Luminai
B2B enterprise software sold to large provider organisations. THE CONTRACT FORM IS NOT DISCLOSED — no retrieved source states whether Luminai prices per workflow, per seat, as a subscription, on a share of recovered dollars, or on shared savings, and the company publishes no pricing. What is established is the BUDGET: administrative and operational labour inside a health system, funded out of collected revenue, framed by the company against roughly $1 trillion of annual US hospital operations spend and administrative activity at 'up to 25% of total healthcare spending' (Series B release). That is Waystar's payer class — paid from money the provider RECOVERS or labour it stops paying for, never from a fee schedule — and it is the most important structural fact here, because canon REG-CODE-001/002 records 26 CPT codes describing clinical AI, only 3 Category I (Jan 2026, Bipartisan Policy Center, secondary), against 1,524 FDA authorisations from FDA's own list (primary) — 508:1 on a mixed basis. No code pays for automating a referral, and none is needed. Architecture is three layers: unstructured-to-structured transformation of faxes, PDFs, charts and handwritten notes; a knowledge graph encoding that system's routing rules, policies, exceptions and institutional judgement; and agentic execution — classify, match patient and provider, route, trigger — escalating to a human whose correction feeds back as training signal.
Earnings, margins, COGS & capex
There are no financials, and that is the finding. Luminai discloses no revenue, ARR, margin, burn, runway, headcount or valuation at any round, and no customer count of its own — lead investor Peak XV supplies the only count in the record, 'dozens of provider institutions', unaudited and undated. That is a more complete absence than the private comparators canon tracks, where Abridge reported ~$100-117M of contracted ARR on a Q1-2025 basis (CM-PRIV-AB-04) and OpenEvidence a reported ~$300M annualised run-rate as of Jul 2026 (CM-PRIV-OE-03; trade press, no primary source). What it publishes is four operating claims and one logo: 12M+ automations, 7 average use cases per customer, a 48-day average time to value and 5.3x ROI. None is a financial quantity of the business and none has a denominator — the fourth is a quantified return claim, so the void is narrower than it looks, and no more testable. So the company is sized by comparison, not multiple. Waystar, in the same payer class, prints TTM revenue of $1.21B (+19.2%), 68-69% gross margin, an adjusted-EBITDA ladder of 40.65% FY2024 / 42.04% FY2025 / 42.76% Q2 2026 / 42.95% 1H 2026, NRR of 108.3% and 1,453 clients above $100k LTM (CF-WAY-05 to CF-WAY-16). The instructive rung is +140bp — the fiscal-year gain delivered in the year AI was the entire narrative, with no disclosure separating a dollar of AI revenue (CF-WAY-06). That is the prior an 80% automation claim must clear. The opposing read of the same ladder is Luminai's best argument: volume-based revenue grew +2.8% in Q2 2026 against +34.5% for subscription (CF-WAY-10) — a commoditising transaction layer, with value migrating up to the workflow-execution layer Luminai occupies.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~32¢ is cost of goods and ~20¢ operating expense, leaving ~48¢ of operating profit.
Revenue trend
Margins
Below Waystar's 68-69% and below pure SaaS. Falling per-document inference cost is the tailwind; a human escalation layer that scales with volume until automation rates asymptote is the headwind. Whose payroll carries the human decides it.
The comparison is Waystar's four-rung ladder, 40.65% to 42.95%, each on its own stated basis (CF-WAY-05 to CF-WAY-08). A Series B company in the same payer class is several ladder-lengths away — the honest statement of stage difference.
Waystar's Q2 2026 GAAP net margin was 12.78% on $40.867M (CF-WAY-11), reached eight years and ten acquisitions into a consolidation strategy — the timeline an entrant is implicitly underwriting.
Even the profitable comparator's conversion is unstable: unlevered FCF margin 20.0% of revenue in Q2 2026 against 40.9% a year earlier (CF-WAY-14; canon attributes the halving to NOLs burning off — an analytical reading, not an issuer statement).
The metric that would matter most, entirely absent. Waystar reports 108.3% with 1,453 clients above $100k LTM, +15% YoY (CF-WAY-12). For a land-and-expand platform whose thesis is that one system buys workflow after workflow, NRR IS the thesis — and it is unpublished.
COGS structure
Undisclosed; reconstructable only from architecture. Four lines. (1) DOCUMENT INFERENCE — OCR and vision over faxes, scans, charts and handwriting, the hardest modality in health IT and the deliberate entry point because incumbents' structured pipelines fail there; metered per page, falling with model prices. (2) AGENTIC REASONING — LLM calls per workflow decision, and a multi-step agentic workflow consumes several times the tokens of a single classification. (3) HUMAN-IN-THE-LOOP ESCALATION — the explicit design point, and the line separating a software gross margin from a services one; at '>80% automation for certain document types' the residual is up to one in five, and the qualifier concedes the average is lower. (4) DEPLOYMENT AND KNOWLEDGE-GRAPH BUILD — 48 days is fast, but 48 days of engineering-plus-operator attention per logo is a real per-customer cost until genuinely productised. The shape rhymes with the comparator: Waystar's 68-69% sits below software peers partly BECAUSE its patient-payments line carries interchange and processing cost. In both, the sub-SaaS margin is structural, not operational.
Capex
Minimal capitalised capex — no owned compute, no data centre; inference and hosting are cloud opex. The genuine investment is expensed: engineering, deployment headcount and the per-customer knowledge graph. The contrast with the comparator is instructive rather than flattering to either: Waystar's balance sheet is 89.3% goodwill and intangibles ($4,014.8M plus $1,223.9M against $5,868.8M of assets, CF-WAY-15) from ten acquisitions since 2018 — most recently Iodine Software at $1.25B, closed 1 Oct 2025 (CF-WAY-17) — with R&D at 5.0% of revenue. The incumbent BUYS capability and carries it as purchase accounting; the entrant BUILDS it and expenses it, which is why capex intensity is near zero on the balance sheet and high in the P&L.
Latest earnings
n/a (private) — no guidance to beat or miss. The comparator's most recent print is dated and specific: Waystar reported Q2 2026 on 29 Jul 2026 with revenue of $319.7M, +18% as released (+18.1% at the 10-Q decimal), adjusted EBITDA of $136.725M and GAAP net income of $40.867M (CF-WAY-09/07/11).
None — no financial guidance and no forward operating target. The only forward statements are the use of proceeds. Waystar, by contrast, raised FY2026 guidance at Q2 to $1.276B-$1.294B revenue and $535M-$545M adjusted EBITDA (CF-WAY-13, ~42.0% at the midpoints).
- Workflow automations
- 12 million+ workflow automations supported — company-stated, on luminai.com and in the company's own 9 Apr 2026 announcement post; it does not appear in the PR Newswire release. No period, no per-customer split, no denominator, and the '+' is the company's own floor.
- Average time to value
- 48 days average time to value — company-stated, quoted by HIT Consultant 9 Apr 2026; 'value' is the company's own undisclosed definition.
- Automation rate (early deployments)
- >80% automation for certain document types in early deployments — Forbes, 9 Apr 2026, attributing the company. 'Certain document types' is the load-bearing qualifier.
- ROI
- 5.3x ROI — published on luminai.com's homepage metric strip. Company-stated and unaudited, with no denominator, no period, no customer sample and no methodology; the company does not say 5.3x of what, measured how, at how many customers. It is the only quantified economic claim the company makes, and as published it cannot be tested.
- Use cases per customer
- 7 average use cases per customer — luminai.com homepage, company-stated. An average implies both a customer population and per-customer expansion, which is the motion the platform thesis requires; the customer count that is its denominator is unpublished, and lead investor Peak XV's unaudited 'dozens of provider institutions' is the only count in the record.
- Series B
- $38M Series B, 9 Apr 2026, led by Peak XV Partners; Define Ventures new; General Catalyst and Y Combinator returning; Scribble Ventures is the only further participant named in the round, and it appears in the company's own post rather than in the release. Moxxie Ventures, Craft Ventures, Underscore VC and Unshackled Ventures appear in the release's standing 'backed by' list as backers of the company, not as participants in this round.
- Total raised
- $60M total capital raised, company-stated in the 9 Apr 2026 Series B release. The announced rounds sum to $57.4M — ~$3.4M seed (itself an aggregate of a $3M round in Aug 2020, a $250k Unshackled Ventures cheque in Mar 2020 and Y Combinator's $150k in May 2020), $16M Series A (9 Jun 2022, General Catalyst lead, TechCrunch; announced together with the rename from DigitalBrain), $38M Series B (9 Apr 2026) — leaving ~$2.6M unaccounted for. The two live candidates are an unannounced 2022-2026 bridge and rounding in the company's own headline; TechCrunch's 'almost $20 million' in Jun 2022 is consistent with $3.4M + $16M, which makes a large undisclosed early round unlikely. No source resolves it.
- Valuation
- No post-money disclosed at any round — checked across the primary release, Forbes, MedCity News, HIT Consultant and MobiHealthNews. An absence in the record, not an omission here.
- Named customers
- One named customer: Cleveland Clinic (23 hospitals, ~300 outpatient facilities, nearly 16M patient encounters against ~3.6M unique patients, and $18.3B of operating revenue at a 5.0% operating margin in 2025, per its own Facts + Figures), announced 9 Apr 2026 — AI classifying and routing complex faxed referrals across thousands of possible destinations. No second customer is named in any source; lead investor Peak XV states Luminai is 'already working with dozens of provider institutions', an unaudited investor characterisation rather than a disclosed count.
- Advisers
- Advisers, company-listed: Bob McGrew (former OpenAI CRO), Kevin Weil (former OpenAI CPO), Toby Cosgrove (former Cleveland Clinic CEO), Bruce Broussard (former Humana CEO), Frank Williams (Evolent Health co-founder). Team provenance: Palantir, Cruise, Google, Coinbase, Brex, Epic, Banner Health.
- Founded
- Founded 2019 per Y Combinator's own directory — the founders met in May 2019 and DigitalBrain took pre-batch cheques in March and May 2020; Y Combinator Summer 2020 batch, originally DigitalBrain; San Francisco. Co-founders Kesava Kirupa Dinakaran (CEO) and Dmitry Dolgopolov; Dinakaran was named to Forbes 30 Under 30, Enterprise Technology, 2023, as a Luminai co-founder while the company was still the general-purpose workflow business.
Growth drivers
- The administrative cost base — roughly $1 trillion of annual US hospital operations spend and administrative activity at 'up to 25% of total healthcare spending' (company-stated market context, not an independent estimate). Budget that size needs no new appropriation, only a labour line to shrink.
- Hospital labour cost and staffing constraint — the buyer's motivation is a wage line, not a coverage decision — and it does not wait on a code (REG-CODE-001/002, 508:1 FDA authorisations to Category I codes — the code count is Bipartisan Policy Center, secondary).
- CMS-0057-F clocks — 7 calendar days standard and 72 hours expedited for prior-auth decisions from 1 Jan 2026, with Patient Access, Provider Access, Payer-to-Payer and Prior Authorization API compliance due 1 Jan 2027 across MA, state Medicaid and CHIP FFS, Medicaid and CHIP managed care and FFE QHP issuers (REG-PA-001, primary, 89 FR 8758). Faster payer clocks raise the value of a first-pass-complete packet. It cuts both ways and reappears in threats.
- CMS-0062-P extending electronic prior authorisation to DRUGS — published 14 Apr 2026, comments closed 15 Jun 2026 (REG-PA-003, primary, 91 FR 19890) — a second workflow surface on a known clock, subject to canon's caveat that this is payer-side plumbing, not a provider data-access mandate.
- The Cleveland Clinic reference as distribution — 23 hospitals, ~300 outpatient facilities and nearly 16M patient encounters against ~3.6M unique patients in 2025, with a former Cleveland Clinic CEO advising. In health IT the marquee academic-centre logo is the channel, and referral management is a workflow every peer system runs.
- Land-and-expand from a single integration — referral intake, then prior auth, claims, eligibility, forms, billing exceptions. Once the knowledge graph exists each further workflow should cost less than the first: the only mechanism by which this business reaches a software gross margin.
Bull & bear
Luminai attacks the one part of healthcare where the waste is unambiguous, from the one direction incumbents cannot copy. Administrative activity absorbs up to a quarter of US healthcare spending; none of it is reimbursed, all of it is labour, and the buyer can measure the saving in its own general ledger without waiting for a CPT code that — at 508:1 — is not coming. The incumbent stack moves structured EDI transactions; the work that actually jams a health system is a faxed referral with a handwritten note that must reach one of thousands of destinations under routing rules nobody wrote down. Luminai encodes those rules per system and executes against them, and the evidence it is a product rather than a demo is that Cleveland Clinic put it into referral management and that the company can publish a 48-day average time to value. Its lead investor also describes dozens of provider institutions, though that is a promotional characterisation by the firm that led the round, not a disclosure. The land-and-expand step has a claim behind it: the company publishes an average of 7 use cases per customer, the exact shape the thesis requires, and a 5.3x ROI as the return that buys the next one. Both are unaudited seller-side claims and neither should be banked. But if the second workflow at each customer really does cost less than the first, this becomes a multi-budget-line platform inside the largest cost pool in American healthcare.
- No reimbursement dependency in a sector defined by it. Every imaging- and diagnostic-AI vendor is fighting for a code that a 508:1 authorisation-to-code ratio (code count per the Bipartisan Policy Center, secondary) says is not coming; Luminai sells against a payroll line that already exists and is already under pressure.
- The entry point is the incumbents' structural blind spot. Waystar's network is built on transactions — roughly 7.5 billion a year, company-stated — and that layer is precisely what is losing pricing power (+2.8% volume-based against +34.5% subscription, CF-WAY-10). Margin is migrating to whoever executes the workflow, not whoever routes the message.
- The flagship logo IS the distribution strategy and it is already secured; its lead investor says the book is wider — dozens of provider institutions — which is an investor's word, not a count. Cleveland Clinic unlocks the peer academic-centre market, referral management is common to all of them, and a former Cleveland Clinic CEO sits alongside a former Humana CEO and Evolent's co-founder on the advisory bench.
- 48 days to value is genuinely differentiating in a market conditioned by multi-year implementations, and with a published 5.3x ROI beside it the company answers both enterprise objections at once — how long until it works, and what it returns. Averages across several metrics imply a population of completed deployments and at least partial productisation. Neither figure is audited and the ROI carries no sample, no method and no statement of what it measures, so this is a sales-cycle advantage rather than a proven one.
- Regulation delivers mandated workflow on a published clock: CMS-0057-F's decision clocks from 1 Jan 2026 with APIs by 1 Jan 2027 (REG-PA-001) and CMS-0062-P extending electronic PA to drugs (REG-PA-003). Forced re-tooling moments are when incumbents lose accounts.
- The cost curve runs for the product and against the alternative — document inference gets cheaper each generation, the wage of the person doing the work does not. Automation rate and gross margin should improve passively, and today's COGS drag is a proprietary correction-data flywheel later entrants will not have.
This is a company with no disclosed price, no disclosed revenue, one named customer and a company-stated $60M raised over six years — $57.4M of it traceable to announced rounds — competing for a budget line against rivals that have each raised more in a single round. What it says about its own economics cannot be checked, and the one number in its 9 Apr 2026 announcement that can be checked is wrong: it put its only named customer's patient base at over 15 million where that customer's own published figures give ~3.6 million unique patients, an overstatement of roughly four-fold. Meanwhile the best available evidence says the automation premium it sells has not shown up anywhere it could be measured. Waystar, the listed name in the identical payer class, moved adjusted-EBITDA margin +140bp from FY2024 to FY2025 in the year AI was the entire narrative, and separates no AI revenue at all. PHTI's verdict on the adjacent category is that health-system leaders report gaps in the evidence on financial ROI. The two forces that decide the outcome are outside Luminai's control: Epic, at 43.7% of US acute-care hospitals and gaining, can ship the same workflow inside a contract it already owns — including at Cleveland Clinic — and CMS-0057-F's 1 Jan 2027 APIs standardise the very integration difficulty the product monetises.
- There is no denominator, and unusually so — no revenue, ARR, margin, burn, runway, headcount or valuation at any round, less than Abridge or OpenEvidence disclose. The void is not total: the company publishes a 5.3x ROI and an average of 7 use cases per customer, and Peak XV publishes 'dozens of provider institutions'. None of the three carries a sample, a period, a method or an auditor, and the company does not say 5.3x of what. The finding is therefore not that nothing is claimed but that what is claimed cannot be falsified — and canon CM-PHTI-02 records health-system leaders already discounting financial-ROI claims in the adjacent category.
- The incumbent's margin ladder is the sharpest evidence against the thesis: 40.65% FY2024 to 42.04% FY2025, +140bp with no AI revenue separated (CF-WAY-05/06) — canon records this as the figure that demoted the 'AI RCM' framing. Cash conversion moved the wrong way over the same span: unlevered FCF margin 20.0% of revenue in Q2 2026 against 40.9% a year earlier (CF-WAY-14).
- Epic is the falsifier and there is no answer. 43.7% share and a fifth straight year of gains (CM-KLAS-01) while Oracle Health lost 56 hospitals and 14,676 beds as the largest net share loser for a third year (CM-KLAS-03). Epic ships adjacent workflow at zero acquisition cost inside a contract it already holds — at Luminai's own flagship account.
- The regulatory tailwind is also the moat's demolition schedule. The four CMS-0057-F APIs are due 1 Jan 2027 (REG-PA-001); to that exact extent payer-connectivity know-how stops being a product. And 42 CFR 422.566(d) already requires a qualified human to review any adverse determination before it issues (REG-PA-004), capping the end-to-end cycle time a provider-side vendor can promise.
- Weakest capital position in its own set: $60M lifetime against Candid Health's $120M Series D alone, Cohere Health's ~$200M total, Adonis's ~$95M, and Commure's $70M May 2026 raise on a $7B mark (CM-PRIV-CM-01). In a category with long sales cycles where the reference customer is everything, being the least-capitalised credible vendor compounds against you.
- The proof points degrade under scrutiny and the buying evidence is unsupportive. Canon CM-PHTI-02 records leaders reporting evidence gaps on financial ROI in the adjacent AI category, and CM-BENCH-07 records that benchmarks have minimal impact on health-system purchasing. A 5.3x ROI published with no sample and no definition is precisely the proof point this buyer has learned to discount. In a market that buys on reference and politics rather than measured performance, the least-funded vendor with one nameable reference is not the natural winner.
What it is worth
NONE APPLICABLE — the finding, not a gap. Luminai has no exchange price, no audited financials, no disclosed revenue and, uniquely among the private names in the QAI healthtech canon, NO DISCLOSED POST-MONEY AT ANY ROUND. All three approaches fail explicitly. (1) LAST-ROUND MARK — unavailable; the $38M Series B disclosed a round size and a syndicate and no price, checked across the company release, Forbes, MedCity News, HIT Consultant and MobiHealthNews. (2) IMPLIED FROM THE RAISE — REFUSED as a matter of method: $38M is consistent with post-moneys from roughly $150M to $600M on dilution alone, and canon's record of that arithmetic going wrong is on file (Sword Health's $4B mark set by a $40M round, 13.5 months stale, CM-COH-05). (3) REVENUE MULTIPLE — impossible; no revenue figure exists, public or reported. The company's published 5.3x is not a revenue multiple and not a company financial, and since the company does not define it — it does not say 5.3x of what — it cannot be an input to any of the three approaches and is not used as one here. What CAN be said is the payer-class comparison, and it is the honest substitute: Waystar trades at $4.05B on $1.21B of TTM revenue — roughly 3.3x TTM revenue (derived: 4.05 / 1.21) — at 68-69% gross margin, ~42% adjusted-EBITDA margin and 108.3% NRR. That is what the mature, profitable, listed version of this payer class is worth per dollar of revenue at 7 August 2026, and it is a low multiple. Any private mark for an entrant in the same class implies either far faster growth than Waystar's +19.2% or a belief that the listed comparator is mispriced — and the burden sits with whoever asserts the mark.
The wedge closes from three directions at once. Epic ships referral routing and authorisation workflow in-suite at zero acquisition cost into 43.7% of US acute-care hospitals and rising (CM-KLAS-01) — including Cleveland Clinic, which runs Epic. The CMS-0057-F APIs land on 1 Jan 2027 (REG-PA-001) and standardise the payer-connectivity difficulty a workflow vendor monetises, while 42 CFR 422.566(d) keeps a human in the payer's denial path (REG-PA-004) so the end-to-end cycle-time promise never fully materialises. And the buyer's scepticism holds: the listed comparator's +140bp of fiscal-year margin gain in the year AI was the entire narrative (CF-WAY-06) is the evidence a CFO actually has, alongside PHTI's finding of unproven financial ROI in the adjacent category (CM-PHTI-02). In that world $60M of lifetime capital against rivals raising $120M in a single round is decisive, the flagship stays at one workflow, the 5.3x ROI and the 7-use-case average are never defined or substantiated with a sample, and the outcome is an acqui-hire or a modest strategic sale — plausibly to the very incumbent this dossier compares it against, at a price no one will disclose either.
Luminai stays a well-regarded, capital-constrained Series B company with a marquee reference and a small book of enterprise accounts — growing, but not fast enough to force a disclosure. It adds workflows at Cleveland Clinic and lands a handful of peer systems on that reference; deployment headcount grows roughly with logos, so gross margin stays a software/services blend; the 1 Jan 2027 API date arrives and helps about as much as it commoditises. No valuation is disclosed at 7 August 2026 or the next, no revenue is published, and the company raises again in 2027-2028 at a mark that is a negotiation rather than a verdict. Nothing about the base case is bad — it is simply the case in which no external party can distinguish this company from the several rivals that have each raised more. An undefined homepage 5.3x does not close that gap: Adonis publishes a growth rate and a net-retention figure, which are testable in a way an undefined figure is not. What binds any valuation judgement here is unverifiability — the company says things about its economics, and none of them can be checked.
Luminai becomes the operations platform inside a meaningful set of large systems: Cleveland Clinic expands from referral management into prior auth, claims and eligibility; the knowledge-graph encoding proves genuinely reusable so the second and third workflow land at a fraction of the first's cost; the escalation rate falls far enough that gross margin resolves toward software; and the company starts disclosing an ARR figure and a customer count of its own — itself the signal, because companies that have those numbers publish them rather than an undefined homepage ROI figure. In that world it raises a large priced round well above anything $60M of lifetime capital would suggest, and the likeliest terminal owner is a strategic: Waystar has bought capability ten times since 2018 and carries 89.3% of assets as goodwill and intangibles, Iodine at $1.25B the most recent. The value driver is not the automation rate — it is net revenue retention above the 108.3% the listed comparator prints, because that is the only number that proves the platform claim.
Three things travel with any figure quoted here. FIRST, a negotiated round price is not a clearing price even when one exists — it carries liquidation preferences, ratchets and pro-rata rights a public price does not, and it re-prices only when someone chooses to raise. Here there is not even that, so nothing should be quoted as Luminai's valuation. SECOND, the capital-raised figures are dated but do not reconcile: the company states $60M total while the announced rounds sum to $57.4M, leaving ~$2.6M unaccounted for — ~$3.4M seed, $16M Series A (9 Jun 2022, TechCrunch, General Catalyst lead; announced as Luminai, with DigitalBrain given as the former name), $38M Series B (9 Apr 2026, Peak XV). Capital raised is not value created and must never be a valuation proxy. THIRD, the operating metrics that would ordinarily substitute for financials do not carry the weight: no period or denominator on the automation count, a company definition behind '48 days to value', an automation rate scoped to certain document types. Reference for the disclosed private cohort: six last-round marks dated Jun 2025 to May 2026 totalling ~$33.05B (CM-COH-04) against a 17-name listed pure-play cohort of ~$33.90B of market cap at the 31 Jul 2026 close (CM-COH-01). Luminai belongs to neither. Not investment advice; there is no security here to buy.
SWOT
Strengths
- The right payer class — sold against administrative labour and recovered dollars, not a reimbursement code. Canon counts 26 CPT codes for clinical AI, 3 Category I (Jan 2026, Bipartisan Policy Center, secondary), against 1,524 FDA authorisations from FDA's own list (primary) — 508:1 on a mixed basis. Every clinical-AI vendor waiting on a code is waiting on nothing; Luminai never needs one.
- The hardest input modality as the deliberate wedge. Faxes, scans and handwriting are where every structured-data pipeline in health IT fails and where an EDI-native clearinghouse has no answer — a strategic choice, not a limitation.
- A flagship reference that functions as distribution — Cleveland Clinic, 23 hospitals and nearly 16M patient encounters a year against ~3.6M unique patients, with a former Cleveland Clinic CEO advising. Referral management is a workflow every peer academic centre runs, so the reference is transferable rather than bespoke.
- Time to value is the right axis, and 48 days is a credible number to publish into a market conditioned by multi-year implementations. Publishing an AVERAGE also implies the knowledge-graph encoding is at least partly productised.
- Capital and cap table after April 2026 — $38M fresh against $19.4M raised in the prior six years — the documented pre-Series-B total, corroborated by TechCrunch's 'almost $20 million' in Jun 2022, with Peak XV leading on a platform-consolidation thesis, General Catalyst returning, healthcare-dedicated Define Ventures entering, and an advisory bench spanning frontier-AI leadership and health-system CEO experience.
Weaknesses
- Total financial opacity — no revenue, ARR, margin, burn, runway, headcount or valuation at any round, and no company-stated customer count behind lead investor Peak XV's unaudited 'dozens of provider institutions'; more opaque than Abridge (~$100-117M of contracted ARR on a Q1-2025 basis) or OpenEvidence (a reported ~$300M annualised run-rate as of Jul 2026, no primary source). No denominator exists against which any claim can be tested.
- One named customer, and no company-stated count behind it. Peak XV describes 'dozens of provider institutions', an unaudited seller-side characterisation, which leaves a single flagship as the only reference that can be checked. If Cleveland Clinic does not expand past referral management, that reference is what the commercial narrative loses.
- The published metrics do not survive close reading — no period or denominator on '12 million automations', a company definition of 'value' behind the 48 days, and an automation rate scoped to 'certain document types'.
- Human-in-the-loop is architecturally load-bearing, putting a services cost inside the product. Until the escalation rate falls, gross margin is a software/labour blend — and neither the rate nor whose payroll carries the human is disclosed.
- Nearly four years between the Series A (Jun 2022) and Series B (Apr 2026), spanning a full pivot out of general-purpose operations automation — through the most permissive funding window healthtech has had. Revealed preference, and not favourable.
Opportunities
- Expand from referral intake into prior auth, claims, eligibility and forms at the same customer — the motion the company already claims to be running, at an average 7 use cases per customer. If the second workflow costs less than the first, NRR here can exceed Waystar's 108.3% (CF-WAY-12), because each workflow is new spend rather than a price increase. The 7 is unaudited and its denominator unpublished, so it raises the prior on this opportunity without settling it.
- The commoditising transaction layer — Waystar's volume-based revenue grew +2.8% in Q2 2026 against +34.5% for subscription (CF-WAY-10) — per-transaction routing losing pricing power while decision and workflow software gains it.
- The 1 Jan 2027 CMS-0057-F API date forces every provider-side vendor to re-tool payer connectivity at once, and an entrant with no legacy EDI estate re-tools far cheaper than an incumbent with thousands of certified connections to migrate.
- Drug prior authorisation as a second regulated surface (CMS-0062-P, REG-PA-003) — a newly mandated workflow on a known clock is the cheapest market entry a workflow platform can get.
- Platform consolidation as the buyer's stated preference — Peak XV's thesis is that large systems want one platform instead of hundreds of vendors. If it holds, the winner takes several budget lines at once; if not, Luminai is one of the hundreds.
Threats
- Epic. 43.7% of US acute-care hospitals in 2025, a fifth consecutive year of gains (CM-KLAS-01), roughly $6.7B of 2025 revenue (CM-EPIC-01, company-stated to a trade outlet and unauditable), and a marketplace toll of roughly $1,700-1,900 per vendor per year (CM-EPIC-02). Epic can ship referral routing and authorisation workflow inside a contract, BAA and security review it already owns, at zero acquisition cost. Cleveland Clinic runs Epic, so the dependency is concrete — and it is unhedgeable for a private company.
- The incumbent's own margin evidence says the automation premium has not appeared: +140bp from FY2024 to FY2025 (CF-WAY-05/06) in the year AI was the whole narrative, with no AI revenue separated in any disclosure. If AI were re-basing administrative cost, the company running the transactions would show it first.
- CMS-0057-F standardisation is two-sided — the same APIs due 1 Jan 2027 that make provider-side automation easier also commoditise the integration labour a workflow vendor monetises. Once a payer connection is a documented FHIR endpoint, 'we know how to talk to every payer' stops being a product — on a published schedule.
- Full-loop prior-auth automation is statutorily capped on the counterparty's side: 42 CFR 422.566(d) requires an adverse organisation determination to be reviewed by a physician or other appropriate professional with relevant expertise BEFORE the MA organisation issues it (REG-PA-004, verbatim from eCFR). Perfect provider-side submission still waits on a human at the payer.
- A crowded field where rivals disclose more and have raised more — as reported in trade coverage of company announcements: Candid Health $120M Series D (Jul 2026, Sixth Street Growth, described as ~3x its Feb 2025 price); Cohere Health $90M Series C led by Temasek to ~$200M total; Adonis $40M Series C (Mar 2026, Quadrille Capital, ~$95M total, company-claimed 4x+ 2025 revenue growth and >130% net retention); Commure — $7B mark (CM-PRIV-CM-01) — raising $70M in May 2026. Luminai's $60M lifetime is the smallest number in that set.
Moats, dependencies & bottlenecks
Moats
the encoded routing rules, policies, exceptions and institutional judgement of one specific health system Medium-strong where deployed The most defensible asset, because it is the part written down nowhere else: the CEO's own example is that the system infers where a referral should go even when the routing decision was never documented. Once in production it is expensive and risky to rebuild. Two limits — it is per-customer, so it confers nothing at the next sale, and its value depends entirely on how much of the encoding is reusable, the undisclosed variable that decides the business.
Human-in-the-loop correction data as a compounding training signal Every escalation returns a labelled decision from a trained operator in the customer's own context — supervision that cannot be bought. Not proprietary in kind: every serious workflow vendor runs the same loop, and the value decays if frontier models get good enough at the underlying document reasoning to make the correction set redundant.
Medium-strong once past pilot High while the workflow runs Referral intake and authorisation sit on the revenue path; a system routing live referrals through a vendor for a year does not casually swap it, because the failure mode is lost patients and lost revenue. Same property that gives Waystar 108.3% NRR — but Waystar has 1,453 clients above $100k LTM to prove it and Luminai has one named logo.
reputational, not structural Cleveland Clinic plus Cosgrove, Broussard, Williams, McGrew and Weil is an unusually strong bench for a Series B, and in health IT the reference is the go-to-market. But it is rented credibility: it opens doors rather than closing deals, any competitor landing a peer system matches it, and an adviser list is not a contract.
Unstructured-document competence at the fax-and-handwriting layer weakening Honestly read, a temporary advantage rather than a moat. Choosing the hardest modality was smart because EDI-native pipelines cannot touch it, but document understanding is the capability improving fastest in general frontier models. What survives the catch-up is not reading the fax — it is knowing what to DO with it, which is the knowledge graph, not this.
Dependencies
discretionary opex funded from collected revenue, with no reimbursement backstop Waystar's payer class, and the reason the two compare directly: paid from money the provider recovers or labour it stops paying for, never a fee schedule. The advantage is that no coverage decision gates the sale; the exposure is that hospital operating margin decides whether the sale happens at all, and canon CM-PHTI-02 records leaders reporting evidence gaps on financial ROI for adjacent AI purchases.
Commercial concentration One named customer carries the whole external narrative. If the referral deployment does not expand into a second and third workflow there is no public evidence of land-and-expand — and land-and-expand is the thesis. Unlike a listed comparator there is no client count to reassure against it (Waystar prints 1,453 above $100k LTM).
Technical dependency that is simultaneously the primary competitor The defining asymmetry of provider-side health IT. Luminai must read from and write to the systems of record; Epic holds 43.7% of US acute-care hospitals with a fifth straight year of gains, TruBridge 7.6%, Altera 2.9% (CM-KLAS-01/02). Marketplace access costs roughly $1,700-1,900 per vendor per year (CM-EPIC-02) — trivial in dollars, decisive in permission. Cleveland Clinic is an Epic system.
COGS and capability Document vision, OCR and agentic reasoning are bought, not built. Falling token prices are a direct gross-margin tailwind. Two-sided risk: a model provider moving up into healthcare workflow, and the fact that any capability arriving free in the base model stops differentiating everyone at once.
Regulatory and counterparty Prior-auth value depends on payer behaviour, and payers are on a federal clock: decision clocks from 1 Jan 2026, APIs by 1 Jan 2027 (REG-PA-001), drug ePA proposed 14 Apr 2026 (REG-PA-003). Standardisation raises the automation ceiling and lowers the price of reaching it. And 42 CFR 422.566(d) keeps a qualified human in the payer's denial path (REG-PA-004).
Advantages
- Sells into a budget that already exists and is already under pressure — administrative labour, framed against roughly $1 trillion of annual US hospital operations spend — with no coverage decision, no CPT code and no payer negotiation between the pitch and the purchase order.
- Attacks the unstructured surface (faxes, scans, handwriting) where the EDI-native incumbent stack has no native capability: a real wedge rather than feature parity against a clearinghouse.
- A measurable, publishable implementation speed (48 days average time to value) plus an explicit escalation design that gives a risk-averse buyer an answer for the cases the agent gets wrong.
- Cleveland Clinic as first reference — 23 hospitals, ~300 outpatient facilities, nearly 16M patient encounters against ~3.6M unique patients in 2025 — with a former Cleveland Clinic CEO advising: the combination that opens peer academic medical centres.
- A cap table pairing a healthcare-dedicated fund (Define Ventures), a returning Series A lead with deep provider relationships (General Catalyst) and a new lead with an explicit platform-consolidation thesis (Peak XV) — capital aligned to multi-workflow land-and-expand.
- Correct exposure to the AI cost curve — the alternative to the product is a wage that does not fall, while the product's own COGS falls with each generation of cheaper document inference.
Weaknesses
- Zero financial disclosure — no revenue, ARR, gross margin, burn, runway or headcount, no post-money at any of the three rounds, and no company-stated customer count; the only count in the record is lead investor Peak XV's unaudited 'dozens of provider institutions'. Nothing about commercial traction is testable.
- One named customer, against a comparator disclosing 1,453 clients above $100k of LTM revenue (CF-WAY-12).
- The least capital in its competitive set — $60M lifetime against Candid Health's $120M single round, Cohere Health's ~$200M total, Adonis's ~$95M, and Commure's $70M May 2026 raise on a $7B mark.
- Nearly four years between priced rounds (Jun 2022 to Apr 2026) spanning a full pivot out of general-purpose operations automation — through the most permissive funding environment healthtech has seen.
- Metrics that do not survive close reading — no period on the automation count, a company-defined threshold for 'value', an automation rate scoped to 'certain document types'.
- Human-in-the-loop is architecturally required rather than transitional in the current design, keeping labour inside gross margin and making the margin trajectory depend on an escalation rate nobody outside can see.
Bottlenecks
- No published unit economics of any kind. The company's 5.3x is not a company unit economic, and since the company does not define it — it does not say 5.3x of what, measured how, over what sample — it cannot substitute for one; without revenue, ARR, gross margin or an escalation rate, neither efficiency nor the direction of unit economics is assessable from outside — the binding constraint on every other judgement here.
- The human-in-the-loop layer sits inside COGS. At '>80% automation for certain document types', up to one in five items needs a person and the platform-wide residual is higher; until that falls, gross margin is a software/services blend.
- Per-customer knowledge-graph encoding may not amortise. 48 days of engineering-plus-operator attention per logo is a real cost that only disappears if the encoding is genuinely reusable — the undisclosed number that decides the business.
- Enterprise sales cycles plus the security, BAA, privacy and IT-governance review preceding any production deployment — a structural throughput limit on a company with $60M raised against rivals with two to three times the capital for deployment headcount.
- One reference account that can be named. In a market that buys on peer reference — canon CM-BENCH-07 records benchmarks having minimal impact on health-system purchasing — a single logo caps the funnel, not just the concentration risk.
- Statutory human review on the payer side of prior authorisation (REG-PA-004) caps the end-to-end automation that can honestly be promised on the workflow that most needs it.
Top signals & trends
Top signals
Verified at the company-issued release and corroborated by Forbes, MedCity News and HIT Consultant the same day. A new institutional lead plus a healthcare-dedicated co-investor is the credible form of a Series B. NO POST-MONEY WAS DISCLOSED and none is inferred — a raise size is not a valuation.
Company-announced, reported by Forbes, MedCity News and HIT Consultant. The description — classifying and routing complex faxed referrals across thousands of possible destinations — reads as a live deployment rather than an MOU. No contract value or term disclosed. One figure in the announcement is wrong at source, and the source is Luminai itself: its own 9 Apr 2026 post says Cleveland Clinic 'serves over 15 million patients each year across 23 hospitals'. Cleveland Clinic's 2025 Facts + Figures gives ~3.6M unique patients against nearly 16M patient encounters, so the flagship customer's patient base is overstated roughly four-fold. A company mis-stating the one number in its announcement that can be checked against a primary source is a calibration datum for the numbers that cannot.
The most informative fact in the funding history, and the one the announcement coverage does not dwell on. The interval spans the pivot into healthcare and covers the 2023-2025 window in which healthcare-AI peers raised repeatedly. A company compounding fast enough to raise, generally does.
Company-stated and unaudited. Genuine operating scale if the automations are recent and spread across customers; near-meaningless if cumulative across six years and concentrated in one or two accounts. The source states neither, which is why this reads neutral rather than bullish.
Forbes, 9 Apr 2026, attributing the company. As a performance claim it is respectable for unstructured healthcare documents. As a COGS statement it says up to one in five items still needs a human — and the qualifier concedes the platform-wide rate is lower and unpublished.
Company-listed, and unusually senior for a company with $60M of lifetime capital. It spans exactly the two constituencies that decide this market. Discount appropriately — advisers signal access, not revenue.
As reported in trade coverage of company announcements: Candid Health $120M Series D (Jul 2026, Sixth Street Growth, ~3x the Feb 2025 price); Cohere Health $90M Series C led by Temasek to ~$200M total, with company-stated 660,000+ providers and 12M+ prior-auth requests annually; Adonis $40M Series C (Mar 2026, Quadrille Capital, ~$95M total); Commure $70M for autonomous RCM (May 2026). The category is being funded aggressively and Luminai is not the best-funded participant.
Trends
Cleanest evidence is inside the incumbent's P&L: Waystar's volume-based revenue grew +2.8% in Q2 2026 against +34.5% for subscription (CF-WAY-10) — with the caveat that part of the subscription growth is Iodine, closed 1 Oct 2025 and not anniversaried until Q4 2026. Routing a message is commoditising; deciding and executing is not.
CMS-0057-F sets 7-day standard and 72-hour expedited decision clocks from 1 Jan 2026 and four APIs due 1 Jan 2027 (REG-PA-001, primary); CMS-0062-P extends electronic PA to drugs (REG-PA-003). Genuinely two-sided: mandated change forces re-tooling that entrants win, and standardised endpoints erode the integration difficulty vendors monetise. The commonly-quoted federal savings estimate is deliberately omitted — canon marks it NOT USABLE because the Regulatory Impact Analysis could not be retrieved at the primary source.
Peak XV's stated thesis is that large systems want one platform rather than hundreds of vendors. The preference is real, which is why every point solution now calls itself a platform — meaning the platform claim is table stakes, and the consolidation instinct can just as easily favour Epic or an incumbent RCM suite.
43.7% of US acute-care hospitals in 2025, up from 42.3%, a fifth consecutive year of gains, +77 hospitals and +18,679 beds; TruBridge 7.6%, Altera 2.9%; Oracle Health down 56 hospitals and 14,676 beds as the largest net share loser for a third year (CM-KLAS-01/02/03). Every point of Epic share is a point where the system of record can bundle adjacent workflow at zero acquisition cost.
The structural tailwind under every automation business here — the machine's cost falls each model generation while the person's does not. Also entirely non-proprietary: every competitor gets it, so it lifts the category's economics without differentiating anyone inside it.
Canon CM-PHTI-02 records PHTI's verdict on the adjacent ambient category: burnout improves, financial ROI unproven, with health-system leaders themselves reporting evidence gaps. Administrative automation has a better claim to a measurable general-ledger benefit than a scribe does — but the buyer's scepticism is now pre-loaded across every health-AI deal.
Canon's private cohort of six disclosed marks totals ~$33.05B — OpenEvidence 12 + Commure 7 + Abridge 5.3 + Sword 4 + Hippocratic 3.5 + Ambience 1.25 (CM-COH-04) — but they are last-round marks dated Jun 2025 to May 2026, summed, and set against a 17-name listed pure-play cohort of ~$33.90B of market cap at the single 31 Jul 2026 close (CM-COH-01). Directional only. Luminai is in neither and cannot be added to the private one: it has no disclosed mark. A statement about both its stage and its disclosure.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
OCR, agentic reasoning) Bought, not built. Reading a faxed referral with a handwritten annotation rides on general models whose price falls each generation — a direct gross-margin tailwind and a strategic dependency, since any capability arriving free in the base model stops differentiating Luminai and every competitor at the same moment. The advisory bench (former OpenAI CRO and CPO) suggests unusual awareness of that roadmap; it confers no supply advantage.
Hosting and inference consumed as opex inside an undisclosed gross margin; no owned compute, no data-centre capex. Azure = MSFT, GCP = GOOGL. For a document-heavy workload the meaningful cost is inference throughput, not storage or egress.
EHRs and systems of record as the integration surface Epic, Oracle Health, MEDITECH, athenahealth and the practice-management systems are what the agents read from and write to — simultaneously the technical dependency and, for Epic above all, the competitor. Identical in shape to the asymmetry canon records for Waystar.
The unusual supply line, and the one that decides gross margin. Cases the agent cannot resolve route to a person with full context whose intervention becomes training signal. Whether that person is a Luminai employee, a contracted operator or the customer's own staff is UNDISCLOSED — worth more to an analyst than any other single fact about this company.
The only customer NAMED in public sources; lead investor Peak XV puts the unnamed remainder at 'dozens of provider institutions', unaudited. Per Cleveland Clinic's own 2025 Facts + Figures: 23 hospitals, ~300 outpatient facilities, 83,000 caregivers, nearly 16 million patient encounters against ~3.6 million unique patients, $18.3B of operating revenue and a 5.0% operating margin; the 6,725-bed total is derived by summing the four regional counts the document publishes, not printed there as a total. Deployment announced 9 Apr 2026, referral management as the initial use case. No contract value, term or expansion commitment disclosed. Former CEO Toby Cosgrove is listed among Luminai's advisers — context worth stating plainly.
Large integrated delivery networks and academic medical centres The target segment: high document volume, thousands of routing destinations, and enough administrative headcount for a saving to be visible in the general ledger. Also the slowest-buying segment in American healthcare, with security, BAA, privacy and IT-governance review preceding any production deployment.
Regional health systems and multi-site provider groups The volume market beneath the flagships, and where a 48-day time to value matters most because there is no internal integration team to carry a long implementation. Lower ACV, and the segment most exposed to an EHR vendor simply bundling the capability.
Revenue-cycle and patient-access departments as the specific buyer The economic buyer is an operations leader with a labour line — not a CIO with a software budget, not a clinical chief with a quality mandate. That is what places Luminai in Waystar's payer class, and also why the sale is measured against a headcount plan: easier to justify, and easier to defer.
The direct comparison and the reason this dossier is framed as it is — the listed name in the same payer class, paid out of the provider's collected revenue rather than an IT opex line or a fee schedule. Scale gap: TTM revenue $1.21B, +19.2% to 30 Jun 2026, gross margin 68-69% (CF-WAY-16); adjusted-EBITDA ladder 40.65% FY2024 / 42.04% FY2025 / 42.76% Q2 2026 / 42.95% 1H 2026 (CF-WAY-05 to 08); NRR 108.3% with 1,453 clients above $100k LTM (CF-WAY-12); FY2026 guidance $1.276-1.294B revenue and $535-545M adjusted EBITDA (CF-WAY-13); $4.05B market cap at $21.11 on 31 Jul 2026 (CF-WAY-01/02). Overlaps Luminai directly in prior auth, claims and eligibility, and added clinical documentation integrity with Iodine ($1.25B, closed 1 Oct 2025, CF-WAY-17). Two qualifiers cut Luminai's way — only +140bp of fiscal-year margin gain in the year AI was the whole narrative (CF-WAY-06) and volume-based revenue growing just +2.8% (CF-WAY-10). One cuts against: 89.3% of assets are goodwill and intangibles from ten acquisitions (CF-WAY-15), so Waystar buys entrants rather than out-building them — simultaneously the competitive threat and Luminai's most plausible exit.
PRIVATE and permanently uninvestable — never took outside capital. The falsifier for the category, not merely a competitor: 43.7% of US acute-care hospitals in 2025 and a fifth consecutive year of gains (CM-KLAS-01), roughly $6.7B of 2025 revenue (CM-EPIC-01, company-stated to a trade outlet, unauditable because Epic files nothing). Referral routing, authorisation workflow and document intake are all natural in-suite extensions shipped inside a contract, BAA and security review Epic already owns, with a marketplace toll of roughly $1,700-1,900 per vendor per year deciding who else plays (CM-EPIC-02). Cleveland Clinic runs Epic, so Luminai's flagship is also the clearest live test of whether a third party can hold a workflow inside an Epic shop.
Owns the largest competing clearinghouse and sits on the PAYER side of the transactions a provider-side automation vendor is trying to win — a competitor that is simultaneously the counterparty. Vertically integrated across payer, provider and payments, with a balance sheet no private entrant can match, and with discretion over how it implements the CMS-0057-F APIs that decides how easy provider-side automation becomes.
PRIVATE. The most direct prior-authorisation competitor and Luminai's mirror image: it sells to HEALTH PLANS, automating utilisation management from the payer side. As reported in trade coverage of the company's announcement: ~$90M Series C led by Temasek taking total capital to ~$200M, with company-stated coverage of 660,000+ providers, 12M+ prior-auth requests annually and roughly 85% approved in real time. Note the symmetry — if the payer side automates approval, provider-side submission automation is worth less. Competitor and partial substitute at once.
PRIVATE. AI-native revenue-cycle automation and the best-funded direct analogue: a reported $120M Series D led by Sixth Street Growth in July 2026 with Oak HC/FT, 8VC and Y Combinator, described in trade coverage as roughly 3x the February 2025 Series C valuation. Overlaps in claims, billing exceptions and eligibility from a billing-first rather than intake-first start — and is the best evidence that capital is available here to companies that can show numbers.
PRIVATE, marked at $7B on a $70M round led by General Catalyst with Sequoia, Morgan Stanley and Kirkland & Ellis, announced 19/20 May 2026 and directed at autonomous RCM (CM-PRIV-CM-01; canon also flags that the widely-quoted '~$750M raised to date' CONFLATES equity with a $200M General Catalyst Customer Value Fund financing, CM-PRIV-CM-02). The roll-up assembling ambient documentation, RCM and operations tooling into one health-system platform — the same consolidation thesis, at a $7B mark, sharing an investor with Luminai's Series A. The most awkward competitor on this list.
PRIVATE. AI orchestration for revenue-cycle operations; a reported $40M Series C led by Quadrille Capital in March 2026 taking total equity to ~$95M, with company-claimed 4x+ revenue growth in 2025 and net retention above 130% (trade coverage of the company's announcement, unaudited). The relevant contrast is disclosure discipline rather than scale — Adonis publishes a growth rate and a retention figure; Luminai publishes neither.
PRIVATE. Payer-founded clearinghouse and the other large independent provider-payer network, overlapping directly in eligibility, claims and prior-auth traffic. Its ownership roots in the health plans give it payer-side connectivity advantages and a structurally different pricing incentive — and make it one of the parties best positioned to absorb the CMS-0057-F API transition rather than be disrupted by it.
BOTH PRIVATE. R1 was taken private on 19 Nov 2024 by TowerBrook and Clayton, Dubilier & Rice at $14.30 per share, valuing it around $8.9B, and can now invest without quarterly scrutiny; Ensemble is an end-to-end revenue-cycle OUTSOURCER rather than a software vendor. They compete for the same budget from the opposite direction — taking the function over instead of tooling it. Where an outsourcer wins, automation software may still sit underneath but the customer relationship and pricing power move to the outsourcer, whose own incentive to automate is the strongest in the market because the labour is on its P&L.
ALL PRIVATE. The adjacent agentic-operations cohort Forbes named alongside Luminai (Ambience Healthcare, Artera, Assort Health, UnityAI, Dyania Health), plus the intake and payer-call automation names. Individually narrower than Luminai's platform claim — Assort and Infinitus on voice and payer calls, Artera on patient communication, Notable on intake and scheduling — but collectively they are why the platform pitch is necessary: a health system can assemble most of Luminai's surface area from four point vendors, and the buyer's willingness to consolidate is an assumption rather than a fact.