Choosing Healthcare AI: What Questions Should NHS Leaders Be Asking

6 days ago
6 min read

Clinical capacity is under pressure. Referral volumes are climbing, waiting times are stretching, and patients and staff are frustrated. Meanwhile, your inbox is full of vendors promising AI solutions that will revolutionise care delivery. The question isn't whether technology can help, it's how to separate genuine opportunities from expensive distractions.

Dr. Dan Mullarkey, Medical Director at Skin Analytics and practicing GP, explains the reality: "You need many yeses for any deployment to succeed, but you only need one no to stop it." That single "no" often comes down to three factors.

1. Underestimating change management capacity

2. Unclear performance baselines

3. Risk aversion without a proper understanding of the risk of the status quo

As Dan straddles both the NHS and industry, he can see both perspectives — as a potential purchaser of health tech solutions and as the Medical Director of a health tech company.

We asked him for his advice on what questions and information should inform our decision-making process, and here is what he shared.

Choosing Healthcare AI: What Questions Should NHS Leaders Be Asking

Start with Data, Not Vendor Promises

Before evaluating any AI solution, you need to understand your current performance. This isn't about finding blame—it's about identifying genuine opportunities for improvement.

Map your pathway performance: Take dermatology as an example. Do you know what percentage of your urgent suspected cancer referrals actually turn out to be cancer?

National data suggests it's less than 10%. Are you comfortable knowing that significant clinical capacity will be spent on those 90% of patients whose lesions are benign, potentially leading to a patient with a malignant lesion waiting longer? How does this change when you see that national data shows that as many as 1 in 3 melanomas will be found in GP referrals not on the urgent suspected cancer pathway? How about when that patient’s five-year survival rate declines significantly if their cancer is caught at a later stage.

Quantify the real problems: How many patients are waiting? How long are they waiting? What's the downstream impact on secondary care? "If you haven't quantified that

performance, it can feel relatively easy to be ignorant of a risk that exists in your pathway," notes Mullarkey.

Look beyond your patch: The problems you're experiencing likely span primary and secondary care. In dermatology, for instance, primary care continues to generate increasing referral volumes (more than 700,000 referrals were made through the urgent suspected skin cancer pathway in 2024 – an increase of 11% YOY) while one in four consultant dermatologist posts remain unfilled. This isn't just your problem to solve; it's a system problem that requires system solutions. Working in partnership with Trusts and ICB colleagues is essential.

The Real Business Case: Beyond Simple Cost Savings

It’s tempting to focus solely on direct costs, but the strongest business cases consider broader impact:

Population health impact: Understandably clinicians must consider "what's best for the patient in front of me", but from a planning and system-level perspective, teams also need to ask "what offers the most good for the entire population.” This requires different decision-making frameworks and different metrics.

Risk mitigation: Consider the risks of not acting. In dermatology, as many as one in three high-risk skin cancers are found on wrong pathways, with patients waiting on routine backlogs while cancers progress. What's the real cost of maintaining the status quo?

Cutting Through the AI Hype: Questions That Matter

When vendors approach you, focus on these essential questions:

Regulatory status: Is this a medical device? What class? What is the intended use? Vendors should be crystal clear about regulatory status and committed to appropriate oversight.

Performance data: Confirm where the performance has been validated. How have they navigated the transition from research to the real world? How many patients have been assessed? What external scrutiny has taken place for this data e.g. notified body, independent evaluations, publications etc.

Implementation reality: Can the vendor provide templated approaches based on previous deployments? How long does implementation typically take? What training is included? Most importantly—what change management support do they provide?

Ongoing monitoring: How will you measure success? What data will you have access to? How does this compare to your baseline metrics, and will this raise the bar?

Learning from Real-World Implementation

Following a number of successful deployments in secondary care, Skin Analytics started to explore how AI could be deployed in primary care. They found that deploying AI at individual GP practice level led to predictable results: excellent uptake in a third of practices, variable adoption in another third, and no uptake in the final third. Sound familiar?

The solution they identified – and which has now been successfully deployed in several ICBs – was moving to community diagnostic hubs where medical photographers capture standardised images for AI analysis. This approach provides:

● Consistent implementation across the network

● Proper equipment maintenance and operation

● Standardised protocols and training

● The lesson for healthcare leaders: think beyond practice-level deployment toward network-wide solutions that address workflow reality.

Change Management: Where Most Deployments Fail

Technology deployment isn't primarily a technical challenge—it's organisational. Successful leaders approach this systematically:

Resource allocation: Dedicate specific staff time to implementation and ongoing support. This isn't something to squeeze into existing job plans.

Stakeholder engagement: Involve clinical champions, practice managers, and administrative staff from planning through deployment. Resistance usually stems from workflow disruption rather than technology concerns.

Phased approach: Start with limited implementation and scale as processes become embedded. Plan for 8-12 weeks for initial deployment, then gradual expansion based on performance data.

Performance monitoring: Establish clear metrics from day one and review regularly. Use data to drive continued adoption and identify improvement areas.

Evidence from implementing AI at scale

Skin Analytics now processes approximately 15% of NHS urgent suspected skin cancer referrals, providing real-world evidence of scaled AI deployment. Key lessons include:

● Start with a clear problem definition and performance baselines

● Invest properly in change management from day one

● Focus on solutions that create genuine capacity rather than shifting workload

● Work with vendors committed to appropriate regulatory pathways

● Plan for system-wide impact rather than practice-level fixes

Making Strategic Decisions in Uncertain Times

AI in healthcare isn't about replacing clinical judgment; it's about supporting decision-making where it adds genuine value – including using proven solutions to triage cases and ensure the patients with the most need can access care when they need it most. For healthcare leaders, this means developing a different way of thinking about risk. New technologies no doubt bring a degree of risk; but when we compare this risk with the risk of not acting – growing referral numbers, rising waiting lists, and no organic increase in workforce capacity on the horizon, the risk of using highly-regulated technologies for their intended purpose pales in comparison. When and how to rely on them depends on an organisation’s ability to evaluate, implement and monitor solutions and understand the impact they are having.

The organisations that master this approach won't just solve today's capacity pressures. They'll build sustainable foundations for delivering high-quality care in an increasingly complex healthcare landscape.

Your next steps: start with your data, be honest about organisational readiness, and focus on solutions that address genuine problems across care pathways. The technology exists—the question is whether your organisation is prepared to harness it effectively.

The choice isn't whether to engage with AI—it's whether to do so strategically or reactively.

If your PCN is exploring new ways to support patients with AI and skin cancer pathways, it may be worth talking to Dan and the team at Skin Analytics.

Contact the Skin Analytics team here: skin-analytics.com/contact-us

We also recorded a podcast with Dan from Skin Analytics which you can listen to here.

About Us

THC Primary Care is an award-winning healthcare consultancy specialising in Primary Care Network Management and the creator of the Business of Healthcare Podcast. We've supported more than 200 PCNs through interim management, training, and consultancy.

Our expertise spans project management and business development across both private and public sectors. Our work has been published in the London Journal of Primary Care, and we've authored over 250 blogs sharing insights about primary care networks.