TTP hosted a fringe event during Cambridge Tech Week 2025 – “How will AI transform innovation in physical products from advanced therapies to autonomous UAVs?”. We convened a cross-industry group of technologists and business leaders to examine how artificial intelligence (AI) is revolutionising the innovation and development of physical products, from advanced therapies in healthcare to autonomous unmanned aerial vehicles (UAVs). It centred around expert speakers from industry, including Uday Phadke (Triple Chasm), Vincenzo Di Cerbo (Cell and Gene Therapy Catapult), and Anthony Spouncer (Viasat). Our event provided a platform for deep discussion of the practical, technical, and organisational challenges facing those striving to embed AI meaningfully in new physical products.
Integrating AI into real-world systems
The session combined a keynote, panel discussion, and networking to examine how AI is transforming the development of physical products—and the persistent barriers that prevent many from reaching commercial maturity. Discussion focused on the interplay between technical feasibility, organisational capability, and regulatory compliance across sectors such as healthcare, life sciences, defence, and aerospace.
In his keynote, Uday Phadke analysed why almost 90% of products fail to transition from prototype to market, highlighting the importance of balancing innovation with focus on technical, organisational and commercial needs. He introduced a structured framework for integrating AI into product development, encouraging end-to-end alignment between technical opportunity and market value.
Uday then joined Vincenzo Di Cerbo and Anthony Spouncer for a panel that extended these ideas through applied case studies. The discussion moved from strategic alignment to implementation detail, highlighting both sector agnostic and sector specific learnings, as summarised in the sections below.
Core takeaways & implications
Value over hype
Across all sectors represented, a universal theme emerged: the necessity to focus resources on high-value AI use cases, not chasing trends for their own sake. Attendees were cautioned against being swayed by industry hype cycles. Instead, leaders were encouraged to:
- Clearly identify pain points or process improvements where AI delivers demonstrable outcomes
- Prioritise investments based on proven or rigorous potential to generate economic or operational value, rather than hype or general optimism
This principle is vital not only from a risk-management perspective but also in response to broader market trends, such as the large share of venture capital chasing AI opportunities. This was reported as now topping 40% (at the expense of other high valuable areas of deep tech), and the societal risks of resource misallocation, including environmental impact from data centre proliferation.
Safety, provenance, and regulation
The consensus among speakers and the audience was that the success of AI in physical products, particularly in regulated markets, is fundamentally reliant on achieving trust. This is not just amongst end users, but also the buyers of AI and therefore relies on foundational attention to safety, data provenance, and contributing to evolving regulatory landscapes.
- Certification, traceability, and explainability need to be designed into products from day one. This extends to the concept of a traceable “AI Bill of Materials,” reflecting not just software components but also the origin and handling of all data used to train, validate, and operate AI models
- Data and model provenance are crucial for version control, auditability, and for countering risks such as data poisoning
- Designing for trust is a core requirement, since both market acceptance and regulator confidence depend on visible, documented control over AI decisions and data integrity
Near term wins at the edge and the importance of data relevance over data volume
Edge computing and strategic data collection were highlighted as areas of near-term AI wins. Rather than indiscriminate data gathering, organisations were urged to:
- Focus on collecting only actionable, relevant data that supports high-value insights, especially in bandwidth-constrained or latency-sensitive environments such as autonomous UAVs
- Design data pipelines from sensors to AI models that are both secure and cost-effective, thus supporting deployable real-world systems. Creative approaches, like using multiple inexpensive sensors to gain richer insights rather than relying on fewer high-end sensors, were highlighted
- Optimise AI models to operate efficiently at the edge were relevant, improving speed and reducing the need to send data to the cloud, and with it, reducing compute and communication costs
Sector-Specific Summaries
Advanced therapy manufacturing (healthcare & life sciences)
AI holds the promise of transforming advanced therapy manufacturing, a field plagued by bottlenecks, capacity constraints and overall complexity. Parallels were drawn with the relative complexity of advance therapies compared with manufacturing of small molecules of building an aircraft carrier rather than a bicycle. Current limitations mean only 2% of eligible advanced therapy patients can access necessary treatments, with therapy manufacturing failures and the consequential delays in treatment aggravating patient risk.
Key AI opportunities and best practices include:
- Prioritising time–to–release by accelerating quality control (QC) and process analytics using real-time AI-based insights, rather than relying solely on lengthy offline analysis
- Implementing AI for QC at critical process checkpoints, directly impacting the timeliness and reliability of therapy release
- Supporting regulators by contributing to frameworks that balance the need for rapid QC (to get therapies to patients faster) and the requirement for thorough safety validation
- Focusing on discovering new, actionable biomarkers and hence carefully selecting data parameters to reduce the cost and improve the quality of data needed for high-impact AI applications
- Exploring AI-driven process design (digital twins) that will in time enable real-time, adaptive control to boost manufacturing throughput and reduce failure rates
Aviation & unmanned aerial vehicles (UAVs)
AI’s role in aviation, and specifically in UAVs, is rapidly expanding, carrying major implications for operational efficiency, safety, and regulatory or airspace management. The growing volume of autonomous and semi-autonomous UAVs will soon exceed the managing capacity of human pilots and air traffic controllers.
Panellists discussed:
- Embedding safety and regulatory compliance at the initial design stage, accounting for issues like trusted data sources, resilient communications, and the latency/consequences of on-board versus cloud-based AI
- AI’s utility for predictive maintenance, utilising real-time flight data to identify potential equipment failure (even before standard scheduled maintenance could reveal a problem)
- Evaluating which mission-critical processing must occur directly on the UAV (at the "edge") to ensure both operational autonomy and real-time responses, while remaining mindful of SWaPC (size, weight, power and cost) constraints
- Foreseeing regulatory challenges as UAV use becomes pervasive, with AI needed to support not just individual missions, but broader fleet operations and airspace management (e.g. AI-driven traffic control scenarios)
Practical framework for driving AI innovation in physical products
For Strategy, Product Management, and Development Leaders:
- Clarify Outcomes and Understand Value: Begin with an understanding of the value AI must deliver to the business and end-user. Structure pilots and implementations so that learning and impact are measurable wherever possible
- Embed Trust and Safety: From initial design, include robust approaches for logging, traceability, and model/data provenance. Participate actively in regulatory discussions so contributions shape, not just react to, future compliance norms
- Optimise Data Strategy: Treat the gathering and use of data as a key design criterion. Be ruthless in filtering for actionable signals and avoid overwhelming teams (and cloud infrastructure) with non-essential data
- Build for the Edge: Even if deploying to the edge is not an immediate requirement, consider designing products as if it might be—including lightweight models and modular sensor systems—to futureproof solutions and control costs
Cross-sector lessons and overarching themes
- The next era of physical product innovation will be defined less by generic “AI capability” and more by the thoughtful, sector-specific deployment of AI to address impactful, measurable problems
- Interdisciplinary collaboration, combining technical vision (from multiple science and engineering disciplines), commercial realism, and regulatory expertise, is essential for reducing the high product failure rate cited at the event (90% not reaching full commercialisation)
- AI’s effect on resource allocation (venture capital, operational budgets, and energy/environmental impact) heightens the need for disciplined program management and value-led development
Conclusion -the road ahead for AI in physical products
Our TTP event at Cambridge Tech Week 2025 underscored that while AI is a catalyst for extraordinary transformation in physical products, its success depends on marrying ambition with discipline. Future market leaders will be those who follow value, not hype, equip their teams to address safety, quality and regulatory needs from the start, and are relentless in focusing on relevant data engineering and edge deployment strategies where needed.
For those shaping the future of AI in physical products, these insights provide an indication of the areas to be considered when moving from exciting prototypes to commercial reality in some of the world’s most demanding markets.
