The thesis in five sentences
The regulated segment gets prior-authorization APIs by 1 January 2027; ERISA self-funded employer plans, drug prior authorization and commercial plans outside the federal exchanges do not.2 During the transition, coordinators work a mixed environment, some payers structured and most not, which is the most expensive configuration a desk can be in. Even where the API arrives, it carries the request, not the decision, and the denial that follows still needs an appeal someone writes. We already have a working in-session runtime, not a specification, and we reached healthcare by testing one engine on several kinds of portal work until the evidence pointed here. What we are raising for is the cost of putting that runtime on the first 10 desks and publishing what it measures.
1. What the rule actually leaves on the desk
CMS-0057-F names its impacted payers: Medicare Advantage organizations, state Medicaid and CHIP programs and their managed-care plans, and qualified health plan issuers on the federally facilitated exchanges. Everything outside that list keeps its portal and its fax line.
| Segment or workflow | Covered by CMS-0057-F? | Residual portal work after 2027 |
|---|---|---|
| Medicare Advantage, Medicaid and CHIP, exchange QHPs | Yes. A FHIR prior-authorization API is required. | Shrinking. Structured exchange replaces the portal for covered requests. |
| ERISA self-funded employer plans | Not an impacted payer. | Persists. Portal and fax remain the mechanism. |
| Drug prior authorization | Excluded from the API provisions. | Persists. Specialty and step-therapy work stays manual. |
| Requests routed to clinical review | The API carries the request, not the decision. | Persists. Documentation, attachments and appeals are still assembled by hand. |
| Denials, appeals, status chase | Partly. Denial reasons are required; the appeal workflow is not. | Persists. The rebuttal is still human work. |
| The 2026–2028 transition | Phased, payer by payer, often late. | Peak pain. Mixed environments cost more than uniformly manual ones. |
Why this is a window, not a permanent market
We say this plainly because it sets the shape of the company. The regulated slice of prior authorization is being automated by mandate, and we are not competing with a federal API. Our addressable work is the exempt and excluded remainder, the appeals that follow denials everywhere, and the multi-year transition. That work is large, urgent and getting more concentrated. A desk running a mixed book in 2027 has fewer manual payers but no less confusion about which is which.
2. We sell a measurement, not a promise
We withdrew the pass/fail savings guarantee on purpose. It made us the underwriter of the client's cohort, portal family and exception rate, which are variables we do not control. What replaces it is a published 30-day protocol on one named portal family.
- First 10 desks
- 30 days free on one named portal family, in exchange for one coordinator, a pre-period baseline and a blinded ledger we may publish.
- After day 30
- The desk stops, or subscribes to S-Practice at $490/mo ($5,400/yr), S-Desk at $1,950/mo ($21,000/yr), or S-Agency at $4,900/mo ($54,000/yr). No case metering and no share of recovered revenue.
- What the desk keeps
- The minutes. We publish the blinded ledger, with no recovered-revenue claim attached.
- What an investor gets
- A measured operating number from a named protocol instead of a testimonial.
3. Five things that are true of us and rarely true at this stage
The runtime exists and runs
embed.js, the operator overlay, screen-grounded staging and the Operator Extension v2.6.8 are working software on synthetic cases, not a roadmap. A pre-seed check here funds distribution and proof, not the question of whether the thing can be built.
The wedge is a conclusion, not a hunch
One engine was tested on several kinds of portal-heavy work before healthcare. Denials and prior authorization won on observed friction, buyer urgency and willingness to engage. We tested and dropped the alternatives at our own cost.
The human gate is strategy, not a limitation
Teams chasing full autonomy meet a compliance committee that will not accept unattended filing against a payer. We stop before the irreversible click on purpose. That is what lets a desk say yes without a risk review that kills the pilot. It costs us the "fully autonomous" headline and buys us the deployment.
We can find our own buyers
Our targeting work ranks operators with residual portal work by payer surface, workflow type and desk shape. Most companies at this stage cannot name their first twenty customers. We can produce that list on request and show the scoring behind it.
The commercial structure removes the buyer's risk first
A free, frozen-cohort month with a published ruler means a desk can say yes without betting on a vendor's projection. This is why the first desks are reachable at all.
4. What compounds after the first desks
- Per desk
- Portal behavior corpus. Each measured lane teaches us how a payer family behaves: field meanings, which attachments are accepted, which rejections are formatting and which are substance, and where staging breaks. This only comes from live desks, and a competitor with identical architecture starts it at zero.
- Per case
- The evidence ledger. Hashed provenance of what was staged, by whom, and what the payer returned. Months of audit-ready record become something a desk depends on and would have to rebuild elsewhere.
- Per relationship
- Reference desks. In healthcare administration, a named peer describing measured results outsells any feature list. The first references are the hardest thing to get and the most valuable thing to hold. This is where an investor with a healthcare portfolio changes our trajectory rather than only funding it.
On the architecture itself
In-session execution, screen grounding, a human gate and no chart data in DeskGate databases are the right engineering answers to hospital VPNs, MFA, bot management and compliance review, and other serious teams are reaching the same conclusions. We do not present them as a barrier to entry. They are why we work today where cloud RPA does not. The durable position is built on top of them, from the three assets above, and it is earned desk by desk.
5. The case against us, stated by us
We would rather you test these than discover them.
- FHIR adoption could run faster and broader than the rule requires. Our residual would shrink sooner. We think the exempt segments make that a long tail rather than a cliff, but it is the central risk.
- The architecture can be copied. A funded team could build it in a quarter. Our answer is speed to live desks and what accumulates there.
- We have no paying customers yet. No logos, no revenue, no completed measurement. That is what this round is for.
- Healthcare sales cycles are slow and reference-gated, which is why we lead with a free 30-day measurement and why a portfolio introduction is worth more to us than the check alone.
- We are a small team. Two founders. The runtime is built; the distribution is not.
6. What we are not claiming
- Not claimedThat we replace Epic, a clearinghouse or any system of record.
- Not claimedThat we make clinical or medical-necessity decisions. The coordinator and the treating clinician do.
- Not claimedThat anything is filed without a person clicking Submit. Autonomous submissions are zero.
- Not claimedThat nothing is retained anywhere. Chart notes and member IDs are not written to DeskGate databases; screen reads go to Vertex AI under the Google Cloud BAA executed 28 September 2026, whose terms govern Google's side (Master Terms §4).
- Not claimedA guaranteed savings figure. We publish a measured ledger instead.
- Not claimedThat any of this is proven at scale. It is not yet. That is the point of the first desks.
7. Sources
- CMS, Interoperability and Prior Authorization Final Rule (CMS-0057-F). Impacted payers must implement the Prior Authorization, Patient Access, Provider Access and Payer-to-Payer APIs generally by 1 January 2027; decision-timeframe requirements apply from 1 January 2026.
- CMS-0057-F final rule text. Self-funded ERISA plans are not impacted payers; the prior-authorization API provisions cover items and services, not drugs; requests may still route to clinical review.
- American Medical Association, prior authorization physician survey, fielded late 2024, published April 2025: 39 prior authorizations and 13 hours per physician per week.
- KFF, Claims Denials and Appeals in ACA Marketplace Plans in 2023: about 20% of in-network claims denied; under 1% of denied claims appealed.
- Cloudflare, What is bot management? Bot-management systems use fingerprinting, IP reputation and behavior signals that flag automated headless browsers.
- He et al., WebVoyager: Building an End-to-End Web Agent with Large Multimodal Models (2024). An agent reading screenshots completed more live-website tasks than the same setup reading page text alone.
- HHS, HIPAA Security Rule, technical safeguards at 45 CFR § 164.312. Keeping chart facts out of DeskGate databases narrows what those safeguards must cover; it does not remove DeskGate's duties as a business associate.
These references describe the regulatory and technical environment. They are not evidence of DeskGate's own performance. That is what the 30-day measurement exists to produce, and it has not been produced yet.
The ask
A pre-seed SAFE, and an introduction to one or two desks you already know.
DeskGate is raising a $150k–$250k post-money SAFE on a $2M–$3M valuation cap. The capital funds the deployment. The introduction is what turns a working runtime into a published ledger, and that ledger is what prices the next round. If you hold a healthcare portfolio, you can do something for this company that money alone cannot.