crossbeamHEALTH

Specialty judgment at the threshold of every referral.

Concept draft for clinician feedback   |   July 2026   |   US market   |   Confidential draft, not an offering

DRAFT

"Crossbeam Health" is a draft working name. Nothing in this deck is an offer of products, services, or securities. Photo: AI-generated illustration.
The problem

The referral decision is made alone, and the patient pays for it.

A PCP decides "this needs a specialist" with no specialist input, no protocol, and no feedback. The patient is referred, waits weeks, arrives without the right workup, or never arrives at all.

100M+
specialty referrals a year in US ambulatory care; only about half are ever completed1
31 days
average new-patient wait across 15 major metros (2025); gastroenterology 40 days, Boston up to 652
40 to 70%
of referrals leak out of network;3 30 to 50 percent never complete at all1
206M
Epic In Basket system messages in 2024, up from 58M in 2017; the follow-up lands on primary care4
Nothing about the episode, whether the referral was needed or what happened to the patient, ever feeds back into the next referral decision.
Sources  1 Mehrotra et al., Milbank Quarterly, 2011 · 2 AMN Healthcare, 2025 · 3 Advisory Board, 2024 · 4 Journal of Arthroplasty, 2025 · Full citations: slide 14. Photo: AI-generated illustration.
2 / 14
The problem

Six steps, and it fails at every one.

STEP 1 · THE DECISION
PCP decides aloneMid-visit or in the In Basket, with no specialist input and no protocol at the moment it matters.
The intervention point everyone skips
STEP 2 · THE ORDER
Order plus free textThe referral goes out info-poor, with no standard workup attached and the question unstated.
Specialists start over on arrival
STEP 3 · THE QUEUE
Scheduling and authThe pile is sorted by scheduling and prior authorization, not clinical need.
31-day average wait; GI 40; Boston 652
STEP 4 · THE PATIENT
Navigates, or does notCalls, portals, travel, cost. Many quietly give up along the way.
30 to 50% never complete;1 40 to 70% leak3
STEP 5 · THE VISIT
Starts from scratchLow-acuity cases crowd out urgent ones; the visit is spent re-ordering the missing workup.
Wasted slots on both sides
STEP 6 · THE REPORT BACK
Maybe a letter returnsThe loop rarely closes. The next referral starts exactly as blind as the last one.
Nothing is learned, ever
Referral visits grew 159 percent from 1999 to 2009 (40.6M to 105M).5 The queue is structural, not cyclical, and every current fix starts AFTER step 1.
Sources  1 Mehrotra et al., 2011 · 2 AMN Healthcare, 2025 · 3 Advisory Board, 2024 · 5 Barnett et al., JAMA Internal Medicine, 2012 · Full citations: slide 14.
3 / 14
The problem

The hidden work, and why today's fixes stay bolt-ons.

The problem · the hidden work
58M → 206M
Epic In Basket system messages, 2017 to 2024. The referral system's dropped work lands here.4
~40%
higher odds of burnout with above-average In Basket load.6
50 to 98%
of interruptive decision-support alerts are overridden. Interruptive fixes fail.7
Today's fix · and why it falls short
The workaround: the eConsult

Ask a specialist before referring. It works clinically, in the US, UK, and Canada. But every US version is bolted on OUTSIDE the decision moment, with questions hand-drafted every time.

The clinical idea is right; the delivery is wrong.

What it costs today
  • Vendor networks: $150 per consult8 or $250 to $350 per provider per month9
  • The billing rail is thin: CPT 99446 to 99452 pays $19 to $76, chronically under-billed10
Why bolt-ons underperform
  • They start after the decision; the avoidable referral already exists.
  • They add a new destination for the busiest person in the building.4
  • They measure activity, not patient outcomes.
The gap is not another queue. It is judgment, at the decision moment, inside the tool already open.
Sources  4 Journal of Arthroplasty, 2025 · 6 Tai-Seale et al., Health Affairs, 2019 · 7 PMC systematic review, 2023 · 8 AristaMD · 9 RubiconMD · 10 CodingIntel · Full citations: slide 14.
4 / 14
The solution

Crossbeam: the specialist's judgment, at the decision moment, in Epic.

THE REFERRAL DECISION MOMENT. The PCP opens the panel by choice, inside Epic. No pop-ups, no hard stops, ever.
1OPEN

Mid-visit doubt: "does this need GI?" The PCP opens the Crossbeam panel by choice. Uninvited support gets overridden;7 invited support gets used.

2SEE

The matched protocol (P1): this system's own pathway, version-stamped, authors named. The chart context it used (P2), listed item by item.

3CHOOSE

Three real doors, below. The clinician picks; suggested orders are only ever pended for signature. The plain referral stays one click away.

4LEARN

What was chosen and what happened over 6 to 12 months, from claims and chart data (P4), feeds the next protocol version.

Step 3, expanded · the three doors
1 · MANAGE IN PRIMARY CAREprotocol says safe to treat or watch here; suggested orders pended for the PCP's signature
2 · SEND eCONSULTone click; question and context pre-assembled, routed on Epic In Basket rails to the specialist rota
3 · REFER, ENHANCEDthe specialty's preferred workup attached; a complete referral, the right queue, kept in network
Advisory only. The basis is always visible, the clinician always decides, and the full referral is always one click away.11
Sources  7 PMC systematic review, 2023 · 11 FDA, Clinical Decision Support Software guidance, 2022 · 25 Crossbeam venture record (product workflow), 2026 · Full citations: slide 14.
5 / 14
Product walk-through · 1 of 3 · illustrative concept design

The panel at the decision moment.

EHR SIDEBAR · ADVISORY PANEL · DEMO DATA
1
Opened by choiceNever interruptive. Decision support that fires uninvited gets overridden 50 to 98 percent of the time.7
2
Shows its workProtocol version, authors, and every data point used are on screen before anything is signed.
3
Three real optionsManage, eConsult, or refer enhanced. The plain referral stays one click away, always.
4
Defensible for the PCPFollowing your own institution's versioned protocol is a documented act, not a leap of faith.
Illustrative concept design with demo data and a fictional patient; not a screenshot of any EHR product. Sources  7 PMC systematic review, 2023 · Full citations: slide 14.
6 / 14
Product walk-through · 2 of 3 · illustrative concept design

The eConsult, pre-assembled; answered in minutes, not months.

1
Drafted, not dictatedQuestion and attachments assembled by software from the pathway and the chart; the PCP edits and sends.
2
Complete on arrivalThe specialist answers a real question in minutes instead of excavating a cold chart. LA County (US) runs a median 1-day reply at scale.12
3
Nothing lostReply, disposition, and protocol version are written back to the referral record and tracked against claims (P4).
eCONSULT COMPOSER · DEMO DATA
Illustrative concept design with demo data; fictional patient and clinicians. Sources  10 CodingIntel · 12 Barnett et al., Health Affairs, 2017 · Full citations: slide 14. Photo: AI-generated illustration.
7 / 14
The solution

Four pillars, and autonomy that is earned rung by rung.

P1 · EncodeSpecialist-authored, system-specific protocols, versioned and refreshed on a named cadence.
P2 · ContextualizeAuto-assembled chart context with the basis visible item by item.
P3 · EmbedAt the referral decision moment in Epic, PCP-initiated, never interruptive.
P4 · LearnOutcomes tracked on claims and chart data, feeding the next protocol version.
RungWhat the product may doUnlocked by
v1 · Advisory floorOn-demand panel, protocol match, auto-context, disposition menu with pended orders. The clinician signs everything.Day one. This is the permanent floor.
v2 · Auto-draftedeConsults and referral workups arrive fully drafted; the clinician reviews and signs.Measured accuracy and edit-rate thresholds on v1, reviewed by the clinical governance committee.
v3 · Narrow, earned auto-resolutionLowest-risk pathway steps only (repeat-lab scheduling per protocol), opt-in per authoring division, audit-trailed.Claims-verified avoidance at a pre-registered effect size AND a sub-1 percent adjudicated miss rate, re-reviewed with FDA counsel.
The question is never "do you trust AI." It is "what has this specific protocol earned, on evidence your own committee reviewed."11
Sources  11 FDA, Clinical Decision Support Software guidance, 2022 · 25 Crossbeam venture record (autonomy ladder and evidence gates), 2026 · Full citations: slide 14.
8 / 14
The market

Who pays: the buyer must keep the savings.

+$490K
per 1,000 deflected GI referrals for a fully capitated system (estimate, internal model)25
-$405K
for a pure fee-for-service system: deflection destroys procedure revenue (estimate, internal model)25

The honest framing: under fee-for-service, our success costs the buyer money. So Crossbeam sells where risk is real. 53.4 percent of Traditional Medicare lives are in accountable care (January 2025; CMS targets 100 percent by 2030),14 but only 28.7 percent of payments carried downside risk in 2024.13 We underwrite on buyer P&L value; no medical-specialty collaborative-care code exists through CY2027.16

Payment rail todayPaysReality
Interprofessional eConsult, CPT 99446 to 9945210$19 to $76 per consultOnce per 7 days, patchy coverage, chronically under-billed
Behavioral CoCM (99492 to 99494, G2214)15~$48 to $145 per patient per monthThe precedent: CMS built a rail AFTER the evidence. Behavioral only
Medical-specialty collaborative PMPM16Does not existNothing on the CMS calendar through CY2027
Risk-bearing system P&LThe real railPayer-provider systems, capitated groups, and MSSP savings keepers buy on kept savings
Proof the buyer exists: Baylor Scott & White kept $76.8M of the $104.5M it saved CMS in PY2024.17 That retained-savings budget line is the customer.
Sources  10 CodingIntel · 13 HCP-LAN, 2024 · 14 NIC, 2025 · 15 Providers Care Billing · 16 CMS, CY2027 PFS proposed rule · 17 Baylor Scott & White, 2025 · 25 internal model (inputs: 18, 19) · Full citations: slide 14.
9 / 14
The market

Market size: two honest builds, side by side.

Optimistic build directional · all figures USD · assumes premium pricing no buyer has yet paid25
TAM SAM SOM
TAM · near-term software + services
~$2B to $4B25, 20
US specialty-referral decision and eConsult software, bounded by the referral-management market
SAM · reachable via system GTM
~$1.0B to $1.5B25
Health systems sellable with a direct enterprise motion
SOM · 5-year ARR
~$25M to $75M25
At ~$4,500 per clinician per year, a price point no buyer anywhere has yet paid
Where these numbers come from: internal market build, July 2026,25 top-down from US clinician counts at premium per-seat pricing (~$4,500 USD per clinician per year), bounded by the US patient-referral-management software market (Grand View Research).20 Circles illustrative, not to scale.
Diligence rebuild estimate · all figures USD · at prices the category has actually paid25
TAM SAM SOM
Theoretical TAM
~$450M to $750M25
397 US health systems21 at realized eConsult-category price points
ICP-filtered SAM
~$15M to $80M25
~150 to 350 qualifying risk-bearing systems, the only buyers whose P&L improves
SOM · 5-year
~$3M to $15M25
At realized ~$100K to $400K per-system ACV
Where these numbers come from: internal diligence rebuild, July 2026,25 bottom-up from 397 US health systems (AHA)21 at realized eConsult-category prices (USD; vendor list pricing and public program economics).8, 9, 10 Circles illustrative, not to scale.
Both builds are estimates in USD, and the divergence IS the finding: the workflow is validated in the US, UK, and Canada, the premium software price on none. We either prove premium value with claims-verified evidence, or reprice into the per-provider anchor that already exists.13, 14
Sources  13 HCP-LAN, 2024 · 14 NIC, 2025 · 20 Grand View Research · 21 AHA, 2025 · 25 internal market build and diligence rebuild, July 2026 · Full citations: slide 14.
10 / 14
The opportunity

Everyone holds a piece. Nobody runs the loop.

Player (27 profiled; key rows)25P1 EncodeP2 ContextP3 At the momentP4 LearnTotal
Crossbeam target22228/8
Epic (platform)12216/8
NHS Advice & Guidance (UK program)11215/8
Oshi Health (GI provider, not software)11125/8
eConsult vendors (AristaMD class)11114/8
Ontario eConsult (Canada, provincial program)01113/8
Ambient AI (Ambience, Abridge)01102/8
P4
The empty columnAcross all 27 profiles, no horizontal software closes the outcome loop. The only 2 belongs to a provider group grading itself.
E
Epic is the real clockClosest at 6/8. The competitive question is speed to the loop, not the eConsult incumbents.
$
Why the center stayed emptyUnder fee-for-service, deflecting referrals costs the deflector money. The whitespace is an economics gap, not an ideas gap, and risk adoption is closing it.13
W
The wedgeOne specialty (GI), one institution's own protocols, one decision moment, outcomes measured from claims.
Pillar scores: internal whitespace matrix, July 2026 (0 absent, 1 partial, 2 shipped); labeled an evidence gap, not confirmed-empty space. Sources  13 HCP-LAN, 2024 · 25 Crossbeam venture record · Full citations: slide 14.
11 / 14
Evidence · product walk-through 3 of 3

What we must prove, and the console that keeps us honest.

51% (self-reported)
Ontario (Canada) PCPs reporting a contemplated referral avoided, on a mandatory closeout census of 60,474 eConsults22, 23
~6%
the category's only RCT, on claims data: a non-significant reduction in actual referrals (CI 0.85 to 1.03)24

Both numbers can be true; only one is bankable. Vendors quote the first kind. A buyer's actuary finds the second. So Crossbeam commits to claims-verified outcomes from day one, and the protocol console (right) is that commitment built into the product: every KPI is labeled by how it was measured, and self-report is demoted to secondary telemetry.

Three questions we are validating now: Is the referral decision moment where clinicians actually want help? Would your system pay, given its payer mix? Does advisory guidance change behavior enough to show up in claims?
PROTOCOL CONSOLE · CLAIMS-VERIFIED · DEMO DATA
Console: illustrative concept design, all numbers fictional. Sources  22, 23 Ontario eConsult program studies · 24 RCT, read as "not demonstrated on claims," never "proven zero" · Full citations: slide 14.
12 / 14

We are validating with clinicians now.
Tell us where this is wrong.

  • Where in your day would this panel genuinely help, and where would it annoy?
  • Would you trust pathways authored by your own system's specialists more than generic guidelines?
  • Who in your organization would sponsor and pay for this, honestly?
  • If referrals dropped 10 to 15 percent, who in your building would fight it?

DRAFT

We'd love your feedback

docs.google.com/forms/d/e/1FAIpQLSd4RZGojA5Xu4BtR1aX9DELviqutIoHa0PhGe2HYstL_UyHYA/viewform

Scan or click. About 5 minutes.
Blunt is better than polite; we are looking for the reasons this fails.

Confidential draft for clinician feedback, July 2026. "Crossbeam Health" is a draft working name, a concept in validation. Not an offering; no product claims are made. Photo: AI-generated illustration.
Appendix

References.

1. Mehrotra, Ateev, Christopher B. Forrest, and Caroline Y. Lin. "Dropping the Baton: Specialty Referrals in the United States." The Milbank Quarterly, 2011. pmc.ncbi.nlm.nih.gov/articles/PMC3160594/

2. AMN Healthcare. "2025 Survey of Physician Appointment Wait Times." AMN Healthcare Insights, 2025. amnhealthcare.com/amn-insights/physician/whitepapers/2025-survey-of-physician-appointment-wait-times/

3. Advisory Board. "How to Reduce Referral Leakage." Advisory.com, Dec. 2024. advisory.com/topics/physician/2024/12/reduce-referral-leakage

4. Analysis of Epic In Basket message volume growth, 2017 to 2024. The Journal of Arthroplasty, 2025. sciencedirect.com/science/article/abs/pii/S0883540325008204

5. Barnett, Michael L., et al. "Trends in Physician Referrals in the United States, 1999 to 2009." JAMA Internal Medicine, 2012. jamanetwork.com/journals/jamainternalmedicine/fullarticle/1108675

6. Tai-Seale, Ming, et al. "Physicians' Well-Being Linked to In-Basket Messages Generated by Algorithms in Electronic Health Records." Health Affairs, 2019. healthaffairs.org/doi/10.1377/hlthaff.2018.05509

7. Systematic review of override rates for interruptive clinical decision support alerts. PMC, 2023. pmc.ncbi.nlm.nih.gov/articles/PMC10491420/

8. AristaMD. "eConsult ROI Calculator." AristaMD.com. aristamd.com/specialty-care/care-delivery/econsults/econsult-calculator/

9. RubiconMD. Published per-provider pricing. RubiconMD.com. rubiconmd.com/buy-now

10. CodingIntel. "Interprofessional Internet Consultations (CPT 99446 to 99452)." CodingIntel.com. codingintel.com/interprofessional-internet-consultations/

11. US Food and Drug Administration. "Clinical Decision Support Software: Guidance for Industry and FDA Staff." FDA.gov, 2022. fda.gov/media/191560/download

12. Barnett, Michael L., et al. "Los Angeles Safety-Net Program eConsult System Was Rapidly Adopted and Decreased Wait Times to See Specialists." Health Affairs, 2017. healthaffairs.org/doi/10.1377/hlthaff.2016.1283

13. Health Care Payment Learning and Action Network. "2024 APM Measurement Effort." HCP-LAN.org, 2024. hcp-lan.org/2024-infographic/

14. NIC. "Progress Toward Value-Based Care in 2025." NIC.org, 2025. nic.org/blog/progress-toward-value-based-care-in-2025/

15. Providers Care Billing. "Complete Guide to CoCM Billing: 99492, 99493, 99494, G2214." ProvidersCareBilling.com. providerscarebilling.com/complete-guide-to-cocm-billing-99492-99493-99494-g2214/

16. Centers for Medicare & Medicaid Services. "Calendar Year (CY) 2027 Medicare Physician Fee Schedule Proposed Rule." CMS.gov Newsroom, 2026. cms.gov/newsroom/fact-sheets/calendar-year-cy-2027-medicare-physician-fee-schedule-proposed-rule

17. Baylor Scott & White Health. "Baylor Scott & White Quality Alliance Sustains Strong Performance in Quality Measures, Saving CMS $104.5 Million." BSWHealth Newsroom, 2025. news.bswhealth.com

18. Economic model input: specialty care cost analysis. PMC, 2024. ncbi.nlm.nih.gov/pmc/articles/PMC10872000/

19. Economic model input: eConsult program economic evaluation. PMC, 2023. pmc.ncbi.nlm.nih.gov/articles/PMC10724760/

20. Grand View Research. "US Patient Referral Management Software Market Report." GrandViewResearch.com. grandviewresearch.com/industry-analysis/us-patient-referral-management-software-market-report

21. American Hospital Association. "Fast Facts: US Health Systems 2025." AHA.org, June 2025. aha.org/infographics/2025-06-18-fast-facts-us-health-systems-2025-infographic

22. Mandatory closeout census of 60,474 Ontario eConsults: PCP-reported referral avoidance (self-reported). PMC, 2022. pmc.ncbi.nlm.nih.gov/articles/PMC9199054/

23. Ontario eConsult program evaluation. JMIR Formative Research, 2022. formative.jmir.org/2022/4/e32101

24. Randomized controlled trial of eConsult on claims-measured specialty referrals: non-significant ~6 percent reduction (CI 0.85 to 1.03). PMC, 2019. pmc.ncbi.nlm.nih.gov/articles/PMC6558850/

25. Crossbeam Health venture record (internal, unpublished working papers): market build and diligence rebuild (08_market_overview), economic deflection model, 27-entity whitespace matrix (09_deep_dives), product workflow and autonomy ladder (10_gtm). July 2026.

Superscript markers on slides 2 to 12 refer to these numbered sources (abbreviated MLA format; descriptive titles are used where a formal title is not restated). Internal analyses are labeled as estimates where they appear.
14 / 14