Cancer second-opinion copilot that assembles fragmented records, surfaces treatment discordance, and helps avoid overtreatment.

1. A single lymphoma patient needed 12 opinions after opposite chemotherapy recommendations, which shows real demand for software that can organize disagreement before treatment starts.
2. The case proves that multimodal patient evidence can already be compiled into a useful AI-assisted question workflow without asking clinicians to trust autonomous diagnosis.
3. An ambiguous PET scan almost led to unnecessary radiotherapy, making overtreatment avoidance a concrete ROI story rather than an abstract efficiency claim.
4. Because the clinically acceptable AI role is support around physician judgment, a workflow product can be adopted sooner than a full diagnostic system.

Catalyst. The Keragon case shows that general AI plus patient-generated data can now materially improve question quality in a complex cancer workflow, while the consequences of missing an alternative explanation can be unnecessary chemotherapy or radiotherapy.

Oncology Discordance Copilot plugs into document retrieval feeds, patient uploads, imaging-report text, labs, and optional wearable or symptom data to build a single longitudinal case file before a specialist touches the chart. The product does not diagnose cancer; it structures evidence, surfaces missing records, compares treatment recommendations across prior opinions, and flags ambiguous follow-up findings that deserve targeted questions rather than blind escalation. Coordinators get a work queue for incomplete or contradictory cases, while physicians receive a short briefing with timelines, unresolved conflicts, and patient-specific questions for the consult. The first module is sold as second-opinion operations software, which keeps regulatory exposure lower while still saving clinician time and reducing avoidable overtreatment.

What's different. This is not a generic patient chatbot and not another LLM note summarizer for clinics. The product is purpose-built for one painful workflow: external cancer cases where records are fragmented, prior specialists disagree, and the next decision could lock a patient into months of toxic therapy. Over time the moat comes from case-assembly integrations, dispute-resolution datasets, and a proprietary map of which evidence gaps and conflict patterns actually change final oncology decisions.

Jobs to be done

flowchart LR
  Buyer[Second-opinion program director] --> Pain[Complex cancer referrals arrive as fragmented conflicting records]
  Pain --> Product[Oncology Discordance Copilot]
  Product --> Outcome[Faster case review safer decisions and fewer unnecessary interventions]

Executive takeaways

• The most credible opening is an operations layer for expert cancer review, not an autonomous oncology recommender.
• Rare-cancer and pathology-heavy referrals create the sharpest ROI because specialist re-review often changes diagnosis or management.
• The best early accounts are cancer centers that already run remote or online second-opinion programs and need more throughput from the same specialist labor.
• Competition is real but fragmented; differentiation has to come from external-record reconciliation and disagreement mapping rather than generic summarization.

Market definition

U.S. software and workflow infrastructure for assembling external oncology referrals into specialist-ready second-opinion cases, initially focused on rare or aggressive cancers at academic centers and large oncology networks. The wedge sits upstream of the consult and tumor board: timeline reconstruction, missing-record detection, pathology and imaging conflict surfacing, and question-pack generation.

Customer and buyer

Primary users are nurse navigators, second-opinion coordinators, tumor-board staff, and disease-specific physicians who inherit outside records from community settings. The economic buyer is usually the oncology service-line leader, CMO, or second-opinion program director who owns throughput, specialist utilization, and referral quality.

Buying triggers

• Centers that launch or scale written or virtual second-opinion services need a repeatable intake and record-assembly layer behind them. [50][51][53][58]
• Rare-cancer and pathology-heavy referrals justify budget because specialist re-review often changes diagnosis, treatment choice, or treatment intensity. [26][27][28][30][38][39]
• Navigator and tumor-board teams hit manual prep bottlenecks well before physician demand is exhausted. [40][42][78]

Willingness to pay

Willingness to pay is already visible in the market. Leading cancer centers sell self-pay second opinions at meaningful price points—Cleveland Clinic lists $1,690 to $1,990 in the U.S. and Dana-Farber lists $3,000—while AccessHope reports strong employer and payer demand for expert cancer review. That makes a sub-$1,000 software value per complex case a conservative modeling assumption rather than an aggressive one. [51][53][72]

Category dynamics

Growth signal +3.6% YoY estimated U.S. new cancer cases (2025→2026)

Validation signals

• Specialist second opinions materially change diagnosis or management in sarcoma, lymphoma, gynecologic, and breast imaging contexts.
• AccessHope shows asynchronous oncology review can turn around cases in about five days and still surface management changes and cost savings.
• Leading cancer centers now openly market remote or online second opinions, often with explicit cash-pay pricing.
• Oncology programs are already adopting AI documentation and virtual tumor-board tooling, lowering skepticism toward workflow assistance.

Regulatory & technical constraints

• ONC interoperability and information-blocking rules create a path to electronic health information access but also impose strict expectations on how actors share or withhold data.
• If the product crosses from evidence assembly into treatment recommendation logic, it risks moving closer to FDA-regulated clinical decision support.
• Cancer data still spans USCDI, FHIR, mCODE, DICOM, PDFs, and external portals, making normalization a central technical risk.
• CAP places explicit restrictions on commercial and AI-system reuse of its current cancer protocols, which affects pathology-template strategy.

Competition comes from expert-review services, tumor-board and care-coordination platforms, oncology navigation vendors, and ambient documentation tools, plus the entrenched substitute of nurse navigators assembling packets by hand inside the EHR. The white space is a cancer-center-owned operating layer that reconciles fragmented outside evidence before a specialist review begins.

Why incumbents do not win by default

• Expert review services. They prove demand for remote oncology expertise, but they remain service-heavy and do not give cancer centers their own internal case-assembly operating layer.
• Tumor board workflow platforms. They are stronger at internal conference orchestration and data aggregation than at reconciling messy external referral packets before the first specialist touch.
• Ambient documentation platforms. They automate conversations and notes after clinicians engage, but they are not built around outside-record reconciliation, pathology disagreements, and pre-consult question packs.
• Navigation and patient-engagement platforms. They help with triage, ePROs, and patient messaging, but they do not own the specialist-facing disagreement map that changes second-opinion prep quality.

Oncology Discordance Copilot is an operations-layer software product for U.S. academic cancer centers that converts fragmented, conflicting external referral records into specialist-ready case briefings before a second-opinion consult or tumor board begins. The beachhead pain is measurable and severe: rare-cancer patients require 12 or more opinions when two oncologists recommend opposite chemotherapy regimens, coordinators spend hours assembling PDFs and portal screenshots, and ambiguous follow-up imaging has led to unnecessary radiotherapy that could have been avoided with cleaner evidence synthesis. The product does not diagnose cancer; it structures timelines, surfaces recommendation conflicts, flags missing records, and generates targeted specialist question packs—workflow tasks that manual navigation and generic note-summarization tools cannot perform. The wedge is second-opinion program operations software sold to NCI-designated cancer centers receiving 50 or more complex lymphoma and sarcoma referrals per month, a segment where specialist re-review is already documented to change diagnosis or management and where centers charge $1,690–3,000 per self-pay case, making sub-$1,000 software value per case a conservative model. Research across 84 verified sources supports a U.S. beachhead SAM of roughly $22.5M and a modeled Year-3 reachable share of $3.6M across six landed centers. The primary regulatory posture is evidence-assembly software below FDA clinical-decision-support thresholds, which allows faster commercial deployment than an autonomous oncology recommender while still delivering overtreatment-avoidance ROI. The long-run moat is a proprietary dataset linking outside-record conflict patterns to eventual diagnosis or treatment changes—data no incumbent currently owns. The seed round of $3–5M funds two paid 90-day pilots, four core engineering hires, HIPAA/SOC 2 compliance, and the first enterprise sales hire, targeting three signed production contracts and $1.2M ARR by month 18.

Problem

• Rare-cancer second-opinion programs receive external referrals as scattered PDFs, portal screenshots, imaging reports, and symptom notes spread across systems; nurse navigators spend multiple hours per case manually assembling packets before a specialist can begin clinical review.
• When prior specialists disagree on regimen choice or follow-up imaging is ambiguous, no current tool maps the conflicts, identifies missing evidence, or generates targeted questions—forcing avoidable treatment delays and raising overtreatment risk at the highest-stakes moment of a patient's illness.

Solution

• A coordinator-facing case-compilation engine that ingests uploaded and retrieved outside records—PDFs, imaging reports, labs, wearable exports, symptom histories—and outputs a structured longitudinal timeline, a missing-record checklist, and a recommendation-conflict map before the first specialist touch.
• A specialist-facing pre-consult briefing module that presents unresolved treatment conflicts and an AI-generated question pack, enabling physicians to focus on judgment rather than chart assembly and reducing avoidable escalation when ambiguous findings have alternative explanations such as thymus rebound.

Why we win

• No existing product reconciles fragmented external oncology records into a specialist-ready disagreement map: AccessHope is service-heavy, OncoLens focuses on post-intake conference orchestration, Abridge and Ambience target in-session ambient documentation, and navigation platforms manage ongoing care rather than pre-consult evidence synthesis.
• Each case reviewed enriches a proprietary conflict-to-outcome dataset—linking evidence gaps and recommendation clashes to final diagnosis or management changes—that compounds into a moat competitors cannot replicate without years of similar workflow deployment.
• Operations-software regulatory positioning (evidence assembly + question generation, never treatment recommendation) allows clinical trust to build faster than autonomous-AI competitors while satisfying CMO liability requirements through full audit trails and mandatory human sign-off.
• Beachhead accounts already market and charge for second opinions at $1,690–3,000 per case, making a sub-$1,000-per-case software value a demonstrably conservative ROI argument rather than an aspirational one.

Milestones

flowchart LR
  Wedge[Second-opinion program director\nat NCI cancer center] --> Pain[Fragmented external referrals\ndelay specialist review]
  Pain --> MVP[Case-compilation and\ndiscordance copilot MVP]
  MVP --> Proof[Documented prep-time\nreduction in 90-day pilot]
  Proof --> Expand[Tumor-board module +\nemployer/payer pathway]
  Expand --> Moat[Conflict-to-outcome\ndataset + integrations]

Founding team

Experiment roadmap

Risk assessment

What must be true

• Second-opinion program directors at NCI-designated centers will authorize $60–120K+ annual software spend when a 90-day pilot documents ≥30% reduction in coordinator prep time per case.
• The copilot can normalize at least 70% of incoming external referral documents—PDFs, imaging reports, labs—into a usable structured timeline without requiring manual re-entry by coordinators.
• Specialist oncologists will act on AI-generated disagreement maps and question packs when those outputs include explicit human sign-off checkpoints and a full audit trail, satisfying CMO liability requirements.
• At least 20–25 NCI-designated or peer cancer centers have sufficient referral volume (≥50 complex eligible cases per month) and program maturity to justify the initial platform fee within a 2-year commercial window.
• Recommendation conflicts and missing-evidence patterns are common enough in rare-cancer referral cohorts that a 90-day single-pathway pilot surfaces enough decision-relevant signals to constitute defensible ROI evidence.

Open diligence questions

• How many eligible external referrals does this center process per month in lymphoma, sarcoma, and other rare cancers, and what share arrive with digital pathology and imaging access versus PDF-only reports?
• What liability sign-off language and human-review checkpoints would your CMO require before allowing an AI-generated disagreement map into pre-consult workflow?
• Who captures first-year ROI—the center's own second-opinion program, an employer or payer contract partner, or both—and does that affect where the software budget sits?
• How long does your typical oncology software procurement cycle run from pilot start to production contract, and is the pilot budget owned by the program director or requires VP or C-suite approval?
• Which of AccessHope, OncoLens, Abridge, an in-house EHR workflow, or nurse navigator headcount is the closest current substitute, and what would an incumbent need to ship to make this solution redundant?
• Have you seen cases where an ambiguous follow-up finding—imaging, pathology, or conflicting prior regimen—delayed treatment or led to an avoidable escalation in the past 12 months, and how was it resolved?

Model sanity

• Revenue engine. Base-case revenue comes from converting two paid pilots into 6 live centers by Q4Y3 and expanding mature-center annualized revenue toward about $600K through case fees and early module upsells.
• Must go right. The company must keep pilot-to-production timing near the 6-month base case because sales-cycle slippage is the largest single hit to both Y3 revenue and cash runway in the sensitivity table.
• Model breaks if. If mature-center ARPU stays nearer $500K or gross margin slips to 67%, the downside case pushes cash below zero before the next round closes.
• Next-round proof. The next financing is justified once the team proves roughly $1.2M ARR by month 18 and then shows a repeatable path from 4 live centers to 6 live centers with module expansion.

Scenarios

Sensitivity

flowchart LR
  Leads[Target cancer centers] --> Pilots[90-day paid pilots]
  Pilots --> LiveCenters[Production centers]
  LiveCenters --> Cases[Eligible referral cases]
  Cases --> Revenue[Platform + case-fee + module revenue]
  Revenue --> GrossProfit[70% gross profit]
  GrossProfit --> Cash[Runway and next-round timing]

Flags: The base case requires mature-center revenue to reach roughly $600K annualized by Y3 exit, which is meaningfully above the entry platform minimum and therefore depends on case-fee and module expansion materializing. · Cash still falls to about $422K by Y3 exit while EBITDA remains negative, so the company still needs a timely Series A rather than assuming self-funding from operations. · Revenue concentration is high because only 6 centers drive the full Y3 plan; one delayed flagship account would materially change both ARR and runway.

• Clinical-liability fear. Cancer centers may worry that any AI touching treatment workflows creates medico-legal exposure if clinicians over-trust the output. Mitigation: Keep the first product focused on evidence assembly and question generation, require human sign-off everywhere, and publish workflow audit trails instead of diagnostic claims.
• Thin initial evidence base. The source signal is powerful but comes from one founder case study, so adoption could be slower if the problem is less frequent or less budgeted than it appears. Mitigation: Pilot with centers already marketing second opinions, measure case-prep time and decision-delay reduction, and concentrate on referral-heavy rare-cancer programs where pain is easiest to prove.
• Integration drag. Outside oncology records span portals, PDFs, imaging reports, and pathology systems, which can slow implementation and reduce early ROI. Mitigation: Start with lightweight patient-upload and coordinator-upload workflows, then add deeper retrieval integrations only after proving value on a narrow referral pathway.