This case study is published under the InnovateUK BridgeAI - Bespoke AI and Data Science Advice for SMEs offer. The Alan Turing Institute Independent Scientific Advisors (ISAs) offer transformative support to SMEs across BridgeAI sectors, enabling them to harness AI for both practical and strategic benefits. This initiative is supported by Innovate UK BridgeAI. We thank Paul Coker, CEO, and Rebecca Cole-Coker, Creative Director at FarmSmarter, as well as Dr. Po Yang, Independent Scientific Advisor for BridgeAI based at The Alan Turing Institute, and Megan Yates, Impact Evidence and Evaluation Manager at Hartree Centre for their significant contributions to the development of this case study. Special thanks to Alexandra Araujo Alvarez, Senior Research Community Manager for BridgeAI, Dominica D'Arcangelo, Programme Manager, and Punita Maisuria, Project Coordinator, for their leadership and support. We also acknowledge Stuart Gillespie for his role as the technical writer for this and other case studies in the programme. Additional thanks go to Aida Mehonic, Principal Researcher for Research Applications, and Shakir Laher from The Alan Turing Institute for their valuable reviews and feedback. This work is led by Dr. Vera Matser, Head of Skills and Principal Investigator for BridgeAI at The Alan Turing Institute. For any comments, questions, or collaboration opportunities with BridgeAI, please email: bridgeAI@turing.ac.uk.

FarmSmarter Improving decision-making in farmingWhether for smallholder farms or large-scale agribusinesses, AI is becoming a critical ally in managing resources, improving crop yields and controlling pests. FarmSmarter’s free, easy-to-use smartphone app wasoriginally designed for use in developing countries. It uses AI models to generate advice for farmers, overcoming barriers such as lack of access to specialist equipment or lower levels of literacy. The app supports smallholder farmers by giving them accurate, real-time information and advice, including weather forecasts, field mapping, crop health reports and yield predictions – there’s a huge appetite for this type of information.To do this effectively, efficient machine learning models are essential, yet making sense of the vast number of data points collected is often difficult, such as when sampling images of crops to detect disease. The answer is to combine two techniques: automated deep learning and human- supervised active learning