Preparing an FDA De Novo request for a novel medical device requires more than demonstrating that the technology is innovative. Sponsors must explain why the device belongs in Class I or Class II, demonstrate that general controls or general and special controls can provide reasonable assurance of safety and effectiveness, and provide enough evidence for FDA to conduct a substantive review.

For companies developing AI-enabled or software-based devices, the challenge is even greater. The regulatory landscape is evolving, comparable products may be difficult to identify, and relevant evidence can be spread across FDA databases, guidance documents, decision summaries, clinical literature, and competing technologies.

This is where competitive intelligence in healthcare can support De Novo preparation. AI can help sponsors search FDA precedents, organise comparable devices, identify regulatory patterns, map evidence gaps, and structure submission content. The important principle is that AI should accelerate regulatory research and analysis rather than replace qualified regulatory judgment.

What Is the FDA De Novo Pathway?

The FDA De Novo pathway provides a route for certain novel medical devices that have no legally marketed predicate but for which general controls, or general and special controls, can provide reasonable assurance of safety and effectiveness. Devices classified through De Novo can generally be placed into Class I or Class II and may serve as predicates for future 510(k) submissions where appropriate.

FDA currently describes two routes for submitting a De Novo request.

The first is after receiving a high-level not-substantially-equivalent determination following a 510(k) submission.

The second allows a sponsor to submit a De Novo request directly when it determines that there is no legally marketed device on which to base a substantial-equivalence determination.

For a truly novel device, the second route can be particularly relevant.

The strategic question is therefore not simply whether a product is innovative.

It is:

Can the sponsor demonstrate a reasonable regulatory classification and establish controls that provide reasonable assurance of safety and effectiveness?

Why FDA Intelligence Matters for a De Novo Request

A De Novo request is fundamentally evidence-driven.

FDA needs to understand:

• What the device does

• What its intended use is

• How the technology works

• What risks it creates

• How those risks can be controlled

• What evidence supports safety and effectiveness

• Why the proposed classification is appropriate

FDA's public De Novo database provides a valuable source of precedent. As of August 2026, the database contains hundreds of De Novo decisions, including recent decisions involving software, robotics, diagnostic technologies, and other novel devices.

This creates an opportunity for AI-assisted regulatory intelligence.

Instead of manually reviewing every potentially relevant decision, teams can use AI to organise precedents around:

Device type → Intended use → Technology → Risk → Evidence → Classification → Special controls

The regulatory team can then examine the most relevant examples in detail.

Step 1: Confirm That De Novo Is Appropriate

Before using AI to build the submission, confirm that the device is actually suitable for the De Novo pathway.

The absence of an obvious predicate is an important consideration, but it is not the only one.

The sponsor should evaluate:

• Intended use

• Indications for use

• Technological characteristics

• Known risks

• Potential harms

• Available predicates

• Applicable product codes

• Existing classifications

• Potential general controls

• Potential special controls

• Expected device classification

FDA recommends that sponsors consider a Pre-Submission before submitting a De Novo request to obtain feedback from the appropriate premarket review division.

AI can support this initial assessment by building a structured landscape of potentially relevant devices and classifications.

However, the final pathway decision should remain with the regulatory team.

Step 2: Define the Intended Use Before Searching for Precedents

One of the most important inputs into regulatory intelligence is the intended use.

A poorly defined intended use can produce a misleading competitive or regulatory landscape.

Before asking AI to find comparable devices, define:

• Intended use

• Indications for use

• Target population

• User

• Clinical setting

• Disease or condition

• Inputs

• Outputs

• Method of operation

• Human involvement

• Treatment or diagnostic role

For an AI-enabled medical device, the intended use should also clarify how AI contributes to the device's function.

For example, is the AI:

• Detecting a condition?

• Supporting diagnosis?

• Predicting risk?

• Prioritising patients?

• Supporting treatment decisions?

• Automating a clinical function?

• Generating measurements?

• Analysing images?

FDA notes that different AI-enabled device applications can require different assessment approaches, including different performance metrics and reference standards.

That makes intended-use definition a foundational step in regulatory strategy.

Step 3: Build an FDA Regulatory Intelligence Landscape

Once the intended use is defined, AI can help search FDA intelligence sources.

Useful sources include:

• FDA De Novo decisions

• 510(k) databases

• Device classification information

• FDA guidance documents

• FDA safety communications

• AI-enabled medical device information

• Public decision summaries

• Relevant clinical literature

FDA maintains an AI-enabled medical device list that can provide insight into devices that have met applicable premarket requirements. The agency states that the list is intended to provide transparency and help innovators understand the current AI-enabled device landscape.

A regulatory intelligence workflow can classify relevant precedents according to:

This gives the sponsor a more structured view of the regulatory environment.

Step 4: Use AI to Identify Similar De Novo Decisions

AI is particularly useful for narrowing a large database.

The search should not rely only on exact product names.

Instead, AI can look for similarities in:

• Clinical indication

• Intended use

• User population

• Device function

• Software functionality

• Technology

• Risk profile

• Performance endpoint

• Classification

• Special controls

For example, a sponsor developing an AI-assisted imaging device may want to examine previous De Novo decisions involving image analysis, computer-aided diagnosis, automated detection, or related clinical functions.

The objective is not to copy another submission.

It is to understand the types of evidence FDA has considered relevant for comparable risk profiles.

Step 5: Analyse Why FDA Classified Comparable Devices

Classification precedent can be especially valuable.

FDA's De Novo process is explicitly risk-based. The agency evaluates whether general controls or general and special controls can provide reasonable assurance of safety and effectiveness.

AI can compare previous decisions and identify recurring patterns around:

• Device risk

• Intended use

• Clinical evidence

• Nonclinical testing

• Software validation

• Performance testing

• Human factors

• Labelling

• Special controls

This can help the sponsor formulate a preliminary classification rationale.

The key is to distinguish between:

Observed precedent

and

Assumption about future FDA action.

A previous decision can inform strategy, but it does not guarantee the outcome for a new device.

Step 6: Build the De Novo Classification Rationale

The De Novo request needs to include the proposed classification and supporting rationale.

FDA's current De Novo information identifies classification information and supporting data as core components of the request, including a complete discussion of why general controls or general and special controls provide reasonable assurance of safety and effectiveness.

AI can help structure this rationale around:

Risk → Control → Evidence → Residual risk → Proposed classification

For a proposed Class II device, the sponsor may need to explain the special controls that would allow FDA to conclude that the device is reasonably safe and effective.

AI can compare controls used for similar devices and identify possible areas for consideration.

The regulatory team must then determine whether those controls actually fit the new device.

Step 7: Map Device Risks to Proposed Controls

A strong De Novo submission should connect identified risks to mitigation measures.

For example:

AI can help build and maintain this matrix.

It can also identify situations where a claimed control has no clearly associated supporting evidence.

This is one of the most practical uses of AI because it focuses on traceability, rather than asking AI to make an independent regulatory determination.

Step 8: Prepare the Device Description

FDA identifies device description as a core component of a De Novo request. The description should explain the technology, proposed conditions of use, accessories, components, and other relevant characteristics.

AI can help convert technical documentation into a structured regulatory outline.

For an AI-enabled device, the description may need to explain:

• Hardware

• Software

• Algorithms

• Inputs

• Outputs

• User interface

• Data flows

• Connectivity

• Intended operating environment

• Human interaction

• Performance limitations

The technical team should validate every generated description.

A regulatory submission should never rely on an AI-generated technical explanation that has not been checked against the actual device.

Step 9: Use FDA Intelligence to Identify Evidence Expectations

Different device types can require different evidence.

AI can compare relevant FDA decisions and guidance to identify recurring evidence categories.

These may include:

• Analytical performance

• Clinical performance

• Software validation

• Electrical safety

• Biocompatibility

• Usability

• Cybersecurity

• Human factors

• Bench testing

• Imaging performance

• Algorithm performance

For AI-enabled devices, performance assessment can become especially important.

FDA notes that new AI applications and indications may introduce novel assessment questions, while different AI applications can require different metrics and reference standards.

This makes precedent analysis more valuable than simply following a generic checklist.

Step 10: Analyse the AI Component Separately

If the medical device incorporates AI, the AI component should be evaluated within the broader device context.

Questions may include:

• What data were used?

• How representative are the datasets?

• How was the model validated?

• What performance metrics were used?

• Were relevant subgroups assessed?

• What are known limitations?

• What happens when inputs fall outside expected conditions?

• How are model changes controlled?

• How is performance monitored after deployment?

FDA's AI work increasingly addresses these questions.

In August 2026, FDA issued a discussion paper specifically addressing considerations for regulating generative AI-enabled medical devices, including risk assessment, premarket evaluation, and postmarket monitoring.

This signals the importance of considering AI-specific risks as part of broader device regulatory planning.

Step 11: Use AI to Compare Your Evidence With Precedent

Once the evidence package has been assembled, AI can help identify potential gaps.

For example:

Your device

• Clinical validation available

• Software verification complete

• Cybersecurity testing complete

• Limited demographic subgroup analysis

Comparable precedent

• Clinical validation available

• Software verification complete

• Cybersecurity testing complete

• Broader subgroup analysis

The gap does not automatically mean that additional evidence is required.

It means the regulatory team has identified an area requiring assessment.

This distinction prevents AI from turning precedent analysis into an unsupported checklist.

Step 12: Use a Pre-Submission to Test the Strategy

FDA recommends considering a Pre-Submission before a De Novo request.

A Pre-Submission can be used to obtain FDA feedback on important questions before the formal request.

AI can help prepare the questions by identifying:

• Areas of uncertainty

• Evidence gaps

• Classification questions

• Proposed special controls

• Testing strategies

• Clinical evidence questions

• Software considerations

A useful question structure is:

FDA question → Sponsor position → Supporting evidence → Reason for uncertainty → Potential development impact

This is more useful than asking broad questions such as:

"Is our device safe?"

Step 13: Prepare the eSTAR Submission

An important operational change is that, beginning October 1, 2025, FDA requires De Novo requests to be submitted electronically using the eSTAR format unless an applicable exemption applies.

This means submission preparation should account for the electronic structure from the beginning.

AI can help organise information before it is entered into the required submission structure.

It can support:

• Content mapping

• Section completeness checks

• Cross-reference checking

• Terminology consistency

• Missing-evidence identification

• Document organisation

But the final submission should follow FDA's official eSTAR requirements.

Step 14: Create a Single Evidence Map

A useful approach is to create a central evidence map connecting every major regulatory claim to its supporting source.

For example:

Regulatory requirement → Submission section → Evidence → Source → Test → Owner

This helps prevent one of the most common problems in complex submissions: evidence that exists but cannot be quickly connected to the relevant regulatory argument.

Pienomial's Life Sciences solution can support evidence-driven workflows for life sciences organisations working across scientific, regulatory, clinical, and competitive intelligence.

Its platform is designed to connect information and context across enterprise AI workflows.

For research-intensive teams, Knol AI can support AI-assisted research and synthesis, while Knol Composer can help turn validated research into structured outputs.

The value is not simply generating content faster.

It is keeping research, evidence, and regulatory reasoning connected.

Step 15: Monitor New FDA Decisions

Regulatory intelligence should continue after the initial landscape analysis.

FDA's De Novo database is updated as new decisions are made. Recent 2026 decisions include novel technologies such as autonomous robotic systems, cardiac analysis software, surgical systems, and AI-enabled devices.

For a sponsor preparing a future submission, monitoring these decisions can reveal emerging patterns in:

• Device classification

• Special controls

• Evidence expectations

• AI-related considerations

• Software requirements

• Clinical validation

• Postmarket controls

AI can flag potentially relevant decisions and route them to the regulatory team for review.

Step 16: Perform a Final Regulatory Intelligence Review

Before submission, conduct a final intelligence review.

Ask:

Has the FDA landscape changed?

Check for new guidance, new classifications, and recent De Novo decisions.

Has a new predicate or comparable device appeared?

A newly marketed device could change the strategic analysis.

Have similar devices received different classifications?

If so, understand why.

Are there new AI-specific regulatory considerations?

FDA's AI regulatory work continues to evolve, making this especially important for AI-enabled devices.

Does every major claim have evidence?

If not, identify the gap before submission.

How AI Fits Into a Medical Device Regulatory Strategy

AI is most useful when it supports the research-heavy components of regulatory work.

AI can help with:

• FDA database searching

• De Novo precedent analysis

• Regulatory landscape monitoring

• Guidance comparison

• Evidence extraction

• Risk-control mapping

• Submission consistency checks

• Gap analysis

• Reference management

• Pre-Submission question development

• Regulatory change monitoring

Experts should retain responsibility for:

• Regulatory pathway selection

• Device classification

• Safety and effectiveness conclusions

• Clinical strategy

• Risk acceptance

• Special-control decisions

• Final FDA communications

• Submission approval

The distinction is important.

AI should make the regulatory team faster and better informed.

It should not make unsupported regulatory conclusions on the sponsor's behalf.

A Practical FDA De Novo AI Checklist

Before submitting, the team should be able to confirm:

• The De Novo pathway has been evaluated against alternative pathways.

• The intended use and indications are clearly defined.

• Relevant FDA classifications have been reviewed.

• Potential predicates and comparable devices have been assessed.

• Relevant De Novo decisions have been analysed.

• The proposed classification is supported by evidence.

• General and special controls have been considered.

• Device risks have been mapped to mitigation measures.

• Safety and effectiveness evidence is organised.

• AI-specific performance considerations have been assessed where applicable.

• Software documentation is consistent with the actual device.

• Relevant FDA guidance has been reviewed.

• A Pre-Submission has been considered where useful.

• eSTAR requirements have been incorporated.

• References have been checked against primary FDA sources.

• AI-generated content has undergone expert review.

• A process exists for monitoring regulatory developments before submission.

Conclusion

Preparing an FDA De Novo request AI medical device strategy requires more than assembling technical documents.

The sponsor needs to understand how FDA has approached comparable devices, why those devices received particular classifications, what evidence supported their decisions, and how those precedents can inform—not dictate—the strategy for a new product.

This is where AI-assisted competitive intelligence in healthcare can create practical value.

AI can rapidly search FDA decisions, organise comparable technologies, extract regulatory patterns, map risks to controls, identify evidence gaps, and monitor new decisions. It can also help regulatory teams maintain a clear connection between FDA requirements, submission arguments, supporting evidence, and device-specific controls.

However, regulatory intelligence should remain evidence-led.

A previous De Novo decision does not guarantee the same outcome for another device. A similar technology may have a different intended use or risk profile. And an AI-generated interpretation should never replace review by qualified regulatory and scientific professionals.

The strongest model is therefore:

FDA intelligence → AI-assisted analysis → expert interpretation → evidence mapping → regulatory strategy → submission

As FDA's approach to AI-enabled medical devices continues to evolve, this model can help sponsors prepare for a regulatory environment where technology changes quickly but the need to demonstrate safety, effectiveness, and appropriate controls remains constant.

For novel medical device developers, the objective is not simply to produce a faster De Novo request.

It is to produce a request in which every major regulatory argument can be traced back to evidence, precedent, risk, and a clearly defined control strategy.