FDA AI-Device PCCP Workspace

Analysis

FDA AI-Device PCCP Workspace

1. Title

FDA AI-Device PCCP Workspace for submission-ready model-change evidence

2. One-line thesis

Build a narrow, consultant-friendly workspace that turns one FDA-authorized Predetermined Change Control Plan into a controlled evidence room: planned modifications, modification protocol, validation/test artifacts, risk impact assessment, review state, implementation history, and submission/export readiness for AI-enabled medical-device teams.

3. ICP

Primary buyer: specialist medical-device regulatory consultancies and fractional regulatory leaders helping AI-enabled SaMD / device-software companies prepare 510(k), De Novo, or PMA submissions.

Secondary buyer: small-to-mid AI-enabled medical-device manufacturers with one or a few FDA submissions, especially radiology, diagnostics, monitoring, clinical-decision-support, and software-heavy device teams that expect model/data/performance updates after clearance.

Best initial wedge: teams that already have an eQMS, document repository, Jira/GitHub/ML experiment stack, and outside regulatory counsel, but no shared operational layer for one PCCP. The pain is not “manage all AI lifecycle governance.” It is: can we prove that this specific planned AI/device modification fits the authorized PCCP, passed the pre-specified tests, received the right quality/regulatory approvals, and can be summarized for FDA without reassembling evidence from six systems?

4. Hard-fact grounding

The parent hypothesis is supported by FDA’s own language.

• FDA’s final guidance page for AI-enabled device software functions says PCCPs are intended to support “iterative improvement through modifications to AI-enabled devices” while maintaining reasonable assurance of safety and effectiveness.

• FDA says a PCCP should describe planned device modifications, the methodology to develop, validate, and implement them, and an assessment of their impact. FDA reviews the PCCP as part of the marketing submission so the device can implement described modifications without additional marketing submissions.

• FDA’s broader draft PCCP guidance for medical devices uses the same core structure: what modifications will be made, how they will be assessed, and how FDA can ensure continued safety and effectiveness without a new submission for each modification covered by the PCCP.

• Consultant summaries converge on three named work products: Description of Modifications, Modification Protocol, and Impact Assessment. These are not generic QMS folders; they are a concrete object model for a product.

• Emergo’s summary of the earlier AI/ML PCCP draft says the modification protocol should include data management practices, ML retraining methods, performance evaluation metrics, update procedures, user/stakeholder communication strategies, verification/validation activities, and pre-defined acceptance criteria.

• Clarkston’s summary emphasizes that modifications outside the PCCP, or changes that do not meet PCCP conditions, can require a new filing; implementation also has to be documented through the manufacturer’s quality system.

• FDA’s AI-Enabled Medical Device List shows the market is now large enough to matter: the extracted FDA page contained 1,376 K-number occurrences and at least one listed device explicitly named “Fibresolve (with PCCP)” in November 2025. That is a small but important signal: PCCP is moving from guidance concept toward clearance-list reality.

• FDA’s eSTAR program is mandatory for most 510(k) and De Novo submissions, but eSTAR is a submission template, not a day-to-day PCCP evidence/versioning workspace.

Product implication: the opportunity is plausible only if the product stays narrow. It should not pitch “AI governance for healthcare” or “complete medtech QMS.” It should be the controlled collaboration layer that sits between QMS/design controls, ML experiment tooling, issue trackers, and FDA submission artifacts for a single PCCP.

5. Pain evidence and buyer language

The pain pattern is specialized, but real.

Buyer-language signals from FDA and practitioner commentary:

• “Planned device modifications” implies scope control. A team needs a living map of which algorithm/data/threshold/UI/deployment changes are authorized under the PCCP and which are out-of-scope.

• “Associated methodology to develop, validate, and implement” implies protocol execution. A team needs traceable evidence that retraining, test-data selection, performance evaluation, bias/subgroup checks, risk controls, release gates, and deployment steps followed the authorized protocol.

• “Impact assessment” implies cross-functional judgment, not just file storage. Regulatory, clinical, QA, ML, software, and product need a reviewed risk/benefit rationale for each planned change and for combinations of changes.

• Emergo’s wording is especially product-shaped: clear and traceable descriptions, data-management practices, retraining methods, metrics, update procedures, stakeholder communication, verification/validation, and acceptance criteria.

• Intertek explicitly links PCCP execution to data governance, design controls, version tracking, labeling transparency, performance metrics, and revised instructions for use. That is a workflow coordination problem across systems.

• Clarkston notes that the quality system must document implementation of each change made under the pre-approved PCCP. That makes “PCCP change record” a durable compliance artifact, not a one-time submission narrative.

The best pain evidence is not broad social-media complaint volume. This is a regulated workflow where buyers express pain through guidance, consultant explainers, and paid-service categories. The strongest buying trigger is the cost of avoidable resubmissions or FDA review friction: if the team cannot show that a post-market AI modification stayed inside the authorized PCCP and satisfied the Modification Protocol, the operational benefit of PCCP collapses.

6. What to build

A narrow “PCCP evidence room” with six modules:

1. PCCP scope map

• Structured Description of Modifications: change type, affected AI-DSF, intended-use boundary, user impact, automation/manual implementation, affected populations, affected outputs, labeling/IFU implications.

• Authorized/not-authorized status and “requires new submission?” flag.

• Versioned PCCP baseline from the submitted/authorized plan.

2. Modification Protocol execution tracker

• Protocol steps, owners, due dates, acceptance criteria, required evidence, and reviewer signoff.

• Data-management checklist: dataset version, inclusion/exclusion criteria, representativeness, held-out test set, subgroup metrics, drift checks, provenance.

• Model/software release checklist: training run, model card, software build, cybersecurity dependency note if relevant, deployment plan.

3. Evidence vault and traceability

• Link evidence objects to each PCCP modification and protocol criterion: test reports, validation summaries, risk files, clinical/performance evidence, screenshots, notebooks, requirements, Jira tickets, Git commits, QMS records.

• Hash/version metadata and immutable audit trail.

• “Evidence completeness” dashboard for submission-readiness.

4. Impact Assessment workspace

• Benefits/risks template aligned to FDA PCCP language.

• Risk-control deltas, population/subgroup impact, performance metric comparison, labeling/training implications, residual risk approval.

• Reviewer workflow for regulatory, quality, clinical, ML, software, and executive signoff.

5. Implementation and post-authorization log

• For each implemented modification: PCCP fit rationale, protocol run, final evidence packet, approval state, date implemented, release/version identifiers, labeling/user communication status.

• Exportable change history for internal audits, FDA interactions, and next submission.

6. Submission/export pack

• Generate a PCCP appendix / reviewer packet: Description of Modifications table, Modification Protocol matrix, Impact Assessment summary, evidence index, open issues, reviewer approvals, and “outside PCCP” exceptions.

• Export to Word/PDF/Markdown and eSTAR-friendly copy blocks; do not attempt to replace eSTAR in v1.

7. MVP shape

Weekend-buildable first version:

• Multi-tenant project workspace for one device + one PCCP.

• Templates for Description of Modifications, Modification Protocol, and Impact Assessment.

• Evidence upload/linking with required/optional artifacts per protocol criterion.

• Review states: draft, evidence pending, QA review, regulatory review, approved, implemented, outside PCCP.

• CSV/Markdown/Word export of PCCP tables and evidence index.

• Manual integrations only: paste GitHub/Jira/MLflow/Drive/QMS links instead of building regulated integrations immediately.

• Consultant dashboard across clients: number of PCCPs, modifications in review, evidence gaps, outside-scope flags, next submission deadline.

Avoid in v1:

• Replacing eQMS, design controls, RIM, regulatory intelligence, model monitoring, cybersecurity risk management, or full AI governance.

• Automated FDA submission filing.

• Claiming the product “ensures FDA acceptance.” The value is coordination and submission-readiness, not legal/regulatory guarantee.

8. Distribution wedge

Start with regulatory consultants, not enterprise medtech IT.

• Target boutique FDA/medtech regulatory consultancies publishing about AI/ML SaMD, PCCP, eSTAR, and 510(k)/De Novo submissions.

• Offer a white-label “PCCP readiness room” they can use across clients.

• Create a free PCCP readiness checklist and evidence-matrix template with the three FDA components as columns.

• Run founder-led outreach to consultants who already write PCCP explainers; ask to turn their checklist into a client portal.

• Partner with ML validation / SaMD test labs that need a place to hand evidence back to regulatory owners.

• Use FDA’s AI-enabled device list to identify manufacturers with software-heavy radiology/diagnostic devices likely to face iterative model updates.

Pricing hypothesis:

• Consultant plan: $300-$800/month for 10-25 client PCCP rooms, plus export branding.

• Manufacturer plan: $500-$2,000/month per active device/PCCP workspace, depending on audit-trail and approval features.

• Services-assisted onboarding: $2k-$10k to convert an existing draft PCCP and evidence folder into a structured workspace.

9. Competition / substitutes

Substitutes are real, but none is obviously purpose-built around one PCCP.

• eQMS platforms such as Greenlight Guru, MasterControl, Qualio, and similar systems manage design controls, document control, CAPA, risk, training, and software-release traceability. They are broader systems of record and can store PCCP artifacts, but they are not marketed as PCCP execution workspaces.

• RIM/submission platforms such as Rimsys support regulatory submissions, product registrations, impact assessment, change analysis, and global regulatory operations. They are credible enterprise substitutes, but heavier than a consultant-led PCCP room.

• FDA eSTAR is mandatory for most 510(k)/De Novo submissions, but it is a submission template, not a cross-functional work tracker for evidence, versioning, test execution, and approvals.

• General AI governance platforms can manage model risk, policies, and monitoring, but usually lack FDA-specific PCCP structure, eSTAR/submission export orientation, and medtech quality-system vocabulary.

• Spreadsheets, SharePoint/Drive, Jira, Confluence, GitHub, MLflow, and existing QMS records are the default stack. The product must win by making the PCCP-specific evidence matrix and “inside/outside authorized plan” decision visible.

Positioning line: “Not your QMS. Not generic AI governance. A PCCP control room for one AI-enabled device modification plan.”

10. Why now

• FDA has finalized AI-enabled device PCCP guidance and separately advanced broader PCCP policy for medical devices.

• AI-enabled medical-device authorizations continue to accumulate, and the FDA list already shows at least one listed device with “with PCCP” in the device name.

• AI/ML device teams want iterative updates, but FDA’s benefit depends on disciplined pre-specification and evidence. That creates recurring work after initial submission, not just one-time consulting.

• eSTAR standardizes submission packaging, which can make structured exports more valuable.

• Consultants are educating the market now; a tool that productizes their checklists can ride that education wave before enterprise platforms add specific PCCP modules.

11. Risks and skeptical read

• Market timing risk: PCCPs may remain mostly a guidance/submission construct for a small subset of AI devices rather than a broad operational category. FDA’s AI list currently showed only one obvious “with PCCP” device in extracted text.

• Enterprise platform risk: Greenlight Guru, MasterControl, Rimsys, Veeva-like RIM/QMS suites, or ALM vendors could add PCCP templates quickly.

• Buyer-budget risk: small SaMD companies may prefer consultant-managed spreadsheets until they have multiple devices or active post-market updates.

• Regulatory-liability risk: a tool touching FDA submission evidence must be careful with claims, audit trail, validation, data security, and 21 CFR Part 11 expectations if used as a regulated record system.

• Scope-creep risk: the obvious adjacent problems—AI lifecycle management, model monitoring, cybersecurity, QMS, RIM, eSTAR authoring—are much larger and more competitive. The wedge works only if it stays PCCP-specific.

• Evidence weakness: there is limited public complaint/forum evidence. The opportunity is inferred from workflow complexity, regulatory guidance specificity, consultant activity, and existing substitute gaps.

Self-critique: The opportunity is not yet a screaming market with visible buyer pull. It is a specialist wedge at the moment when a new FDA operating pattern is being adopted. The best validation is not more desk research; it is 10 consultant interviews asking, “Show me the spreadsheet/folder you use to manage one PCCP’s modifications, protocol evidence, and approval state.” If they do not already have a messy version of this, do not build.

12. Scorecard

• Pain: 7/10 — high consequence and cross-functional complexity, but concentrated in a specialized subset of AI-device teams.

• Willingness to pay: 7/10 — consultancies and submission-bound manufacturers pay for regulatory workflow tools and readiness support; stronger if sold with onboarding/templates.

• Reachability: 7/10 — consultants, FDA AI-device list companies, SaMD founders, and medtech regulatory communities are identifiable.

• MVP simplicity: 8/10 — a useful first version is structured CRUD, evidence links/uploads, review states, templates, and exports; regulated integrations can wait.

• Competition: 5/10 — broad QMS/RIM/submission platforms are formidable substitutes, but the narrow PCCP workspace is not obviously saturated.

• Overall: 7/10 — buildable niche wedge, best as a consultant-first workflow product; validate fast before assuming large standalone SaaS demand.

Verdict: CONDITIONAL BUILD. Build only if consultant discovery confirms that PCCP evidence assembly is already handled manually and repeatedly. The right first customer is a regulatory consultancy with multiple AI-device clients, not a single startup hoping FDA will bless future model changes.

13. Sources

• FDA — Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions: https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence

• FDA — Predetermined Change Control Plans for Medical Devices draft guidance page: https://www.fda.gov/regulatory-information/search-fda-guidance-documents/predetermined-change-control-plans-medical-devices

• FDA — Artificial Intelligence-Enabled Medical Devices list: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices

• FDA — eSTAR Program: https://www.fda.gov/medical-devices/how-study-and-market-your-device/estar-program

• Intertek — How FDA’s PCCP Framework Supports AI-Enabled Medical Devices: https://www.intertek.com/blog/2025/03-25-fdas-pccp-framework-and-ai-enabled-medical-devices/

• Emergo by UL — FDA draft guidance synopsis on PCCPs for AI/ML-enabled medical devices: https://www.emergobyul.com/news/fda-draft-guidance-predetermined-change-control-plans-artificial-intelligencemachine-learning

• Clarkston Consulting — PCCP for AI: FDA guidance on AI in medical devices: https://clarkstonconsulting.com/insights/pccp-for-ai/

• Greenlight Guru — medtech QMS / design-control positioning: https://www.greenlight.guru/

• MasterControl — document control / life-sciences quality platform positioning: https://www.mastercontrol.com/quality/document-control-software/

• Rimsys — medtech RIM/submission platform and eSTAR explainer: https://www.rimsys.io/blogs/what-is-the-fda-estar-program