AI has been everywhere for several years, but Paul Barnes, Chief Commercial Officer at Nourish Care, noted that the conversation had always been about the future. I attended a panel at The Care Show in Birmingham which was about “what’s happened already”: AI as business as usual in social care.
Nourish Care led the session and it centred on Nourish’s own tool, and the most useful evidence came from two providers using it. Mike Cleasby is Chief Quality and Risk Officer at National Care Group. Mark Hoult-Allen runs the Nourish rollout at Choice Care Group, which has about 130 residential and supported living services and was paper-based until recently. Carrie Taylor, Nourish’s Head of Clinical Product and Practice, is a registered nurse and the company’s clinical safety officer.
From dip-testing to 100% coverage
Care plan audits have usually meant dip-testing one or two plans a month. Nourish chose them as its first use of AI because providers struggled most with their volume, quality, and consistency.
The tool, called Confidence, scans every care plan each month. It flags “improvement opportunities”, sorts them into categories, and explains why each was raised. Its model is trained on UK social care, not on the general Internet. A manager can then see patterns across a whole service, not just faults in individual plans.
The main value is consistency across long documents, where a human reviewer can’t keep every detail in mind. A plan might say early on that a person has full capacity, then contradict that in the activities section. Another common flag is a plan that describes a condition or medication in detail but never says how staff should act on it.
A tool for managers, not a stick
Nourish expects quality teams to drive adoption, using AI to free time for other work. Across its roughly 75 customers, Barnes said, registered managers and service managers have taken the lead.
At Choice, Hoult-Allen gave the tool only to operations teams and avoided the words “compliance” and “audit”. He first shared results on paper and talked managers through them. Seeing “86 opportunities for improvement” on one plan, he said, can be “a bit terrifying”.
National Care Group took the same approach. It ran a small pilot and made clear that the tool was for learning, not for punishing staff over audit results. In both organisations, staff outside the pilot soon asked when they would get it. Managers no longer have to wait for an internal auditor to tell them what to fix. As Barnes put it: “Suddenly I’m in control.”
Earning trust, keeping humans in charge
The AI does not change care plans; it only points out issues. “Now you need to interpret that and go away and do it,” Hoult-Allen said. Barnes added that the result is a human decision with a traceable record behind it.
As clinical safety officer, Taylor admitted she was nervous about accuracy. Before release, her team of clinicians manually checked the outputs against best practice.
The providers still tested it. “We were probably trying to break it,” Cleasby said, and National Care Group kept running traditional audits alongside it. At Choice, a deputy manager reviewed the AI’s comments on plans she had written a year earlier. Her verdict: “I don’t disagree with anything.”
It remains a tool, not a final ruling. Managers can dismiss flags they disagree with, such as Choice’s continued use of “challenging behaviour”. As Cleasby put it, AI “should support professional judgement, not replace it.”
What regulators made of it
So far, the inspection stories are positive. Barnes described a home manager who, during an unannounced CQC visit, showed inspectors the tool, what it had found, and what had been fixed. The inspector treated this as self-assessment. After hearing about the visit, other home managers in the group asked to have the tool too.
National Care Group has had several CQC inspections in recent months, and the feedback was “really, really positive”. The tool helped show “a learning culture, continuous improvement”.
Taylor said the bigger shift is in how progress is judged. No service will ever have zero issues. Managers can instead deal first with what matters most to the people they support, and show a steady record of improvement. She called inspectors’ acceptance of this “a really positive shift in the sector”.
The outcomes for care homes
Asked what they would report to their boards, the providers named four results.
- Measurably better plans. At Choice, the number of flags on a single support plan has fallen from as many as 84 to five or six. “That’s almost 80 things that could have been improved with that plan that have been improved,” Barnes said. “And that’s measurable.” The fixes include language that wasn’t person-centred, contradictions, and risks recorded at the wrong level
- Safer, clearer guidance for care workers. Hoult-Allen said the most important outcome is that staff can trust what they read. “What I’m seeing on my device is not contradicting something I’m not currently looking at.” Better-written plans mean care workers know how to act on the information. Cleasby said the tool has “undoubtedly helped keep people safe” and supported more person-centred outcomes
- Less searching, more fixing. When a manager updates a medications page, the tool finds every other place that medication is mentioned. Barnes quoted a customer who said staff used to spend “half our time looking for problems, half our time fixing it.” Physical audits are still needed, but time spent searching has fallen sharply
- Quality teams back on the floor. With less time spent reading plans, Hoult-Allen expects quality staff to spend “more time in services talking to the people you support and to colleagues”. That part of quality work “often gets overlooked”
What comes next
Nourish has added better filtering and links from each flag straight to the relevant passage in the plan. Next, organisations will be able to add their own policies and terminology. A risk register is also planned, feeding a central action plan tracker.
The providers had their own wish lists. Cleasby wants “a single pane of glass” across all digital systems, where managers trust that green items are fine and focus only on amber and red. Hoult-Allen wants the AI to compare support plans with scheduled care to spot gaps. For example, a plan that notes continence needs should have pad checks scheduled.
Barnes said the main lesson was the approach, not the technology: prove it at scale first, then expand step by step. For care homes considering AI, the panel’s advice was to start small, involve managers early, and treat the output as evidence for human judgement, not a verdict.
