On average, a new or revised national regulatory document relevant to medtech is published every 53 minutes worldwide. ¹ This frequency transforms staying up to date from a discussion point into a staffing challenge that traditional organizational structures are not equipped to manage.
This is not a knowledge gap. The real issue is a throughput gap. It is impossible to keep up with this demand without solving the throughput challenges.
The math that stopped working
Two realities exist simultaneously, yet most regulatory strategies address only one.
The volume of documentation teams must track has grown 75% in the past five years, and 70% of global requirements are now published in a language other than English. ¹
At the same time, the pool of experienced medical device regulatory professionals remains small in the markets where this kind of expertise is hardest to scale quickly, and not coincidentally where much of the industry’s regulatory complexity is also concentrated.
Costs have risen alongside these trends. From product concept to achieving 510(k) clearance now averages nearly $31 million, with roughly $24 million attributed to obtaining clinical data and the regulatory clearance it unlocks. Class III devices average $75 million; a 64% increase over eleven years. ² MDR and IVDR have pushed clinical evaluation costs in Europe up by as much as tenfold in some cases.
Hiring alone cannot close this gap. More people will not fix a process that cannot keep pace with regulation that changes constantly and grows more complex by the year; closing it takes streamlined workflows and tools capable of expanding what a team can deliver without sacrificing quality. Finally, next to the often unacceptable impact on the bottom line, significant hiring also increases organizational complexity.
In an industry survey on regulatory compliance confidence, only 28% of individual contributors reported feeling very knowledgeable about current regulatory changes, compared to 44% of management and 56% of senior executives. ³ The confidence gap closely aligns with areas where the workload is most concentrated.
This is not a matter of capability. Seniority does typically bring more expertise, but in this context, workload and complexity are the dominant factors: those closest to submission work carry documentation demands that have outgrown what training or onboarding alone can address. The employees on the regulatory shop floor, so to speak, simply cannot read and absorb all that information, turn it into work instructions and apply them before the next change comes out.
Why “Just Adopt AI” is not the solution, and what works
It should come as no surprise to anyone that dedicated regulatory AI tools are ideally positioned to translate this information volume into actionable materials. This does not mean the solution is simply to add AI and move on. AI tools are effective if they are designed with the application in mind. General-purpose AI is smart, like a college graduate, but in real life we don’t give tasks to people just because they are smart (flying a plane, designing a house), but because they are qualified.
“Start with a business problem, then decide whether AI is the right solution” is a more disciplined standard than most organizations currently apply.⁵ It’s what Michael Konings, Senior Director of Regulatory Affairs at Philips Healthcare, put more directly in Raiana’s Augmented Professional webinar: “I always believe that AI needs to be used in a problem-solving way, a strategic way, versus just wanting to use AI tools.”⁵ The distinction matters, it puts the burden of proof on the tool, not on the team asking for evidence that it works.
This approach differs fundamentally from adopting AI due to its popularity. One method addresses a specific capacity problem, while the other focuses on appearances. The distinction is evident in how success is measured. Tools acquired for appearances are evaluated by usage metrics such as logins and adoption rates. Tools intended to close a capacity gap are assessed by their ability to deliver defensible, accurate, audit-ready output more efficiently, and/or with higher quality, than manual processes. Only the latter provides meaningful long-term value.
Where capacity is already coming back and where it’s headed next
AI, when applied to well-defined regulatory tasks, has already demonstrated its ability to increase capacity. Not by replacing professional judgment, but by eliminating tasks that do not require it. Konings anticipates this effectiveness will increase through 2026 as both the tools and associated practices mature. Four areas illustrate this progression:
- Regulatory intelligence is starting to enable tools to directly identify which products are affected by a given change and what to prioritize, shifting the task from research to triage.
- Compliance alerts apply the same logic across FDA, national authority databases, and notified body trends, linking signals to actual submission history rather than relying on someone to catch the connection manually under deadline pressure.
- Drafting documentation has often disappointed companies, because they expected a finished, expert-level document that required no review. Done right, it produces a strong first draft, structured content reuse, and a traceable path from evidence to conclusion, with review built in rather than bolted on. As Konings puts it: “I would review it with a lens of an AI, of a junior, or an expert level, to make sure the document is end-to-end safe.”
- Translation addresses a genuine bottleneck: with 70% of new requirements no longer published in English, the obstacle isn’t only regulatory expertise; it’s language.
None of these four eliminates the need for professional oversight. They automate the parts of the work that never required specialized judgment to begin with, which is exactly what frees up time for the parts that do.
There’s a fifth capability worth naming separately, because it isn’t something generic AI tooling handles well. It’s where purpose-built regulatory intelligence earns its distinction.
When a standard is revised, there’s often no clean redline. Someone has to manually compare versions of the document across every affected product to find out what changed. A tool built specifically for regulatory content, with the domain curation to know what a substantive change looks like, not just a general-purpose model pattern-matching text, can compare both versions, isolate the changes that matter, and produce a transparent, reviewable record of how it got there. That combination of speed and an auditable trail is the bar in a field where “the AI said so” is never an acceptable answer.
The review is the design, not an afterthought
“Human in the loop” is not a disclaimer added to an AI proposal to address liability. When implemented correctly, it forms the core structure of the process.
Accountability remains with the human professional. AI manages drafting, structuring, flagging, and summarizing, while a person reviews, validates, approves, and assumes responsibility for the final output. This division of labor is not a temporary measure; it represents the expected structure of the role moving forward.
This clarifies a common misconception about “augmentation.” The distinction is not that machines handle the easy 80% and humans with the difficult 20%, but rather that machines process data at a scale and speed unattainable by teams, allowing professionals to focus on their judgment where it is most valuable.
That brings us back to the point of quality. Saying that AI produces higher quality than humans is considered blasphemy. It is also the wrong statement. AI-assisted work, on the other hand, can produce much higher quality in the same or even less time.
Think about it: a regulatory professional gets a new guidance document to read, understand and apply. Just reading it verbatim may take a couple of hours, so they resort to scanning. Implementing the guidance, same thing. AI, on the other hand, will tirelessly “read”, analyze and apply everything in the guidance. The hours saved by the human can then be used to understand the guidance and go back to the text in a targeted fashion, applying what is needed for the product portfolio.
This also explains why generic AI tools often fall short: without regulatory-specific training, curation, and safeguards, they may generate plausible text without understanding what constitutes a substantive change or a defensible document. A domain-specific tool, reviewed by a knowledgeable professional, is fundamentally different from a general assistant operating in a regulated environment.
Trust in any tool develops gradually, as it should. Begin with lower-risk tasks, ensuring a clear understanding of the tool’s capabilities and limitations. Expand delegation only as confidence increases, maintaining an ongoing relationship with the tool developers rather than relying on a single evaluation.
Perhaps the last point of attention is over-reliance. Will we be able to do our jobs without AI in a couple of years? Just like with satellite navigation in the car, the answer is probably no: we will be somewhat helpless without this technology. Does anyone still drive with a paper map on their steering wheel? Is it even allowed to drive while so distracted?
But that’s ok. Technology has made us more effective ever since the steam engine and no, we cannot do without AI anymore soon. The key is, just like you don’t honor the satnav’s wish to ‘turn right’ when you’re in the middle of a bridge, to blindly trust. Keep your judgment.
The 53 minutes don’t stop for anyone
That opening number, a new document somewhere in the world roughly every 53 minutes, isn’t going to slow down because a team is understaffed or exhausted. The regulatory landscape doesn’t wait for capacity to catch up.
Whether this pace becomes a crisis or a manageable workflow depends not on the abstract adoption of AI, but on whether the chosen tool effectively addresses the capacity gap or merely adds another interface. A generic assistant that drafts without regulatory expertise can increase the review burden rather than reduce it. A domain-specific tool, used as Koningsdescribes reviewed, clearly defined, and trusted incrementally; enables efficient triage rather than starting from scratch with each document.
Organizations relying solely on increasing headcount are working against an accelerating timeline. Those successfully closing the gap are redesigning workflows and making deliberate choices about which tools to implement.
References
¹ IQVIA Regulatory Intelligence, Medical Devices & IVDs, May 2024, as presented by Michael Konings (Philips Healthcare) in Raiana’s Augmented Professional webinar.
² Starfish Medical, “How Much Does It Cost to Develop a Medical Device?” (2020, updated Jan. 2025).
³ Greenlight Guru, “2024 State of the MedTech Industry Report.”
⁴ Michael Konings, Philips Healthcare, The Augmented Professional, Ep. 2 (Raiana webinar).
⁵ Paraphrase reflecting the principle Konings articulated in the webinar; the bracketed quote is his verbatim language, not the earlier paraphrase presented as a direct quote.
