Tag: research and development

  • How AI is reshaping innovation ecosystems and speed to market

    How AI is reshaping innovation ecosystems and speed to market

    Artificial intelligence is changing how new products, services and scientific discoveries emerge. It can search possibilities at enormous scale, suggest unfamiliar combinations and help people test ideas more efficiently. For innovation ecosystems, the opportunity is to connect these capabilities with specialist knowledge and customer needs. The resulting value depends on what organisations can discover and deliver together.

    An innovation ecosystem includes businesses, universities, investors, public institutions and the relationships between them. AI can strengthen those relationships by making knowledge easier to access and experiments easier to organise. Different technologies contribute differently: predictive models estimate promising outcomes, generative systems propose alternatives, and automated laboratories connect computational recommendations with physical tests.

    AlphaFold illustrates the importance of shared knowledge. Developed by Google DeepMind and EMBL’s European Bioinformatics Institute, its database provides open access to hundreds of millions of protein structure predictions. A February 2026 update incorporated specialist community datasets, including contributions from researchers at Edinburgh and Glasgow. AI capability is therefore becoming infrastructure that other researchers can extend and apply to their own questions. EMBL-EBI

    For entrepreneurs, such resources can reduce the work required to investigate a biological problem. A new venture could combine public predictions with specialist experimental expertise and customer knowledge to explore a particular application. Commercial opportunities emerge around the shared resource, including testing services and specialist analytical tools. Experimental validation remains essential when assessing whether a prediction supports a useful product.

    Materials discovery offers another example. Microsoft and Pacific Northwest National Laboratory combined machine learning with physics models to narrow more than 32 million potential battery materials to 18 promising candidates. Researchers then synthesised and characterised selected materials. PNNL described commercial suitability as requiring further investigation. The achievement demonstrates how computational screening can direct scarce laboratory resources towards a manageable set of possibilities. PNNL

    The University of Manchester and Unilever have taken a related approach to chemical process development. Their laboratory uses AI informed by physics to select valuable experiments and learn from their results. This creates a continuing relationship between prediction and measurement. Each experiment can improve the model used to choose the next, helping researchers explore chemical processes more systematically. manchester.ac.uk

    These examples suggest a shift in how research partnerships should be organised. Businesses need access to useful data, suitable models and facilities where recommendations can be tested. Universities and technology providers bring complementary expertise. Agreeing how experimental results will be recorded and shared becomes part of the innovation process, because those results inform subsequent decisions.

    AI can also help people combine commercial and technical thinking. A 2026 study in Organization Science reported an experiment involving 791 Procter & Gamble professionals working on product innovation challenges. Individuals using AI matched the performance of teams without it, while AI users produced proposals balancing technical and commercial perspectives. These findings concern an early stage development task, rather than evidence of successful product launches. Organization Science

    For smaller businesses, the ecosystem can provide expertise they could struggle to develop independently. InSignLanguage worked with the STFC Hartree Centre through Innovate UK BridgeAI to investigate British Sign Language to English translation. Support worth £15,000 enabled technical assessment and a proof of concept classifying basic hand gestures. This was an initial demonstration towards a broader accessibility service, with substantial development still ahead. Innovate UK Business Connect

    The entrepreneurial lesson is to begin with a defined unmet need. In this case, the ambition concerns communication for sign language users in public transport. The AI model contributes one capability within a larger service. User involvement, practical integration and a viable commercial arrangement would determine whether that capability becomes something people can depend on.

    New business models may follow from these connections. A specialist startup could provide discovery as a service, helping manufacturers identify promising materials without each customer building a research platform. A university laboratory could offer shared experimentation facilities to several firms. Regional innovation hubs could organise common datasets around a sector’s recurring problems. These are opportunities to investigate through customer demand, with clear agreements on access, ownership and how participating organisations share the benefits.

    AI also changes what people need to learn. Universities should give students opportunities to work with businesses on real problems, compare model outputs with evidence and explain their decisions. Employability increasingly involves combining disciplinary understanding with the ability to question computational results. Employers can support this through placements and collaborative projects that develop judgement alongside technical familiarity.

    Creative diversity deserves attention. A study by Anil Doshi and Oliver Hauser found that access to AI ideas improved evaluations of individual short stories, while the stories became more similar to one another. The experiment concerned writing; wider industrial implications require testing. For innovation teams, it provides a reason to preserve independent thinking and actively seek different customer and specialist perspectives. PubMed

    UK businesses can make a practical start by selecting one expensive uncertainty: which formulation to test, which customer problem to investigate, or which design to prototype. Assemble the relevant expertise, establish the current performance baseline and run a bounded trial. Compare the AI supported approach with existing practice, including the costs of verification and implementation.

    Measure progress through the quality of decisions and what follows from them. Useful indicators include the cost of a validated experiment, the proportion of concepts reaching customer trials and revenue from resulting products. Leadership should fund the route from an interesting output to a working solution. That is where AI can help an innovation ecosystem turn greater computational capability into practical economic and social value.

    References

    1. EMBL-EBI (2026). AlphaFold Database welcomes community datasets, 17 February.
    2. Pacific Northwest National Laboratory (2024). From Simulation to Synthesis, 13 September.
    3. University of Manchester (2025). AI-powered ‘self-driving’ labs accelerating chemical process innovation, 18 December.
    4. Dell’Acqua, F. et al. (2026). The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork. Organization Science.
    5. Innovate UK Business Connect (2026). Improving the accessibility of public transport with AI, 15 January.
    6. Doshi, A. R. and Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290.

    Download the editable Word version.

  • The Missing Middle of Innovation: Why Promising Ideas Die Between TRL 2 and TRL 4

    The Missing Middle of Innovation: Why Promising Ideas Die Between TRL 2 and TRL 4

    Innovation rarely fails because people lack ideas. It fails because promising ideas struggle to cross the space between an interesting concept and credible evidence that it works.

    This is the missing middle of innovation: the journey from Technology Readiness Level 2, where a technological concept has been formulated, through TRL 3, where proof of concept is established, to TRL 4, where the technology has been validated under laboratory conditions.

    It sounds like a short step. In practice, it is where many potentially valuable innovations die.

    The evidence gap

    At TRL 2, an innovator may have a strong hypothesis, early research and a convincing description of the problem. What they lack is sufficient evidence. Customers cannot see a working solution. Investors cannot assess whether the principal technical risks have been reduced. Funders cannot distinguish between a bold proposition and an unsupported claim.

    The innovator therefore faces a paradox: they need resources to generate evidence, but they need evidence to secure resources.

    This is often described as a funding gap, but money is only part of the problem. There is also a capability gap. Moving towards TRL 4 requires experimental design, systems engineering, customer discovery, project governance and commercial judgement. The creator may not possess all those capabilities.

    Activity is not the same as progress

    Many early projects appear busy without becoming more investable. Teams build features before confirming the core technical hypothesis. They commission prototypes that demonstrate appearance rather than function. They speak to supportive contacts instead of prospective buyers. They report tasks completed rather than uncertainties reduced.

    This produces the illusion of progress.

    A credible TRL 3 project should be organised around explicit questions. What must be true for the technology to work? Which assumption presents the greatest risk? What experiment would test it? What counts as success or failure?

    TRL progression is not simply a matter of spending time on development. It is the disciplined reduction of uncertainty.

    Technology cannot progress alone

    Another mistake is treating readiness as purely technical. A proof of concept can work while the wider proposition remains unviable. The customer may not consider the problem important. Integration may be too complex. Data may be inaccessible. Security, regulation or operating costs may prevent adoption.

    Technical evidence must develop alongside market and delivery evidence. By TRL 4, a team should understand whether the technology can work, who might buy it, why they would act and what barriers stand in the way.

    Building a bridge across the missing middle

    Promising ideas need a structured evidence plan, not simply encouragement. This should define the critical hypotheses, experiments, acceptance criteria, responsibilities, costs and evidence needed for the next investment decision. Each activity should answer a material question or retire a significant risk.

    Universities, accelerators and innovation agencies also need to rethink their support. Too many programmes reward pitching, generic mentoring and participation. Innovators need technical facilities, product leadership, relevant customers, commercial expertise and patient funding tied to learning milestones.

    The objective should not be to protect every idea. Some should stop. A well-designed experiment that disproves a concept is more valuable than months spent sustaining false optimism.

    The real purpose of the missing middle is selection through evidence. Ideas that survive emerge stronger, more focused and more credible. Those that do not release their people, time and capital for better opportunities.

    Innovation policy often celebrates discovery and later-stage scale-up. Between them lies the less glamorous work of proving, testing and learning. Until we invest properly in that work, too many good ideas will continue to disappear before they are given a fair chance to succeed.