Why do so many AI startups struggle with regulatory strategy?
Whether you like it or not, AI has made its way into healthcare. Both the MHRA’s and the NHS’s recent confirmation that high-functionality ambient scribe tools are medical devices highlights just how embedded these technologies are becoming. From scribes and image analysis to triage, diagnosis, and treatment planning, the technical capability is already there. What’s less clear is a credible plan for how these tools will enter regulated clinical markets - and remain there.
A familiar story we hear from healthcare AI companies is that regulatory strategy gets pushed down the to-do list. The focus tends to be on developing the product, securing funding, and getting something in front of users. By the time regulation becomes a priority, delays have set in, requirements are more complex than expected, and the product needs reworking just to meet baseline standards for clinical use.
So, where do things typically go wrong?
Even with strong technical teams and promising use-cases, many AI companies run into the same set of challenges when it comes to regulation. Some of these issues are structural. Others come down to timing, assumptions, or gaps in experience. Here are five of the most common patterns we see when supporting teams through product development, clinical validation, and market entry.
1. Regulation comes after the product is built
Many startups begin with a working tool, a few pilot sites, and a broad plan for reaching the market. Regulation is frequently pushed aside in favour of getting the product up and running. The thinking is that once the algorithm is built and showing some early promise, regulatory slack can be picked up later.
But by the time regulatory input is brought in, many of the most important decisions have already been made. The intended use is fixed, the product’s functionality is locked, and clinical workflows have already been shaped. Retrofitting regulatory requirements onto a nearly finished product often means revisiting core assumptions, delaying launches, and repeating work that could have been done once, properly, from the beginning.
A more effective approach is to involve regulatory thinking from the start. This allows teams to plan the route properly and shape the product with those requirements in mind. It makes approval more straightforward and reduces the risk of needing major changes further down the line.
In this space, regulation and product are fundamentally inseparable. Prioritising one over the other just doesn’t work.
2. Building for healthcare is often misunderstood
In tech, teams are often used to moving fast, testing products in the wild, and learning through iteration.
In healthcare, the bar is different. Products designed for clinical use must meet clear, documented standards. Regulators and health systems need to see that the tool is safe, that it performs as claimed, and that it fits within existing care pathways. These requirements exist for a reason. Clinical tools are used in high-stakes environments where patient outcomes are directly affected. As a result, the level of residual risk regulators are willing to accept is much lower than in typical software markets.
Importantly, this is not to say that regulators and healthcare professionals don’t accept any risk. In fact, there are countless high-risk devices on the market. But to mitigate that high risk, regulators expect strong safety features, robust testing, and meaningful clinical benefits.
Startups that don’t recognise this early can end up rebuilding key parts of their product or running into resistance from clinicians who are rightly cautious about integrating new tools into patient care.
3. The regulatory framework for AI is still developing
Regulators will always be on the back foot of innovation. In medical devices, this means that a lot of AI-related guidance is currently lacking. While progress is being made, frameworks for machine learning tools are still incomplete, particularly for LLMs, adaptive models and explainable AI.
For companies building in this space, that creates an added layer of complexity, because it’s hard to fit new technologies into old regulatory frameworks. .The best way to manage this is to think from first principles.. Even if new regulatory guidance is released for AI medical devices, the principles of regulation will remain the same: 1) prove your product delivers a meaningful clinical benefit, 2) show that you have mitigated risk as much as reasonably possible, 3) show that your product is consistent and high quality, and 4) show that your product is cybersecure.
That means documenting how the model is trained and updated, having clear change control procedures in place, and preparing to generate additional evidence if needed. This not only helps with approval but also builds trust with clinicians and users who need to understand how the tool works and why they can rely on it.
4. Regulatory expertise is not part of the core team
Regulatory consultants bring valuable expertise, but they are often brought in too late and asked to work around decisions that have already been made. This limits their ability to influence product design or evidence strategy in a meaningful way.
A more effective approach is to bring regulatory thinking into the core team. That does not always mean hiring a full-time specialist from day one, but it does mean involving someone who understands the regulatory landscape in early planning and decision-making..
When regulation is treated as a strategic function rather than a compliance task, it can help shape everything from how the product is positioned to how trials are designed and which claims can be supported. It also makes conversations with potential partners and investors easier, as there’s more confidence in the product’s path to approval.
5. Global growth plans are not matched by regulatory readiness
Many startups are built with global reach in mind. The goal is often to start in one region and expand quickly into others. But regulatory systems are not harmonised, and approval in one region rarely transfers smoothly to another.
A CE mark, for example, might be enough to enter parts of Europe and the UAE, but it does not open access to the United States. An FDA clearance will not automatically apply in Latin America or Asia. Each region has its own definitions, processes, and expectations.
Companies that succeed are the ones that plan ahead. They map out their priority markets, understand what each one requires, and design their clinical evidence strategy accordingly. They also build quality systems that support this kind of scale, with the ability to manage updates and variations across markets.
At Hardian Health, we work with AI and digital health companies to develop strategies across clinical development & scientific evidence, regulation, health economics, and intellectual property. That often starts with regulation, but it doesn’t end there. We help teams map their route to market, avoid common causes of delays, and make evidence-led decisions that support long-term use in healthcare systems.
We also work in discrete stages, with options to offload work to us, or for us to take more of a mentoring role. However you want to approach it, we can work flexibly to your needs, as we have done with dozens of others who’ve successfully made it to market.
Hardian Health is a clinical digital consultancy focused on leveraging technology into healthcare markets through clinical strategy, scientific validation, regulation, health economics and intellectual property.
