payer-specific PA form applies?Routed to correct pathwayMedium — wrong form = rejection
5Clinical evidence assembly APIPatient chart, payer criteria, guidelinesWhich evidence satisfies this payer?Evidence package with citationsHigh — weak evidence = denial
6PA form completion APIEvidence package, payer form templateMap evidence to form fields accuratelyCompleted PA formMedium — errors delay approval
7Electronic submission EdgeCompleted form, portal credentialsCorrect portal, correct formatSubmission confirmation + tracking IDLow — resubmit if rejected
8Status tracking EdgeTracking IDs, response timelinesOverdue? Needs follow-up?Status updates + escalation triggersMedium — delayed = patient waiting
9Denial analysis APIDenial letter, payer criteria, historyWhy denied? What would overturn?Root cause + appeal optionsHigh — missed appeal = abandoned Rx
10Appeal strategy decision Kill gateDenial analysis, clinical context, urgencyPeer-to-peer? Written appeal? Escalate?Appeal pathway selectionCritical — wrong strategy = permanent denial
11Copay/PAP eligibility Kill gatePatient financials, program criteria, AKS rulesQualifies without inducement risk?Eligible / ineligible + assignmentCritical — AKS/FCA violation (active DOJ)
12Enrollment processing EdgeEligibility determination, program rulesCorrect program, correct copay cardEnrolled patient recordLow — administrative correction
13Adherence monitoring APIRefill data, appointment historyRisk of discontinuation? Intervene?Adherence score + outreach triggersHigh — non-adherence = lost patient

Guardian agents

Clinical evidence guardian

Between Evidence assembler → Form completer

Validates evidence package contains required diagnosis codes (ICD-10), lab values, and prior therapy documentation for the specific payer's criteria. Rejects incomplete packages before form completion.

Submission guardian

Between Form completer → Submitter

Validates all required fields populated, correct payer portal selected, patient consent on file, and form version matches current payer requirements.

Eligibility guardian

Between Copay/PAP router → Enrollment

Validates AKS/FCA compliance on every copay assistance determination. Confirms no prohibited inducement patterns, checks against federal healthcare program enrollment.

Trust escalation ladder

1
Full human processing
All cases processed manually by Patient Access Coordinators across all four hub programs.
Current baseline
2
AI pre-populates, human reviews before submission
AI auto-fills PA forms from patient data and payer criteria. Coordinator reviews pre-populated form, corrects errors, and submits. Per-case time drops from 45 min to 15 min.
Target: 0–3 months · Gate: ≥90% form accuracy
3
AI auto-submits routine PAs, human handles exceptions
Standard PAs for known payers with standard indications auto-submit. Human coordinators focus on appeals, new payers, and copay/PAP determinations.
Target: 3–9 months · Gate: ≥75% auto-submission accuracy
4
AI handles full cycle for top-10 payers
End-to-end processing for the 10 highest-volume payers (~60% of cases). Human coordinators handle appeals, new payer onboarding, and all eligibility determinations.
Target: 9–18 months · Gate: ≥95% auto-approval rate
5
Full autonomy — N/A
Copay/PAP eligibility permanently requires human judgment (AKS/FCA — active DOJ). Appeal strategy is irreducibly clinical.

Economics

Current estimated cost
$8–16M/yr
Direct savings at Rung 3
$3–6M/yr
Revenue protection
$10–50M+

Permanent kill gates

Copay/PAP eligibility determination — AKS/FCA prohibits AI from determining financial assistance eligibility. Active DOJ False Claims Act investigation elevates this to litigation defense.

Appeal strategy selection — Clinical judgment required for peer-to-peer vs written appeal vs external review for each patient's specific situation.

HIPAA consent management — Patient authorization for data sharing across hub programs requires informed consent.

Rate this deep-dive analysis
✓ Rating saved

Accounts payable — 9 decision points, 67% edge / 22% API / 11% human

Edge 67% API 22% Human 11%

Assembly line: Data capture → PO matcher → GRN matcher → HCP-spend screener → Duplicate detector → Approval router → Human: above threshold → Payment scheduler. Exception: Discrepancy resolver → Human adjustment

The human operator today

An AP Specialist opens Oracle Fusion each morning to a queue of invoices from hundreds of vendors: CROs billing millions for clinical trials, contract manufacturers for API production, marketing agencies for congress sponsorships, and freight companies for cold-chain biologics shipment. Each follows the same flow: capture, match against PO, match against goods receipt, resolve discrepancies, route for approval, schedule payment.

The friction is in exceptions — approximately 20–30% require handling. A CRO's invoice references a change order; a marketing agency includes out-of-scope travel. Layered on top: every payment that could constitute a "transfer of value" to an HCP must be tracked for Sunshine Act reporting. Given the DOJ/FCA matter, accuracy on HCP-spend tracking is a litigation defense requirement.

Decision-point map

#Decision pointAssignmentInputsJudgmentOutputConsequence
1Invoice data capture EdgeInvoice image/PDFOCR extraction + field mappingStructured invoice recordLow — manual correction available
2PO matching EdgeInvoice line items, PO databaseLine-by-line match within toleranceMatched / unmatched flagsMedium — delays payment
3GRN matching EdgePO match, goods receipt recordsThree-way match: PO + GRN + invoiceClean match or discrepancy flagMedium — overpayment risk
4HCP-spend screening EdgeVendor record, NPI database, payment detailsIs vendor an HCP? Is payment transfer of value?Sunshine Act flag + categorizationCritical — DOJ/FCA defense
5Duplicate detection EdgeInvoice hash, historical paymentsSubmitted before?Duplicate flag or clearMedium — double payment
6Discrepancy resolution APIUnmatched items, PO terms, vendor historyLegitimate change or error?Resolution recommendationMedium-high — vendor relationship
7Approval routing EdgeInvoice amount, cost center, approval matrixWho must approve per SOX matrix?Routed to correct approverHigh — SOX compliance violation
8Payment authorization Kill gateMatched invoice, approval chain, amountAuthorize payment above SOX thresholdAuthorized or heldCritical — SOX, financial decisions policy
9Payment scheduling EdgeAuthorized invoices, payment termsOptimal payment date within termsScheduled payment batchLow — adjustable

Guardian agents

Match guardian

Between GRN matcher → HCP-spend screener

Validates three-way match (PO/GRN/invoice) within defined tolerance (1-2% or $100). Rejects invoices where line-item variance exceeds tolerance before compliance screening.

Compliance guardian

Between HCP-spend screener → Approval router

Cross-references every payment against NPI database for Sunshine Act reporting. Confirms categorization (meals, travel, consulting, research) matches payment description. Flags ambiguous vendor-HCP relationships.

Threshold guardian

Between Approval router → Payment authorization

Validates payment amount against SOX approval matrix. Confirms correct approver(s) in routing chain based on amount, cost center, and vendor category. Blocks payments bypassing required approval levels.

Trust escalation ladder

1
All invoices manually processed
AP Specialists capture, match, and route all invoices through Oracle Fusion.
Current baseline
2
AI captures and matches, human approves all
OCR capture + automated 3-way matching. Human reviews all matches and approves all payments. Fastest beachhead — 2–4 months.
Target: 0–2 months
3
Auto-process clean matches under threshold
Invoices with clean 3-way match below SOX threshold auto-process. Human reviews exceptions, above-threshold, and all HCP-flagged transactions.
Target: 2–6 months · Gate: ≥70% touchless, zero false negatives on HCP screening
4
70%+ touchless, human reviews flagged only
Majority of invoice volume processes without human touch. AP specialists shift to exception management, vendor negotiations, and compliance adjudication.
Target: 6–12 months · Gate: zero compliance exceptions over rolling 90 days
5
Full autonomy — N/A
SOX payment authorization above threshold + Regeneron AI policy ("financial decisions") + HCP-spend adjudication permanently require human approval.

Economics

Current estimated cost
$1–2M/yr
Savings at Rung 3
$0.5–1.2M/yr
Plus: compliance value
100% vs sample

Permanent kill gates

SOX payment authorization — Payments above threshold require human approval per internal control framework. External auditors rely on this control.

Regeneron AI policy — Explicitly bars AI from "financial decisions." Payment authorization is classified as a financial decision.

HCP-spend adjudication — Given active DOJ/FCA case, Sunshine Act compliance adjudication is a litigation defense function requiring human sign-off.

Rate this deep-dive analysis
✓ Rating saved

Talent acquisition — 9 decision points, 33% edge / 45% API / 22% human

Edge 33% API 45% Human 22%

Assembly line: JD optimizer → Sourcing strategist → Resume screener/ranker → Human: phone screen → Interview scheduler → Feedback synthesizer → Human: hire decision → Offer modeler → Background checker → Onboarding configurator

The human operator today

A TA Partner juggles twenty-five open requisitions: seven field sales roles for the COPD launch (cohort hiring for training classes), five manufacturing roles at the new $2B Saratoga Springs facility, four clinical operations, three RGC research positions, and assorted finance/IT/executive searches. With ~972 active postings (34.4% YoY increase), the volume pressure is real.

For Medical Specialist roles, she receives 80–150 applications per posting. First pass: 2–3 minutes per resume. She rejects 60–70% and advances the rest to phone screen. Phone screens take 20–30 minutes each. Then: interview coordination across panels, feedback synthesis, offer development, background checks, onboarding.

Decision-point map

#Decision pointAssignmentInputsJudgmentOutputConsequence
1JD optimization APIHiring manager intake, role requirementsInclusive language, SEO, positioningOptimized job postingLow — poor posting = fewer applicants
2Sourcing strategy APIRole type, talent market data, channelsWhich channels produce best candidates?Channel-prioritized sourcing planMedium — wrong channels = weak pipeline
3Resume screening & ranking APIApplications, JD criteria, hire profilesQualification fit, experience relevanceRanked candidate list + rationaleHigh — false negatives lose top talent
4Phone screen Kill gateResume, role brief, availabilityCommunication, motivation, cultureAdvance / reject + interview briefHigh — relationship assessment
5Interview scheduling EdgePanel availability, candidate preferencesOptimal time slot across constraintsConfirmed interview scheduleLow — reschedule if needed
6Feedback synthesis APIInterview scorecards, panel notesAggregate signal across interviewersSynthesized assessmentMedium — poor synthesis = biased decisions
7Hire decision Kill gateAssessment, headcount, comp bandHire / no-hire / continue searchHiring decision + justificationCritical — wrong hire costs 1.5–2× salary
8Offer modeling Edge + APIComp band, market data, expectationsBase + equity + sign-on within policyDraft offer for approvalMedium — uncompetitive = lost candidate
9Background check & onboarding EdgeCandidate data, vendor integrationInitiate checks, configure onboardingCleared candidate + Day 1 readinessLow — standard process

Guardian agents

Bias guardian

Between Resume screener → Phone screen

Validates no protected-class signals in ranking. Runs adverse impact analysis against the four-fifths rule (EEOC). Flags any ranking where a protected group's selection rate falls below 80% of the highest group's rate. Required for NYC LL144.

Completeness guardian

Between Feedback synthesizer → Hire decision

Validates all interview scorecards received before the assessment package is presented to the hiring manager. Blocks hire/no-hire until every scheduled interviewer has submitted structured feedback.

Compliance guardian

Between Offer modeler → Background check

Validates offer within approved comp band, equity within policy limits, sign-on within budget authority. Confirms OFCCP compliance for federal contractor positions.

Trust escalation ladder

1
All screening manual
TA Partners manually review all applications, 2–3 minutes per resume, 80–150 per posting.
Current baseline
2
AI ranks resumes, human reviews full list
AI produces ranked candidate list with rationale. TA Partner reviews full list, adjusts rankings, selects phone screen candidates. AI output is advisory.
Target: 0–3 months · Gate: ≥85% concordance with TA Partner top-20
3
AI auto-advances top-ranked, human reviews borderline
Top candidates above confidence threshold auto-advance to scheduling. TA Partner reviews borderline and rejections. Capacity: 35–40 reqs per partner.
Target: 3–9 months · Gate: bias guardian passes + satisfaction ≥4.0/5.0
4
AI handles sourcing through scheduling autonomously
JD optimization, sourcing, screening, ranking, and scheduling run end-to-end. Human conducts phone screens and presents feedback to hiring managers. Capacity: 40–50 reqs.
Target: 9–18 months · Gate: 12-month zero EEOC-actionable outcomes
5
Full autonomy — N/A
Hire/no-hire permanently barred by Regeneron AI policy, EEOC guidance, and NYC Local Law 144. The value capture is capacity, not headcount displacement.

Economics

Current estimated cost
$4.5–8M/yr
Efficiency gains
$2–4M/yr
Capacity increase
20–30 → 40–50 reqs

Permanent kill gates

Hire/no-hire decision — Regeneron AI policy explicitly bars AI from personnel hiring. EEOC guidance on automated employment decisions and NYC LL144 create additional regulatory constraints. Triple lock: policy + regulatory + legal.

Phone screen — Relationship assessment and cultural fit evaluation require human judgment. Candidate experience impacts offer acceptance and employer brand.

Offer negotiation — Compensation judgment at the individual level (equity weighting, sign-on structure, start date flexibility) requires human assessment.

Rate this deep-dive analysis
✓ Rating saved

15 intelligence stacks + 5 intelligence factories, ranked by value

Each assembly line maps a complete workflow — every layer, every data source, every kill gate. Scroll to explore all 15.

1 Tier 1 Stack → Factory $13–56M
Intelligence Stack

PA & benefit verification

Ingests patient/insurance data, predicts coverage, pre-populates PA forms, submits, tracks, and routes denials across four hub programs.

1
Case intake

Receives prescription and patient demographics, validates completeness, assigns priority score

AI-assisted
2
Insurance verifier

Looks up coverage status and plan details against payer eligibility databases

AI-assisted
3
Benefit investigator

Determines PA requirements, step therapy, tier placement for the specific drug and indication

AI-assisted
4
PA pathway router

Selects the correct payer-specific PA form and criteria pathway

AI-led
5
Clinical evidence assembler

Compiles diagnosis codes, lab values, and prior therapy documentation matching payer criteria

AI-led
6
PA form completer

Maps evidence package to payer form fields with correct formatting

AI-led
7
Submitter

Transmits completed form to the correct payer portal electronically

AI-led
8
Status tracker

Monitors submission status, triggers follow-up on overdue responses

AI-led
9
Denial analyzer

Parses denial letters, identifies root cause, generates appeal strategy options

AI-assisted
Human: appeal decision

Selects peer-to-peer, written appeal, or escalation based on clinical judgment

Human-only
11
Copay/PAP router

Determines which financial assistance program applies based on patient eligibility

AI-led
Human: eligibility check

Makes AKS/FCA-compliant eligibility determination — litigation defense function

Human-only
13
Enrollment processor

Configures patient in the correct assistance program with appropriate copay card

AI-assisted
14
Adherence monitor

Tracks refill patterns, appointment history, flags discontinuation risk

AI-led

Submitted PA with clinical evidence

40–60 cases/day per coordinator → auto-submission for routine payers

77%AI-automatable
23%Irreducibly human
Throughput gain

Kill gates: Copay/PAP eligibility (AKS/FCA — active DOJ) · Appeal strategy · HIPAA consent. Data: High. Phase 2 would map: Inter-stack handoff architecture and guardian agent specs at every boundary.

Commission full analysis →
2 Tier 1 Stack $1.5–2.5M+
Intelligence Stack

MLR content pre-review

Auto-checks promotional content against approved claims, current PI, and prior MLR decisions before human review.

1
Brief validator

Checks content brief completeness against template requirements — all fields present or return to brand

AI-led
2
Reference checker

Verifies every citation resolves to an approved primary source or PI section

AI-led
3
Claims-to-label mapper

Maps each efficacy claim against current prescribing information and approved claims matrix

AI-assisted
4
Fair balance analyzer

Measures risk/benefit prominence ratio — text area, font size, placement of safety information

AI-assisted
5
Pre-review synthesizer

Produces consolidated review package highlighting issues for human reviewers

AI-assisted
Human: Legal/Reg/Medical review

Three-person review committee applies clinical, legal, and regulatory judgment — permanent FDA requirement

Human-only
7
Revision checker

Validates all reviewer comments addressed in the revised asset before re-routing

AI-led
8
Approval recorder

Records final approval status, version, and complete audit trail in Veeva Vault

AI-led
9
Expiration monitor

Watches for PI updates, label changes, and asset expiration dates — triggers re-review

AI-led

Pre-reviewed asset with flagged issues

Cycle time from 2–3 weeks to 3–5 days

72%AI-automatable
28%Irreducibly human
Throughput gain

Kill gates: Medical review (FDA 21 CFR 202.1) · Legal · Regulatory strategy. Data: High — Veeva PromoMats. Phase 2 would map: Full claims-to-label mapping database and automated fair balance scoring.

Commission full analysis →
3 Tier 1 Stack $0.7–1.5M
Intelligence Stack

Accounts payable automation

OCR capture through three-way match, HCP-spend screening, and payment scheduling in Oracle Fusion. Fastest beachhead.

1
Data capture

OCR extraction from invoice PDF/image into structured fields — vendor, amount, PO reference, line items

AI-led
2
PO matcher

Line-by-line match of invoice items against purchase order within defined tolerance

AI-led
3
GRN matcher

Three-way match: validates goods/services receipt confirms delivery before payment

AI-led
4
HCP-spend screener

Cross-references vendor against NPI database — flags any payment that could be a transfer of value to an HCP

AI-led
5
Duplicate detector

Hashes invoice against historical payment records to prevent double-payment

AI-led
6
Approval router

Routes to correct approver(s) based on SOX approval matrix, amount, and cost center

AI-led
Human: above threshold

Authorizes payment above SOX dollar threshold — required by internal control framework and AI policy

Human-only
8
Payment scheduler

Optimizes payment timing within vendor terms to manage cash position

AI-led
9
Discrepancy resolver

Analyzes unmatched items against PO change orders and vendor history — recommends resolution

AI-informed
Human: adjustment approval

Approves manual adjustments for exceptions that cannot be auto-resolved

Human-only

Touchless invoice processing

67% edge functions — highest automation density of all 15

89%AI-automatable
11%Irreducibly human
Throughput gain

Kill gates: SOX payment threshold · AI policy ("financial decisions"). Data: High — Oracle Fusion. Phase 2 would map: Full 3-way match tolerance calibration and exception routing logic.

Commission full analysis →
4 Tier 1 Stack $2–4M
Intelligence Stack

Resume screening & sourcing

Screens and ranks applicants for ~972 open reqs. Does NOT make hire/no-hire decisions — Regeneron policy bars this.

1
JD optimizer

Rewrites job descriptions for inclusive language, SEO, and competitive positioning

AI-assisted
2
Sourcing strategist

Identifies highest-yield channels for each role type based on historical conversion data

AI-informed
3
Resume screener/ranker

Scores and ranks all applicants against JD criteria and historical hire profiles

AI-led
Human: phone screen

Assesses communication skills, motivation, and cultural fit — relationship assessment requiring human judgment

Human-only
5
Interview scheduler

Coordinates panel availability, candidate preferences, and room booking into optimal schedule

AI-led
6
Feedback synthesizer

Aggregates interview scorecards into a single assessment with consensus and dissent highlighted

AI-assisted
Human: hire decision

Makes hire/no-hire determination — permanently barred from AI by Regeneron policy, EEOC, and NYC LL144

Human-only
8
Offer modeler

Generates draft offer within approved comp band using market data and candidate expectations

AI-assisted
9
Background checker

Initiates vendor-integrated background verification and compliance checks

AI-led
10
Onboarding configurator

Sets up Day 1 readiness: systems access, training schedule, team introductions

AI-led

Ranked candidate pipeline with briefs

Capacity from 20–30 to 40–50 reqs per TA Partner

78%AI-automatable
22%Irreducibly human
Throughput gain

Kill gates: Hire/no-hire (AI policy + EEOC + NYC LL144) · Phone screen · Offer negotiation. Data: High. Phase 2 would map: Bias audit framework and candidate experience pipeline optimization.

Commission full analysis →
5 Tier 1 Stack $2–5M + risk
Intelligence Stack

HCP-spend compliance auditing

Scans 100% of HCP-payment transactions against Sunshine Act and AKS rules in real-time — vs human sampling of a fraction.

1
Transaction ingester

Pulls every expense and payment transaction from Oracle Fusion in real-time

AI-led
2
Vendor-to-HCP matcher

Cross-references vendor IDs against NPI lookup database to identify healthcare professional recipients

AI-led
3
Payment categorizer

Classifies each payment into Sunshine Act categories — meals, travel, consulting, research, education

AI-assisted
4
Meal/travel classifier

Applies per-meal and per-event dollar thresholds per CMS reporting requirements

AI-led
5
AKS rule screener

Screens payment patterns against Anti-Kickback Statute safe harbor provisions

AI-led
6
Aggregate spend calculator

Computes running annual spend per HCP for reporting threshold determination

AI-led
7
Anomaly detector

Identifies statistical outliers — unusual frequency, amount, or vendor patterns

AI-led
8
Pattern analyzer

Detects multi-transaction patterns that individual checks miss — split payments, recurring gifts

AI-assisted
9
Flag router

Routes flagged transactions to appropriate compliance officer by severity and category

AI-led
Human: compliance adjudication

Reviews flagged transactions and makes final determination — humans adjudicate, system advises

Human-only
11
Sunshine Act reporter

Generates CMS Open Payments submission files with complete audit documentation

AI-led
12
Audit trail logger

Records every screening decision with rationale for DOJ/FCA litigation defense

AI-led

100% transaction screening + audit trail

Critical given active DOJ/FCA matter — transforms sample-based to full-coverage

90%AI-automatable
10%Irreducibly human
10×Throughput gain

Kill gates: None — advisory flags to compliance officers. Stack scans, humans adjudicate. Data: High. Phase 2 would map: Exact categorization rules for ambiguous payments and threshold calibration.

Commission full analysis →
6 Tier 2 Factory $10–20M
Intelligence Factory

Patient access hub coordinator

Integrates PA/BV + copay routing + adherence outreach into an end-to-end hub agent consuming 50–65% of the coordinator role across four hubs.

1
Case intake

Receives and triages new patient cases across all four hub programs

AI-assisted
2
Insurance verifier

Validates coverage status and plan details for each patient

AI-assisted
3
Benefit investigator

Maps PA requirements by payer, plan, and indication

AI-assisted
4
PA pathway router

Selects correct submission pathway per payer rules

AI-led
5
Clinical evidence assembler

Compiles payer-specific evidence packages

AI-led
6
PA form completer

Auto-fills payer-specific PA forms from evidence

AI-led
7
Submitter

Electronically submits to correct payer portal

AI-led
8
Status tracker

Monitors submissions, triggers follow-up on overdue

AI-led
9
Denial analyzer

Parses denial letters and generates appeal options

AI-assisted
Human: appeal decision

Selects appeal strategy based on clinical judgment

Human-only
11
Copay/PAP router

Routes patient to appropriate financial assistance program

AI-led
12
Eligibility screener

Pre-screens patient against program criteria

AI-led
Human: AKS/FCA eligibility

Makes compliance-safe eligibility determination — active DOJ investigation

Human-only
14
Enrollment configurator

Sets up patient in correct program with copay card

AI-led
15
Adherence monitor

Tracks refill patterns and flags discontinuation risk

AI-led
16
Refill predictor

Forecasts next refill date and identifies patients at risk of gap

AI-assisted
17
Intervention trigger

Initiates outreach workflow when adherence score drops below threshold

AI-assisted
Human: clinical outreach

Conducts patient calls requiring clinical advice or sensitive communication

Human-only
19
Hub analytics dashboard

Aggregates program metrics: approval rates, cycle times, abandonment rates

AI-assisted

End-to-end patient access workflow

30–60 FTEs redeployed to exception management, appeal strategy, and compliance

70%AI-automatable
30%Irreducibly human
Throughput gain

Kill gates: AKS/FCA eligibility · HIPAA · Clinical advice. Data: High. Phase 2 would map: Inter-stack handoff architecture, escalation criteria, and guardian agent specs at each boundary.

Commission full analysis →
7 Tier 2 Factory $8–15M
Intelligence Factory

Field reimbursement analyst

Formulary-change detection → payer-policy retrieval → territory impact → FRM briefs → denial pattern analysis. Unifies the payer intelligence pipeline.

1
Formulary database monitor

Continuously scans formulary databases for status changes across all payers

AI-led
2
Change detector

Identifies material coverage changes — PA criteria, step therapy, tier moves, exclusions

AI-led
3
Payer-policy retriever

Pulls updated policy documents and coverage criteria from payer portals

AI-assisted
4
Impact quantifier

Calculates revenue impact of each formulary change by product and geography

AI-assisted
5
Territory mapper

Maps formulary changes to field territories and affected HCP practices

AI-assisted
6
Alert generator

Produces prioritized alerts for field reimbursement managers by urgency and impact

AI-assisted
7
FRM brief builder

Generates territory-specific briefs with talking points and payer-specific data

AI-assisted
8
Claim denial analyzer

Analyzes patterns in claim denials by payer, reason code, and geography

AI-assisted
9
Root-cause classifier

Identifies systemic denial patterns vs one-off errors — feeds PA workflow improvements

AI-led
Human: appeal strategy

Sets appeal approach for systemic denial patterns requiring payer negotiation

Human-only
11
CRM push

Logs all payer intelligence and field actions into CRM for territory tracking

AI-led
12
Competitive tracker

Monitors competitor formulary positioning and market access moves

AI-led
13
Territory analytics dashboard

Aggregates field metrics: win rates, denial trends, coverage shifts

AI-assisted

Real-time payer intelligence to field teams

15–30 FTEs consumed across formulary tracking, field reimbursement, and sales data

75%AI-automatable
25%Irreducibly human
Throughput gain

Kill gates: Pricing/contract approvals · Government-price certification · FRM compliance boundaries. Data: Partial. Phase 2 would map: Claim-denial root-cause diagnosis — pattern-based, data-rich, high-volume.

Commission full analysis →
8 Tier 2 Stack $2–4M
Intelligence Stack

Formulary & coverage tracking

Monitors formulary status across all payers, detects coverage changes, quantifies revenue impact, and generates real-time field alerts.

1
Formulary database monitor

Continuously polls formulary databases and payer portals for updates

AI-led
2
Coverage policy scraper

Extracts updated coverage criteria, PA requirements, and step therapy rules from payer sites

AI-assisted
3
Change detector

Identifies material vs routine changes and flags new coverage restrictions

AI-led
4
Material-vs-routine classifier

Separates changes requiring field action from administrative updates

AI-led
5
Revenue impact quantifier

Models financial impact of each change by product, geography, and patient volume

AI-assisted
6
Geographic mapper

Maps each change to affected territories, zip codes, and prescriber practices

AI-assisted
7
Alert priority scorer

Ranks alerts by urgency: exclusion > new PA > tier change > minor update

AI-assisted
8
Field alert generator

Produces formatted alerts with context for field teams via CRM and email

AI-assisted
9
Competitive position tracker

Compares Regeneron formulary status vs competitors across key payers

AI-led
10
Contract data assembler

Compiles coverage data for contract negotiation support

AI-led
11
PA workflow updater

Pushes updated payer requirements to the PA submission pipeline (S1)

AI-led

Continuous formulary intelligence

24/7 monitoring vs periodic manual review — never miss a coverage change

95%AI-automatable
5%Irreducibly human
24/7Throughput gain

Kill gates: None — analytical. Data: High (MMIT, payer portals). Phase 2 would map: Complete change-detection-to-field-alert pipeline at decision-point level.

Commission full analysis →
9 Tier 2 Stack $3–6M
Intelligence Stack

Sales pre-call planning

Daily pre-call briefs for Medical Specialists: territory data → opportunity scoring → HCP-specific brief. Assist layer, not FTE displacement.

1
Territory data assembler

Aggregates prescribing data, payer mix, and market share by territory

AI-led
2
Prescription trend analyzer

Identifies prescribing trends, new prescribers, and declining accounts

AI-assisted
3
Opportunity scorer

Ranks HCPs by conversion probability and revenue potential

AI-assisted
4
Call schedule optimizer

Generates optimal daily call route factoring geography, priority, and access windows

AI-assisted
5
HCP profile builder

Compiles individual HCP dossier: prescribing history, payer coverage, last interaction, publications

AI-assisted
6
Payer coverage mapper

Maps each HCP's patient panel to current formulary status and PA requirements

AI-assisted
7
Clinical data matcher

Identifies relevant new clinical data or label updates for the HCP's specialty

AI-led
8
Pre-call brief generator

Produces one-page brief per HCP with talking points, objection responses, and data highlights

AI-assisted
9
Post-call CRM assist

Pre-populates CRM call notes from scheduled activities for rep to confirm/edit

AI-assisted
10
Territory analytics dashboard

Aggregates call metrics, conversion rates, and territory performance

AI-assisted
11
Weekly review preparer

Generates weekly territory summary with wins, losses, and priority shifts

AI-assisted

Daily pre-call briefs by 7 AM

Throughput multiplier — every rep starts the day with HCP-specific intelligence

85%AI-automatable
15%Irreducibly human
Throughput gain

Kill gates: MLR-approved content only · Sunshine Act. Data: High. Phase 2 would map: The brief-generation pipeline as a daily automated agent with output ready by 7 AM.

Commission full analysis →
10 Tier 2 Stack $2–5M
Intelligence Stack

Gross-to-net calculations

Automates Medicaid rebate, managed care rebate, chargeback, 340B, PAP cost, and returns accruals. GTN is 30–50% of gross revenue.

1
Sales data ingester

Pulls transaction-level sales data by channel, customer, and contract

AI-led
2
Contract-rate lookup

Maps each transaction to applicable rebate, chargeback, or discount rate

AI-assisted
3
Accrual calculator

Computes accruals by channel — Medicaid, managed care, 340B, PAP, returns

AI-led
4
Reconciliation matcher

Matches accrued amounts against actual claims received quarterly

AI-led
5
Variance flagger

Identifies material variances between accrued and actual for investigation

AI-assisted
6
Reserve assessor

Evaluates adequacy of reserves based on historical patterns and pipeline data

AI-informed
Human: Controller sign-off

Certifies reserve adequacy and accrual accuracy for financial statements — SOX requirement

Human-only
8
Audit package assembler

Compiles supporting documentation for external audit review

AI-led

Automated accrual calculations + reconciliation

Accrual accuracy directly impacts reported net revenue ($12.17B)

80%AI-automatable
20%Irreducibly human
Throughput gain

Kill gates: SOX reserve certification · External audit reliance. Data: Low-medium. Phase 2 would map: Which estimation steps are deterministic vs judgment — critical for a $14B company.

Commission full analysis →
11 Tier 2 Stack $1.5–3M
Intelligence Stack

Financial close & reconciliation

Automates month-end: journals, reconciliations, intercompany eliminations, currency translation, variance analysis. Close from ~10 days to ~5.

1
Close checklist orchestrator

Sequences all close tasks with dependencies and deadlines

AI-led
2
Journal poster

Posts recurring journal entries — accruals, prepaid amortization, depreciation

AI-led
3
Account reconciler

Reconciles all balance sheet accounts against sub-ledgers and third-party statements

AI-assisted
4
IC eliminator

Processes intercompany eliminations across Ireland, UK, Japan, and US entities

AI-assisted
5
FX translator

Applies month-end exchange rates for currency translation of foreign subsidiaries

AI-assisted
6
Variance calculator

Computes period-over-period and budget-vs-actual variances by cost center

AI-led
7
Narrative generator

Drafts variance commentary for management reporting using API reasoning

AI-assisted
Human: Controller review

Reviews and certifies financial statements — SOX attestation requirement

Human-only

5-day financial close

Frees capacity from processing to analysis — controller reviews exceptions only

82%AI-automatable
18%Irreducibly human
Throughput gain

Kill gates: SOX · Controller sign-off. Data: High — Oracle Fusion GL. Phase 2 would map: Judgment-layer decisions (accruals, unusual items) that generic close tools miss.

Commission full analysis →
12 Tier 2 Stack $2–4M
Intelligence Stack

Omnichannel HCP campaigns

Automates HCP segmentation, content variant generation, campaign config, A/B testing, and performance reporting across all channels.

1
Audience segmenter

Segments HCPs by specialty, prescribing behavior, and engagement stage

AI-assisted
2
Prescribing behavior analyzer

Identifies prescribing patterns, brand affinity, and switching signals

AI-assisted
3
Content variant generator

Produces content variants tailored to each segment using approved claims

AI-assisted
4
MLR routing

Submits all content variants through the MLR stack (S2) for compliance clearance

AI-assisted
5
Campaign configurator

Sets up multi-channel campaign parameters — targeting, frequency, sequencing

AI-led
6
Channel mix allocator

Optimizes channel allocation (email, rep-triggered, digital, congress) by HCP preference

AI-assisted
7
Send-time optimizer

Determines optimal delivery timing based on HCP engagement patterns

AI-assisted
8
A/B test executor

Runs controlled tests on subject lines, content variants, and channel combinations

AI-assisted
9
Performance tracker

Monitors open rates, click-through, and downstream prescribing impact

AI-led
10
Script lift attributor

Attributes prescribing changes to specific campaign touchpoints

AI-assisted
11
ROI reporter

Generates campaign ROI analysis by channel, segment, and brand

AI-led

Personalized HCP engagement at scale

Content variants by specialty, indication, and engagement stage — all MLR-cleared

80%AI-automatable
20%Irreducibly human
Throughput gain

Kill gates: All HCP content through MLR. Data: High. Phase 2 would map: Segmentation-to-content-to-MLR-to-deployment as a complete agent chain.

Commission full analysis →
13 Tier 2 Factory $4–7M
Intelligence Factory

Finance operations center

Integrates AP + financial close + internal audit/SOX. Orchestration layer manages close sequencing and continuous control monitoring.

1
Invoice capture

OCR extraction from invoices into Oracle Fusion

AI-led
2
PO matcher

Automated purchase order matching

AI-led
3
GRN matcher

Three-way goods receipt validation

AI-led
4
HCP-spend screener

Sunshine Act compliance screening on every payment

AI-led
5
Duplicate detector

Prevents double-payment across the invoice stream

AI-led
6
Approval router

Routes to correct approver per SOX matrix

AI-led
Human: above threshold

Authorizes payments above SOX dollar threshold

Human-only
8
Payment scheduler

Optimizes payment timing within vendor terms

AI-led
9
Close checklist orchestrator

Sequences month-end close tasks across all entities

AI-led
10
Journal poster

Posts recurring entries and accruals

AI-led
11
Account reconciler

Reconciles balance sheet against sub-ledgers

AI-assisted
12
IC eliminator

Processes intercompany eliminations

AI-assisted
13
FX translator

Applies exchange rates for foreign subsidiaries

AI-assisted
14
Variance calculator

Computes period and budget variances

AI-led
15
Narrative generator

Drafts management commentary on variances

AI-assisted
Human: Controller review

Certifies financial statements per SOX

Human-only
17
SOX control tester

Tests 100% of transactions against control criteria — replaces sampling

AI-led
18
Exception flagger

Identifies control failures and unusual patterns for investigation

AI-assisted
19
Audit package assembler

Compiles audit-ready documentation for external auditors

AI-led
Human: audit sign-off

Final sign-off on audit packages — external auditor reliance

Human-only

Integrated finance operations with 100% control testing

Composed of S3 + S11 + Internal audit/SOX · 15–25 FTEs consumed

78%AI-automatable
22%Irreducibly human
Throughput gain

Kill gates: SOX · External auditor reliance · Payment thresholds. Data: High — Oracle Fusion. Phase 2 would map: Continuous control monitoring — testing 100% of transactions vs sampling.

Commission full analysis →
14 Tier 3 Factory $5–10M
Intelligence Factory

Commercial content engine

Full content generation + MLR + localization + DAM + deployment. Requires mature MLR-agent trust (Rung 3+).

1
Content brief ingester

Receives brand team brief with objectives, audience, and key messages

AI-led
2
Brand strategy validator

Checks brief alignment with current brand strategy and messaging hierarchy

AI-led
3
Content variant generator

Produces content variants for each channel, specialty, and indication

AI-assisted
4
Claims-to-label mapper

Validates every claim against approved labeling and claims matrix

AI-assisted
5
Fair balance analyzer

Ensures risk/benefit prominence meets FDA requirements

AI-assisted
6
Reference checker

Verifies all citations trace to approved sources

AI-led
7
Pre-review synthesizer

Produces review package highlighting issues for MLR committee

AI-assisted
Human: Legal/Reg/Medical MLR review

Three-person review — permanent FDA promotional compliance requirement

Human-only
9
Revision checker

Validates all review comments addressed before publication

AI-led
10
DAM publisher

Publishes approved assets to digital asset management system with metadata

AI-led
11
Audience segmenter

Selects target HCP segments for each approved content piece

AI-assisted
12
Campaign configurator

Sets up campaign parameters across channels

AI-led
13
Send-time optimizer

Schedules delivery for maximum engagement

AI-assisted
14
Channel deployer

Distributes approved content through email, digital, rep-triggered, and congress channels

AI-assisted
15
Performance tracker

Monitors engagement and prescribing impact

AI-led
16
ROI reporter

Attributes revenue impact to content investments

AI-led
17
Congress material coordinator

Manages poster, booth, and symposium content timelines

AI-assisted

Brief-to-deployment content pipeline

Composed of S2 + S12 + Congress/med-ed · 20–35 FTEs · Why Tier 3: needs MLR at Rung 3 first

72%AI-automatable
28%Irreducibly human
Throughput gain

Kill gates: FDA promotional (permanent) · Sanofi co-promo terms. Data: High. Phase 2 would map: Content-brief-to-deployed-asset pipeline across all channels and brands.

Commission full analysis →
15 Tier 3 Factory $5–12M + risk
Intelligence Factory

Government pricing & GTN engine

Automates Medicaid Best Price, AMP, ASP, 340B calculations and CMS/HRSA submissions. Highest-value AND highest-risk.

1
Sales data ingester

Pulls transaction-level data by channel, customer, contract, and NDC

AI-led
2
Contract-rate lookup

Maps each transaction to applicable government or commercial contract rate

AI-assisted
3
Channel classifier

Categorizes each sale into correct pricing channel for statutory calculations

AI-led
4
Medicaid Best Price calculator

Computes Best Price per CMS definition across all commercial transactions

AI-led
5
AMP calculator

Calculates Average Manufacturer Price per statutory methodology

AI-led
6
ASP calculator

Computes Average Sales Price for Medicare Part B reimbursement

AI-led
7
340B ceiling price calculator

Derives ceiling price for 340B covered entities per HRSA methodology

AI-led
8
MFN compliance checker

Validates pricing against April 2026 Most Favored Nation framework requirements

AI-led
9
Accrual calculator

Computes government rebate and discount accruals by program

AI-led
10
Reconciliation matcher

Matches accruals against actual government claims and chargebacks

AI-led
11
Variance flagger

Identifies material calculation variances for investigation

AI-assisted
12
Reserve assessor

Evaluates adequacy of government pricing reserves

AI-informed
Human: Controller sign-off

Certifies calculation accuracy for financial statements

Human-only
14
CMS submission preparer

Formats and validates government price reports for CMS submission

AI-led
15
HRSA reporter

Prepares 340B ceiling price calculations for HRSA reporting

AI-led
Human: CFO pricing certification

Certifies government pricing accuracy — FCA liability exceeds $100M if incorrect

Human-only
17
Audit package assembler

Compiles government pricing audit trail for external review

AI-led

Compliant government price calculations + submissions

Composed of Gov pricing + GTN (S10) + Contract analytics · 10–20 FTEs · Why Tier 3: MFN rules unstable

75%AI-automatable
25%Irreducibly human
Throughput gain

Kill gates: FCA pricing certification ($100M+ liability) · MFN compliance (rules evolving) · HRSA 340B. Data: Medium. Phase 2 would map: Exactly which calculation steps are deterministic vs interpretive.

Commission full analysis →

Three-phase rollout: prove in 90 days, scale by month 18

Each phase has explicit stage-gate criteria. Failure to meet gates extends the phase — do not proceed.

Phase 1
0–3 mo
Prove the model
$2–4M est. value

Deploy two beachhead stacks with mature vendor tech, existing data, and fast ROI. These fund the program politically and financially.

MLR pre-reviewAP automationResume screening
Stage gate: ≥50% reduction in MLR reviewer hours per asset and ≥70% touchless AP invoices within 90 days.
Phase 2
3–9 mo
Attack highest value
$13–56M est. value

Build the Patient Access Hub factory — the largest automatable workforce across four hubs. Directly improves speed-to-therapy and protects revenue from prescription abandonment.

PA/BV → Patient hub factoryHCP-spend complianceSales pre-call planning
Stage gate: ≥75% PA auto-submission accuracy and <2% compliance-exception rate. If exceptions exceed 5%, hold at stack level.
Phase 3
9–18 mo
Decision-layer differentiation
$8–20M est. value

Deploy decision-layer factories requiring process-level integration — where the task→decision→process hierarchy creates value. Government pricing deferred until MFN rules stabilize.

Field reimbursement factoryHEOR dossier generationFinancial closeIR earnings prep
Stage gate: Benchmark against Sanofi plai and Lilly ROI data to calibrate targets. Government pricing factory deferred until MFN rules stabilize.

From map to deployment — Phase 2 engagement

This analysis mapped 27 workflows at screening level and 4 at full decision-point depth. Phase 2 goes inside.

What Phase 2 produces

A 4–6 week engagement with internal data access — SOPs, approval matrices, system configurations, staffing data — to produce full Decision-Point Architecture for your highest-priority workflows.

Full decision-point maps

Every judgment node documented with inputs, logic, output, volume, and consequence. Each assigned through the Jagged Edge Trilemma with specific justification.

Agent assembly line designs

Complete pipeline specifications: named agents, guardian agents at every handoff, validation rules specific enough to implement, and trust escalation ladders with promotion/demotion metrics.

Implementation blueprints

Vendor recommendations, build-vs-buy analysis, data integration requirements, and 120-day deployment plan per workflow following the Five-Phase Transformation sequence.

Scope: 7 next-priority workflows (the ★★★ candidates from the inventory) elevated to full DPA depth. Includes the complete trust escalation ladder, guardian agent specifications, and economics model for each — producing a board-ready business case with stage-gate criteria.
TBD Phase 2 · Decision-Point Architecture engagement

4–6 week engagement producing full Trilemma assignments, Agent Assembly Line designs, Guardian Agent specifications, Trust Escalation Ladders, and implementation blueprints for 7 priority workflows. Delivered as an interactive report with board-ready economics.

Book a 30-minute call →

Framework, sources, and limitations

Framework: Intelligence Factory methodology

This analysis applies the Decision-Point Architecture framework from Code Is Speech: The New Fluency (Orr & de Jong, 2026). The methodology maps every decision point in a workflow, documents its inputs, judgment, output, volume, and consequence, and assigns each to the optimal intelligence layer through the Jagged Edge Trilemma: deterministic edge functions (fast, cheap, private), generative API calls (reasoning, novelty), or human operators (judgment, governance, relationships). Decision points are then chained into Agent Assembly Lines — pipelines of specialized agents with Guardian Agents validating quality at every handoff. Autonomy is governed by the Trust Escalation Ladder — graduated authority earned through demonstrated accuracy, not granted by executive mandate.

Sources

Regeneron Q2 2026 earnings release and call transcript · 2025 10-K annual report · regeneron.com (About, Careers, Leadership, Medicines, Pipeline) · Regeneron Position Statement on Responsible Use of AI (July 2025) · Job postings (regeneron.com/careers, LinkedIn, Glassdoor) · Revelio Labs workforce composition estimates · Peer disclosures (Sanofi plai, Eli Lilly/Yseop, Veeva Falcon MLR, Develop Health Calibrate, Pfizer Charlie) · Gartner agentic AI research · Industry workflow standards (AMCP, FDA promotional guidance, NPDES, SOX, AKS/FCA).

Outside-in limitations

This is an outside-in analysis: exact headcounts by function, field-force size, hub call volumes, and internal system configurations are not publicly disclosed. FTE and dollar estimates are directional inferences from peer benchmarks and Regeneron's disclosed SG&A ($2.70B GAAP), not internal data. Peer ROI figures (Veeva "70%+," Sanofi "$300M," Lilly "53%," Develop Health "80–85%") come from vendor or company communications and should be treated as directional, best-case claims. Gartner warns "over 40% of agentic AI projects will be canceled by the end of 2027" — execution risk is real.

What internal data would sharpen this analysis

SOPs and approval matrices for each priority workflow · Hub program staffing models and call volumes · Veeva Vault PromoMats configuration and approved claims matrix · PA approval/denial rates by payer and indication · Prescription abandonment rates during PA · Oracle Fusion ERP configuration details · Current automation baseline in each department · Sanofi's role in Dupixent MyWay operations · Internal AI Advisory Committee evaluation criteria.

With this data, each of the 23 screening-level workflows could be elevated to full Decision-Point Architecture depth — producing complete Trilemma assignments, agent assembly line designs, guardian agent specifications, and trust escalation ladders. That is the natural second phase of this engagement.