Watch the recent podcast on this topic
AI in Design Controls
SYNOPSIS
Artificial Intelligence (AI) has woven itself into nearly every corner of our work lives. But in medical device and combination product development, where regulatory expectations are high and documentation is everything, AI carries both enormous potential and very real risk.
In our recent Life Science Solutions episode, Kymanox Chief Technology Officer, Nick Ciccarelli, unpacked what AI can (and cannot) do within design controls. Here we summarize the podcast discussion and outline how teams can start using AI responsibly.
Why AI Is Suddenly a Serious Topic in Design Controls
While the topic of Machine Learning has been around for a while, mainstream AI tools became widely accessible in late 2022. Early on people used various AI tools for simple tasks such as, emails, literature search and summarization, and writing small snippets of code. However, three years later, the medical device industry has shifted from casual experimentation to targeted application.
Advancement due to several key changes
- The technology is now mature enough to support technical workflows
- Teams are searching for efficiency in documentation-heavy processes
- Regulators are openly embracing AI — the FDA recently announced use of an internal tool called Elsa
- The burden of documentation has grown with more complex and global supply chains, especially in combination products
- Some data sets across clinical trials or high volume manufacturing can be massive, and AI can finally help make sense of them
AI’s role is not hypothetical anymore, it is practical, immediate, and increasingly expected.
Quick Refresher:
What “Design Controls” Actually Are
Design controls are the structured, documented way we move a device from concept to commercialization. They consist of the following process:
- Design Planning
- Design input requirements review/assessment
- Design Outputs
- Verification (does the device meet the inputs)
- Validation (does it meet user needs)
- Design Transfer
- Life Cycle Management
This process is essential for any device or device constituent of a combination product and every step produces documentation that must be clear, consistent, and justified.
That is exactly where AI can help.
AI Can Help Today: Practical, Verifiable Use Cases
AI should be treated like a tool managed by engineers, not as engineers themselves. When applied by a qualified user to focused, rules-based tasks, it adds real value without compromising compliance.
Here is where AI is already proving useful:
1. Requirements and Traceability
AI can review and summarize guidance documentation and support tasks such as:
- Applying rules to ensure requirements are testable, technical, and unambiguous
- Extracting requirements from standards and guidelines
- Rewriting requirements to follow systems-engineering formats (e.g. INCOSE)
- Automating traceability between risk controls and design outputs
2. Rule-Based Documentation Reviews
AI can scan requirements, protocols, or reports to check for issues such as:
- Missing sections
- Ambiguous language
- Inconsistent formatting
- Misaligned justifications
3. Data Aggregation Across Massive Databases
AI can quickly summarize data from literature and other sources and helps teams “learn from the field” without weeks of manual review. It can support tasks such as identifying the following:
- Use-related problems
- Post-market surveillance trends
- Literature for justification of requirements
- Patterns in FDA Complete Response Letters
4. Design Verification (DV) Support
This is one of the most promising areas for the use of AI. AI can significantly speed up the planning and analysis phases with support in areas such as:
- Sample size determination using defined confidence/reliability levels
- Referencing statistical tables to create consistent sampling plans
- Aggregating verification data across large studies
- Helping identify trends or recurring failures
5. Document Consolidation and DHF Alignment
AI can help save enormous amounts of time in remediation and harmonization efforts for companies that have decades of legacy documentation. AI can provide support for tasks such as:
- Comparing versions and identifying gaps or inconsistencies
- Flagging misalignments across similar products
- Suggesting standardization opportunities
Balancing Automation and Accountability in Design Controls
AI never says, “I don’t know.” In regulated development, “I don’t know” is often the correct answer until proper research, testing, and due diligence has been conducted. However AI’s primary function is to produce an answer, not ensure the answer is correct. This creates several risks, including:
- Over-reliance on automation
- Hallucinated references or data
- False confidence in incorrect conclusions
- Error-filled submissions that could erode Regulatory trust
- Wasted time sifting through irrelevant or unusable output
While you can trust AI with the tasks of a data analyst, you should not trust it with the ambiguity that comes with strategic decision making.
Human, qualified teams must remain editors, reviewers, and decision-makers.
How to Begin Using AI Responsibly in Design Controls
This should be a phased, intentional approach consisting of the following steps:
Start with well-bounded tasks
Examples:
- Statistical calculations
- Requirements rule checks
- Document alignment
- Data extraction
Provide clear rules, references, and expectations
For best AI performance, ensure the following:
- Inputs are structured
- Output formats are specified
- Boundaries are defined
- Have a good understanding of ‘what’ the output should look like
Maintain strong human oversight
Engineers must continue to be actively involved every step of the way and make sure to validate at least the following:
- Data sources
- Methods
- Calculations
- Conclusions
Protect the Right-First-Time mindset
AI does not remove the need for rigor on behalf of engineers. AI requires more discipline from its human supervisors, not less.
The Future: What AI-Enabled Design Controls May Look Like in 5 Years
There are several outcomes for the direction that AI in design controls may take in the near future:
1. Shorter development cycles
For example, AI will streamline some of the following:
- Test planning
- Data analysis
- Document harmonization and accelerated internal reviews
- Submission preparation
2. Portfolio-level consistency
AI can act as a “knowledge keeper” and provide support for some of the following:
- Preserving organizational wisdom
- Supporting onboarding
- Maintaining unified processes
3. Smarter submissions and faster responses
Regulators and Product Developers are actively using AI which provides an opportunity for faster, more direct communication. However, caution has to be taken when using AI for regulatory submissions, since unchecked documents full of errors will not only get rejected, but may also increase the scrutiny of regulatory bodies on any future submission.
4. Richer insights across the product lifecycle
AI will help make closing the development loop easier and safer. As it becomes more commonly implemented, AI can help bridge data across Design, Development, Testing Manufacturing, and Post-market surveillance.
5. A sustainability challenge. AI is incredibly energy-intersive
While AI is a revolutionary tool, the data centers that it requires consume city-level amounts of electricity, potentially undermining the industry’s sustainability gains.
The future must balance speed, quality, cost and energy requirements in order to achieve a sustainable way to implement AI.
Final Takeaway
AI is reshaping design controls in ways that are both practical today and transformative tomorrow. As documentation grows more complex and global development accelerates, AI can reduce documentation burden, streamline verification, surface patterns in large datasets, and strengthen the overall design control process.
The future will favor organizations that adopt AI intentionally. Responsible implementation requires structured inputs, human oversight, and a commitment to validating everything the tool produces. Teams that treat AI as a disciplined partner, NOT a decision-maker, will achieve faster development cycles, clearer submissions, and stronger product reliability.
In the end, AI will not eliminate the need for expert engineers. Instead, it will expand their impact. Those who learn to integrate AI thoughtfully will set a new standard for quality, efficiency, and regulatory confidence across the entire product lifecycle.
