By Diwei Zhou, Professor in Statistics, Loughborough University

Artificial Intelligence (AI) is increasingly being used to support decision-making across business, industry, government and academia. From predicting infrastructure failures and optimising transport systems to supporting healthcare planning and informing public policy, AI has become a powerful tool for turning data into actionable insights.

As researchers, we often focus on developing models that are more accurate, more sophisticated and more powerful. However, through my work with external partners, I have come to realise that technical excellence alone is not enough. A highly accurate model can still fail to create impact if decision-makers do not understand what the results mean, how reliable they are, or how they should be used.

This challenge motivated me to write A Practical Guide to Explaining AI Results in Knowledge Exchange, recently published by the UK Knowledge Exchange Hub for Mathematical Sciences (KE Hub) with support from the Engineering and Physical Sciences Research Council (EPSRC). The guide was developed to help researchers, businesses, policymakers and knowledge exchange professionals communicate AI results more clearly, responsibly and effectively.

Over the past few years, I have had the privilege of working with a wide range of organisations through research collaborations, consultancy projects, Knowledge Transfer Partnerships, Innovate UK programmes and national knowledge exchange activities. Across these different settings, I repeatedly encountered a common challenge. Organisations were often less concerned about how an AI model worked and more interested in understanding what the results meant for their decisions.

Questions such as “Can we trust this prediction?”, “How certain is the result?”, and “Why has the model reached this conclusion?” frequently arise when AI is introduced into real-world environments. These are entirely reasonable questions, yet they are often overlooked when communicating AI outputs.

As highlighted in the guide, many AI projects do not struggle because of poor model performance. Instead, they struggle because the results are not explained in a way that supports understanding and decision-making. In some cases, good models are rejected because stakeholders cannot justify their use. In others, AI outputs are trusted too readily because uncertainty and limitations have not been communicated clearly.

One of the key messages of the guide is that accuracy, usefulness and understandability are not the same thing. An accurate model is important, but impact is achieved when AI results are also useful for decision-making and understandable to the people who need to act on them.

To address this challenge, the guide introduces a simple four-step framework for explaining AI results:

• Context – What decision is being supported?
• Result – What does the model output mean in plain language?
• Reason – Why has the model produced this result?
• Uncertainty and Limitations – What should users be cautious about?

The framework is deliberately practical and designed to be applicable across a wide range of AI applications. It encourages researchers to focus not only on model performance but also on interpretation, transparency and responsible communication.

The guide also emphasises the importance of communicating uncertainty. In my experience, decision-makers rarely expect certainty. What they need is a clear understanding of what an AI result means, how much confidence can be placed in it, and where caution is required. Openly discussing uncertainty does not undermine trust; rather, it helps build it.

I am grateful to the KE Hub and EPSRC for supporting this work. Through the KE Hub, I have had the opportunity to engage with academics, businesses, industry and government partners across diverse sectors. These interactions have provided invaluable insights into the challenges of translating AI research into real-world impact and directly informed the development of this guide.

As AI continues to become embedded in decision-making processes, the ability to explain AI results clearly and responsibly will become increasingly important. Responsible AI is not only about developing robust models; it is also about ensuring that their outputs can be understood, challenged and used appropriately.

I hope this guide helps researchers and external partners bridge the gap between technical innovation and practical impact, enabling AI to support better decisions, stronger collaborations and more sustainable innovation.

Further details about the guide can be found in the KE Hub news article: New practical guide helps bridge the gap between AI research and real-world decision-making. The guide is freely available to download from the KE Hub website.

Open Research comment: Zhou’s guide is a great example where Openness increases the likelihood of real world impact. A paywall would have presented a significant barrier to the audience they would be trying to reach. By offering a practical guide, they’re making the knowledge more accessible to readers beyond academia. All of which contributes to making this world a little better 🙂 – Lara Skelly, Open Research Manager for Data & Methods, Loughborough University.