What changed
FACT (Federal Register, 2026-06-17): FDA classified 'radiological machine learning-based quantitative imaging software with predetermined change control plan' into class II with special controls. This converts one-shot clearance into a recurring documentation duty: every model update shipped under a PCCP must be executed and evidenced per the pre-authorized protocols.
Why now
The classification rule is ~3 weeks old, so PCCP-maintenance practice is unformed and no tooling standard exists yet. Each new clearance under this pathway adds a structurally forced recurring buyer. HYPOTHESIS: the window before eQMS incumbents ship PCCP modules is short (12-24 months), not permanent.
Converging signals
One hard signal: the FDA classification rule itself (FORCED BUYER). No PAIN or HIRING/SPEND evidence was provided in this input — the claims that small AI radiology firms staff this with scarce RA hires, and that consultants price out small firms, are INFERENCE, not fact. The convergence's own testable prediction (≥5 job postings citing PCCP burden) is unverified.
Customer pain
HYPOTHESIS: RA teams at small AI-imaging companies must produce verification results, dataset lineage, performance deltas, labeling updates, and change logs for every model release, exactly matching their authorized PCCP. The penalty for drift is existential (adulterated/misbranded device, fresh 510(k), loss of continuous deployment). Pain is structurally plausible but not yet evidenced by complaints or postings in this input.
Who pays
Regulatory affairs / quality lead at FDA-cleared or clearance-seeking AI radiology software companies, mostly startups under ~100 employees (enumerable via FDA's public AI-enabled device list and 510(k) database). Budget-holder is usually the VP RA/QA or founder-CEO at that size.
Solved today
INFERENCE: a mix of RA consultants ($200-400/hr), fractional RA hires, and generic eQMS/document-control systems (Greenlight Guru, Qualio) plus spreadsheets and Word templates. Ketryx already markets continuous-compliance change management for AI/ML medical software, including PCCP support — this is NOT greenfield.
Why current solutions are bad
Generic eQMS tools manage documents but don't ingest a model release (eval results, dataset manifests, performance deltas) and auto-assemble the PCCP-conformant dossier. Consultants are per-hour and don't scale with release cadence. BUT: if a company ships only 1-2 updates/year, consultant hours beat a subscription — this is the falsification condition and it is untested.
Proposed product
'PCCP Ledger': ingest each model release (metrics JSON, dataset lineage, eval reports), map it against the company's authorized PCCP protocol, auto-generate the conformant update dossier + labeling-change diff + cumulative audit ledger, and flag any change that falls outside the pre-authorized envelope before it ships.
MVP version
A dossier generator for ONE published PCCP structure (FDA's PCCP guidance defines the anatomy): upload PCCP + release artifacts, get back a conformance-mapped dossier draft and gap list. No QMS integration, no e-signatures. Sold as a pilot to 2-3 design partners found via the FDA AI device list.
30-day build
Validation only, ~$0 build: (1) enumerate the buyer class from FDA's AI-enabled device list + 510(k) DB, filter to <100-employee firms via LinkedIn; (2) scan RA job postings and RAPS forums for explicit PCCP authoring/maintenance language; (3) book 5 discovery calls with RA leads. Kill if <3 calls confirm recurring PCCP-maintenance work and willingness to pay.
60-day build
If validated: build the single-format dossier generator with 2 design partners' real (redacted) PCCPs; deliberately human-in-the-loop so output is 'draft for RA review', which lowers the trust bar and liability surface.
90-day revenue plan
Convert design partners to paid pilots at $500-1,000/mo or $1,500-3,000 per release-dossier. Realistic first revenue is day 120-180 given regulated-buyer diligence, which the founder's runway can absorb.
Distribution path
Direct and content-led: the buyer class is small and fully enumerable (FDA public lists), so outreach is a finite checklist, not ad spend. RAPS community, LinkedIn RA groups, and a free 'PCCP readiness checklist' lead magnet keyed to the new rule. This fits demonstrated-value selling.
Pricing hypothesis
Hypothesis: $500-1,500/mo subscription for release cadences ≥ quarterly, or per-dossier pricing ($1,500-3,000) for infrequent updaters — per-dossier hedges the low-cadence falsification risk.
Technical difficulty
Moderate. Document assembly + diffing + template mapping is squarely solo-AI-buildable. The hard part is not code: it is encoding correct regulatory interpretation of PCCP conformance, where an error damages a customer's regulatory standing.
Legal / regulatory risk
Material but bounded: the tool is not a medical device (it produces documentation, not clinical output), but errors in audit-facing dossiers create professional-liability exposure and reputational kill risk. Needs 'draft, RA-reviewed' positioning and E&O insurance. This is the founder's stated avoid-zone (heavily regulated medical adjacency).
Platform dependency
Low. Inputs are FDA public databases and customer artifacts; no marketplace or platform gatekeeper.
Founder fit
Mixed, and weaker than the surface pattern-match suggests. The government-portal-mandate lesson (confidence 0.80) applies only partially: unlike ELDT, there is NO portal submission per update — PCCP dossiers are maintained internally for audit, so the per-transaction filing wedge is absent. Fit positives: mandate-reading, enumeration of forced filers, automation, fast prototyping. Fit negatives: founder explicitly avoids heavily regulated medical products and long trust-building plays, and he has zero medtech RA credibility, which matters when the deliverable is audit-facing.
Breakout potential
If PCCP becomes the default pathway across AI-enabled device categories (cardiology, pathology, ophthalmology), the ledger generalizes into 'change-control-of-record for AI medical devices' — a real expansion story, but one that attracts funded incumbents (Ketryx) faster than a solo operator can defend.
Final recommendation
DO NOT BUILD NOW. The forced-buyer mandate is real and freshly minted, but the input contains zero evidence of current spend or pain (no postings, no complaints), the class is small and possibly low-cadence, a credible incumbent (Ketryx) already claims this ground, and the audit-facing medtech trust cycle contradicts the founder's profile. Run the 30-day zero-build validation (enumerate class, scan postings, 5 discovery calls); revisit only if ≥3 RA leads confirm recurring paid PCCP-maintenance work and per-dossier willingness to pay.
Next action
Spend one day pulling FDA's public AI-enabled device list, cross-referencing LinkedIn headcounts, and searching current RA job postings for 'PCCP' — this directly tests the hypothesis's own prediction at zero cost before any further investment.