.png)
There’s a line often attributed to Virgil: “the greatest wealth is health.” Two thousand years on, it still rings true. When health slips, everything else tends to slow down with it, work included.
Fuzzy Marmalade sees this connection play out every day. As a workplace health provider, they help employers connect people with the right clinical support at the moments it matters most.
As demand for workplace health services grows, the early decisions that shape someone’s care come under increasing pressure. Fuzzy Marmalade wanted to strengthen how those decisions are made and reviewed, while keeping judgement and accountability firmly with the people doing the work.
The challenge
Most workplace health work begins with a referral. A manager or employer submits a description of a situation that needs to be interpreted and turned into the right next clinical steps.
These referrals vary widely. Some are clear and straightforward. Others are ambiguous and rely on careful interpretation. Triage Advisors (TAs) work with documented policy, clinical guidance, and client-specific rules, alongside experience built up over time.
As demand for workplace health services continued to grow, it became increasingly important to apply that guidance consistently at scale, and to ensure decisions could be reviewed, explained, and adapted as requirements evolved.
The aim was to support TAs with clearer, more consistent analysis, and to help the process scale without losing oversight or shifting responsibility away from people.
The approach
Before any technical work began, we followed referrals through the process from submission to booking to understand how experienced judgement is applied across triage, and where cases require deeper review and context.
That insight shaped the design. The system does not make decisions on behalf of TAs. Instead, it brings together the relevant guidance and client rules and surfaces a suggested course of action alongside its supporting context for TA review. Where referrals are more complex or involve multiple issues, they are clearly flagged for fuller human consideration, with responsibility remaining unchanged.
The capability was integrated directly into Marmalade’s core referral management system, so TAs see recommendations and supporting context within the workflow they already use. Alongside triage support, an appointment recommendation service was added to surface suitable booking options, helping TAs move from decision to appointment without removing oversight.
How we built it
We began with collaborative modelling workshops with Fuzzy Marmalade’s domain experts, working through real referrals, historic TA decisions, and existing decision flows to understand how triage works in practice.
Using historic referral data, we examined how referral text related to the outcomes chosen by experienced TAs. This helped clarify where consistent patterns already existed and where referrals should always receive closer human review.
The triage capability combines several techniques. Natural language processing is used to extract key features from unstructured referral text, alongside rule-based checks for recognised conditions and client-specific requirements. More complex or ambiguous cases are surfaced clearly for TA judgement, rather than being forced into a category.
To support the handover from triage into booking, simple scheduling support was added. This surfaces suitable appointment options based on clinician availability, making it easier to move from decision to booking without removing oversight.
Throughout, MLOps practices were applied to support safe operation over time. Changes to decision logic and guidance can be tested, reviewed, and rolled back if needed. Feedback from real TA decisions allows the approach to be refined gradually, without introducing risk into live workflows.
The system runs in Marmalade’s cloud environment and meets their security and compliance requirements. Delivery was carried out through close collaboration and pair-programming with Marmalade’s engineers, supporting knowledge transfer and long-term maintainability.
The impact
When tested on historic referral data, the approach aligned with experienced TA decisions in more than 92% of cases, providing a clear baseline that could be monitored as it moved into live use.
Turnaround times have improved and now take minutes, while decisions remain reviewable and grounded in documented guidance. Outcomes are more consistent and easier to audit.
TAs also have clearer support when identifying appropriate clinical expertise and moving referrals into booking, helping maintain quality as demand grows.
Why it matters
When health is affected, the decisions made early on can shape how someone experiences work, care, and recovery.
This work strengthened how referrals are interpreted and reviewed, so guidance is easier to apply and decisions are clearer and more consistent, without changing where responsibility sits.
The result is a process that supports people doing complex work every day, and can grow with demand without becoming rigid or mechanical. It’s a way of supporting people when they need care, and treating them thoughtfully when it matters most.
.png)
