Three AI Agents, One Occupational Health Consultancy, 62% More Cases Completed

The management referral came in on a Tuesday. A 200-person logistics company in the East Midlands needed an occupational health report for an employee who had been off sick for six weeks. Standard work for the four-person Leicester practice. But by Thursday, the report was still a half-finished draft — sitting in a clinician's inbox, behind eleven others — while two consent forms remained outstanding and a corporate client was chasing their monthly management information summary that was four days late. This was not a bad week. It was every week. <figure> <img src="https://images.unsplash.com/photo-1576091160550-2173dba999ef?w=1200&q=80" alt="AI agents for occupational health consultancy — three agents handling management referral reports, case tracking and client MI reporting for a four-person Leicester practice, completing 62% more cases per month" width="1200" height="630" loading="lazy" /> </figure> <h2>The 28-Hour Problem</h2> <figure> <img src="https://images.unsplash.com/photo-1450101499163-c8848c66ca85?w=1200&q=80" alt="Occupational health report writing admin burden — clinicians spending 14 hours per week drafting management referral reports instead of delivering clinical assessments for UK OH practices" width="1200" height="800" loading="lazy" /> </figure> The practice ran a simple exercise: they tracked exactly how their time split across one working week. The result was uncomfortable. Twenty-eight hours — equivalent to nearly one full-time role — was absorbed by three administrative tasks that had nothing to do with clinical decision-making. Report writing was the biggest drain. With twelve management referrals a week, each requiring a structured report delivered to the referring employer, the team was spending an average of seventy minutes per report. That was fourteen hours a week spent on a process that was repetitive, template-heavy, and almost entirely predictable in structure. The clinical judgement lived in twenty minutes of assessment notes. The other fifty minutes was formatting, boilerplate, and administrative scaffolding that reproduced itself identically on every case. Case management admin came second. Chasing GP records, sending consent form reminders, tracking which cases were approaching their SLA window, producing a weekly case status update for the team — eight hours a week, most of it reactive. The team was managing cases from memory and a shared spreadsheet, which worked until it didn't. Cases slipped. Consent forms expired. A GP record that had been requested three weeks earlier had never arrived, and nobody had noticed. Client MI reporting was the third problem. Four corporate clients expected a monthly management information report: cases referred, cases completed, average turnaround, outcomes by category. Each report took roughly an hour to compile manually from the tracking sheet and another thirty minutes to write up. Four hours a month, concentrated at month-end, always arriving at the worst possible time. <blockquote>Twenty-eight hours a week on admin in a four-person practice means one person is doing nothing but paperwork. That is not a capacity problem. It is an architecture problem — and it has a specific fix.</blockquote> <h2>The Three-Agent Stack</h2> <figure> <img src="https://images.unsplash.com/photo-1517048676732-d65bc937f952?w=1200&q=80" alt="Three AI agent workflow for occupational health consultancy — report writing agent, case management agent, and client MI agent working together to automate 23 hours of weekly admin" width="1200" height="800" loading="lazy" /> </figure> The practice deployed three agents over eight weeks, each targeting one of the three time drains. The build used n8n for workflow orchestration and Claude for document drafting and structured extraction. Total setup time was under forty hours across the build period. Agent 1: The Report Writing Agent. This agent watches for new management referral submissions — arriving via email or a simple web form — and immediately gets to work. It reads the referral brief, extracts the key details (presenting condition, job role, absence duration, employer's specific questions), and drafts a structured OH management report in the practice's template. The draft pre-populates every section that does not require clinical judgement: background, employment history, a summary of documents reviewed, and a structured recommendations framework with placeholder text clearly marked for the clinician to complete. When the draft is ready, it is sent directly to the assigned clinician. The clinician adds their clinical observations and fitness-for-work recommendations — which takes fifteen to twenty minutes rather than the original seventy — approves the document, and the agent sends the final report to the referring employer. Report turnaround dropped from an average of seventy-two hours to eighteen hours. The clinicians stopped thinking of report writing as a task. It became a review. Agent 2: The Case Management Agent. This agent runs continuously against the practice's case tracking database. It identifies cases where consent forms have been outstanding for more than forty-eight hours and sends structured, professional chase emails directly to the employee — personalised, clear about what is needed, and automatically logged. It does the same for GP records, sending structured requests to practices and flagging cases where no response has arrived within the agreed window. Each morning it produces a two-minute digest for the clinical team: cases approaching SLA, cases awaiting documents, cases where a clinician response is the outstanding step. The team stopped managing their caseload from memory. The agent manages it for them, and escalates only what genuinely needs a human decision. Agent 3: The Client MI Agent. At the end of each calendar month, this agent pulls the completed case data for each corporate client and drafts their management information report. Cases referred, cases completed, average turnaround time, outcomes by category — return to work, phased return, reasonable adjustments, referral for further treatment. Each report is formatted to the client's preferred template, includes anonymised case narrative summaries for any cases the client has flagged for discussion, and is emailed directly to the client's HR contact with the practice director's signature. Four clients, four reports, thirty minutes of agent run time, zero hours of manual compilation. The practice director reviewed each report once before the agent sent it. That review took twelve minutes total across all four. All three agents operate within clear boundaries. The report writing agent never sends a report directly to an employer without clinician sign-off. The case management agent never contacts a GP practice on behalf of a case that has not been through assessment. Every action is logged. The practice director can see exactly what each agent has done at any point. This is the <a href="/blog/human-in-the-loop-ai-agents-uk">human-in-the-loop architecture</a> that makes AI safe in a regulated clinical environment — not because it slows things down, but because it defines precisely where human oversight sits and removes it from everything else. <h2>The Results After 90 Days</h2> <figure> <img src="https://images.unsplash.com/photo-1590283603385-17ffb3a7f29f?w=1200&q=80" alt="Occupational health AI agent results after 90 days — 22 to 36 cases completed per month, 23 hours recovered weekly, report turnaround reduced from 72 hours to 18 hours average, running cost £80 per month" width="1200" height="800" loading="lazy" /> </figure> Before the agents, the practice completed twenty-two management referral cases per month. Ninety days after deployment, they were completing thirty-six — a 62% increase with the same four-person team. The caseload did not grow because they marketed more aggressively. It grew because the bottleneck had been removed. Referrals that previously had to be declined or delayed because the team was at capacity could now be accepted. The practice could say yes. The twenty-three hours recovered per week redistributed across the team differently. The clinicians gained back most of their report writing time and used it for additional assessments. The practice manager, who had been spending four hours a week on case tracking and MI report compilation, now handles business development and client relationship calls that had previously been crowded out. One of the clinicians started a workplace wellbeing workshop programme for two corporate clients — work that had been on the ideas list for two years but had never had the time to get started. The consent form chase rate improved from roughly 60% completion within the first week to 94%. GP record requests that previously took two to three weeks on average now returned in under ten days, because the case management agent was structured, consistent, and professional in a way that an ad hoc follow-up email never quite managed to be. <blockquote>The agents did not make the practice bigger. They made the practice able to work at its actual capacity rather than the capacity that was left over after the admin was done.</blockquote> <h2>The £80-Per-Month Running Cost</h2> The three agents cost £80 per month to run. That breaks down as approximately £25 for n8n self-hosted on Google Cloud Run, £35 for Claude API usage at the practice's volume — the report writing agent uses Claude Sonnet for drafting, the case management agent uses Claude Haiku for the lighter extraction and chase tasks — and £20 for the hosting infrastructure and incidental tooling. The setup cost was around £3,200, covering the build, testing, integration with the practice's existing email and document systems, and a structured handover with documentation. The practice recovered that in full before the end of month three. At thirty-six cases per month versus twenty-two, the additional referral revenue alone more than covered the initial build and ongoing running cost by a significant margin. This is the pattern seen across every practice-scale professional services firm that deploys an AI operating system properly: the economics are decisive, the payback period is short, and the compounding effect — as the agents improve and caseload grows — continues to widen the gap between what the practice can deliver and what it could before. The <a href="/blog/ai-roi-framework-uk-service-businesses">AI ROI measurement framework</a> covers how to track this accurately in your own firm. <h2>What This Means for Your OH Practice</h2> <figure> <img src="https://images.unsplash.com/photo-1552664730-d307ca884978?w=1200&q=80" alt="AI operating system for UK occupational health consultancies — the three-agent architecture that recovers clinician time, improves case throughput, and keeps human oversight built in at every decision point" width="1200" height="800" loading="lazy" /> </figure> The occupational health sector has specific characteristics that make it particularly well-suited to this type of AI operating system. Reports are structured and template-driven. Case management is rules-based, high-volume, and time-sensitive. Client reporting is repetitive and data-driven. These are exactly the conditions where AI agents perform reliably and where the human value — clinical judgement, client relationships, nuanced assessment — can be protected and concentrated rather than diluted by admin work. The regulatory and GDPR considerations are real but manageable. Every AI interaction in this system involves anonymised or pseudonymised data at the point of processing. No personal health information is sent to external AI systems without confirmation that the relevant DPAs are in place. The <a href="/blog/ai-governance-framework-uk-service-businesses">three-layer governance framework</a> covers the compliance architecture that UK health and professional services firms need before deploying agents against client data. Most OH practices operating at four to ten clinicians are running a version of the same problem the Leicester practice had: a caseload that is constrained not by the availability of referrals or by clinical capacity, but by the administrative load sitting around the clinical work. The agents described here do not replace occupational health expertise. They remove the scaffolding that was taking up the time that expertise should be filling. If you run an occupational health practice and want to understand what a three-agent stack would look like for your specific referral volume and client structure, <a href="/contact">get in touch</a>. We design and build AI operating systems for UK professional services firms — including those operating in regulated environments — with a production-first approach that keeps clinical oversight where it belongs.