Royal College of Ophthalmologists (RCOphth)

Overview

We deployed 52 AI tools across all eight RCOphth departments: 42 standard assistants covering operations, communications, finance and member services, plus 10 custom assistants built with deliberate constraints for a clinical governance context.

Combined productivity reached an estimated 5,800+ hours per year across 44 users. Onboarding reached 98% completion. The Course Accreditation Assessor saves 45-90 minutes per application. The Review Service pipeline represents a 200+ hour annual saving opportunity.

Regulated processes, key bottlenecks

Royal Colleges operate under strict governance frameworks with clinical, educational and regulatory responsibilities. RCOphth manages accrediting training courses, reviewing NHS department performance and maintaining professional standards across the ophthalmology profession. Manual processes created bottlenecks around key individuals.

RCOphth faced five major challenges. Course accreditation reviews required checking applications against a 52-item checklist with frequent back-and-forth on missing information. One staff member wrote all Review Service reports. Each took 40 hours and 8-12 weeks, allowing only 5 reports yearly. The Eye journal received high manuscript volumes requiring line-by-line compliance checking. College tutor nominations, travel awards and CPD applications each required 30 minutes of manual membership verification. GMC specialist register applications arrived as evidence packs of 1,000-3,000 pages requiring manual mapping against 180 learning outcomes.

Standard tools first, custom builds next

We deployed 42 standard assistants personalised with RCOphth's core business documents. These covered everyday tasks across the organisation: drafting content, analysing data, preparing meetings, reviewing documents, managing member queries. Every staff member had a working AI toolkit from day one.

We placed a bigger focus on in-person training sessions for RCOphth than most engagements. We delivered organisation-wide onboarding for all 44 staff, followed by department-specific discovery sessions with each of the eight teams. The leadership team prioritised AI adoption and actively supported every session. Staff brought a genuinely inquisitive attitude from the start.

Each custom assistant was built with deliberate design constraints reflecting the clinical governance context. The Governance Helper has internet search disabled to prevent hallucination; it only references uploaded governance documents. The JD Compliance Checker was calibrated to check for minimum standards, not perfection. The Course Accreditation Assessor uses sequential processing: extract answer, check guidance, reason, then respond. These constraints built the trust staff needed to rely on the tools.

98% adoption across a Royal College

98% of staff completed all onboarding sessions. This was driven by strong leadership, a well-organised team and a genuine curiosity across every department. Standard assistants were in daily use before the first custom build was deployed.

The Course Accreditation Assessor was the first custom assistant, saving 45-90 minutes per application by checking against the full 52-item checklist with evidence citations. The NOD Data Checker was signed off as complete. The Trustee Briefing Analyst generates quarterly briefings from operational data, saving 3-5 hours per quarter.

By month five, all 52 tools were in production or testing. Combined productivity reached an estimated 5,800+ hours per year across standard and custom assistants. 31 operational challenges were formally mapped across eight departments.

Automating the highest-value bottlenecks

The Review Service pipeline is the first automation target. One person currently writes every report, each taking 40 hours over 8-12 weeks. That limits capacity to 5 reviews per year. We have decomposed the pipeline into five addressable components. Automating the data gathering, structuring and first-draft stages will remove the single-person bottleneck and let RCOphth increase review capacity significantly.

GMC specialist register applications are the second target. Evidence packs of 1,000-3,000 pages currently require manual mapping against 180 learning outcomes. An automated extraction and matching pipeline would reduce processing time from days to hours per application.

Beyond these two priorities, we have identified 13 further automation opportunities across the college. These include event certificate generation, guideline literature search automation, CPD point calculation workflows, manuscript compliance pre-screening and membership verification pipelines. Each one removes a manual handoff that currently sits with a single team member.

What we learned

The standard toolkit gave every staff member immediate value while we ran discovery across the specialist departments. A bigger focus on in-person training sessions built shared AI literacy before custom work began. 98% completion set the strongest foundation of any engagement.

The JD Compliance Checker needed recalibrating after staff feedback. The AI was initially too strict, flagging acceptable wording as non-compliant. We adjusted it to check for minimum standards, not perfection.

In clinical governance contexts, deliberate capability constraints build trust. Disabling internet search on the Governance Helper prevented hallucination. Staff trusted it because it could only reference uploaded governance documents.

*This engagement was delivered through the smartAI programme, our reseller partnership with smartimpact. Read the smartimpact case study to learn more.