Algorithmic Assist Project
The Algorithmic Assist project explored how Emergency Medical Retrieval and Transfer Service (EMRTS) staff make fast, high-stakes decisions on whether to dispatch specialist pre-hospital critical care teams to people in urgent need across Wales. These decisions are made at the central EMRTS Critical Care Hub (ECCH), where two experienced staff members monitor all emergency calls for the entire country. Their job is to identify the rare and specific cases that would benefit from pre-hospital critical care, separate from the standard response provided by the Welsh Ambulance Service University NHS Trust (WASUNT).
The ECCH dispatch process is complex and involves three main decisions: whether to examine a call record in more detail, whether to seek more information (such as by listening in on a call or calling back to scene), and whether to dispatch a team. Most calls (more than 99%) do not result in dispatch, so the staff must rapidly and accurately filter information, while handling many calls and frequent interruptions.
The research goal was to understand this real-world decision-making process in detail and explore future tools – like AI or algorithm-based decision support – and how they can be designed to genuinely help rather than hinder. So as not to rely only on written procedures or interviews, the researcher spent over 50 hours observing how the work is done during real shifts, including both day and night hours.
Privacy note:
Strict ethical guidelines were followed to protect patient confidentiality1. The researcher did not have access to patient names or identifiable medical information. Observations were made from a distance that prevented viewing computer screens, and no recordings were made—only pen-and-paper notes were taken. In some cases, the researcher may have overheard staff talking about particular incidents or patients. However, these conversations were usually brief and focused on practical decisions. Any details overheard were not linked to personal identities, and the researcher did not access full case records.
What was observed was the flow of work: how staff move between calls, talk to one another, manage multiple information systems, and maintain constant awareness of unfolding events. One key finding was that staff often share updates without being asked—demonstrating strong team awareness and coordination. They constantly monitor changing information and anticipate each other’s needs, which helps them make quick decisions under pressure.
Attempts to help this process with automated flagging systems in the past have failed, often producing too many false alerts. This study shows that for AI tools to be useful in this setting, they must fit seamlessly into the human workflow, respect the experience of staff, and support—rather than distract from—their judgement.
The experimental work in this research aimed to explore how algorithmic tools could support the decision-making process at the critical care hub. Building on detailed observations of how dispatch decisions are made at the ECCH, the researcher collaborated with clinical staff to design and test prototype decision-support systems. These systems were developed with a deep understanding of the workflow and included simulated versions of the call list, allowing researchers to evaluate how clinicians interact with algorithm-generated insights.
The experiments focused on identifying where in the decision chain an algorithm could provide useful input—particularly during the early triage steps when staff must rapidly review a large number of incoming calls. Clinical practitioners were involved throughout, ensuring the work remained grounded in practical reality. The experiments tested how algorithmic recommendations, such as call prioritisation scores, affected clinicians’ review and selection of cases.
Rather than aiming to replace human judgement, the goal was to find ways algorithms could reduce cognitive load, highlight potentially overlooked calls, or assist with case re-evaluation. The experimental findings showed that, when carefully integrated, algorithmic tools could support staff without undermining their control or expertise. Success depended on the timing, presentation, and perceived trustworthiness of the algorithmic input.
In summary, this research gives a close-up view of how life-saving decisions are made in a high-pressure environment. It shows the need for deep understanding of real working conditions before introducing any technology. The approach allowed observation of the complexity of this vital work without breaching patient privacy. The findings will inform how future technology could assist critical care teams—ensuring that new tools help the people who save lives, without getting in the way.
1 An NHS Confidentiality Advisory Group (CAG) provided ethical oversight and approval for this work under 24CAG0106