The 'stuck paradox': what we learned from leaders about AI and digital transformation

Over the past few months, we have spoken to nearly 60 health and care leaders about the opportunities and challenges presented by AI, and how they are leading their organisations through a period of rapid technological change.

While many of the insights they shared are not unique to health and care, they speak directly to the realities of leading in an already stretched system.

At the heart of this sits a central tension: the ‘stuck paradox’. Leaders feel an urgency to accelerate the use of AI alongside real constraints on their ability to act. These constraints range from limited resources and transformation capability to gaps in knowledge or confidence to make the right decisions. For many it is a combination of all of these.

We are building on this phase of discussion to support leaders through AI and broader digital change. In the meantime, we share ten key themes that have emerged from our initial conversations.

1. The 'stuck paradox'

Leaders shared that they feel both energised by AI’s potential and overwhelmed by its speed and unpredictability. It is impossible to keep up to date with the latest developments, including available tools and use-cases. Some leaders and organisations are already reaping benefits from AI tools, such as ambient AI for notetaking and consultation summaries. Others are still experimenting. Across the board, however, there is awareness of the risks associated with deploying AI and the potential consequences for employment. Leaders know that decisions taken now may prove sub-optimal within the next 12 months on account of the speed of change, and this feels difficult to navigate.

Some described this as a ‘stuck paradox’, with leaders experiencing both a sense of urgency to act alongside constraints on what they can responsibly deliver. AI is everywhere in the media and policy discourse; there is a sense of needing to ‘do something about AI’ and a worry about falling behind. At the same time, leaders and organisations face constraints including capability gaps that can be paralysing and slow the pace of change.

‘I remember reading an ICB email like nine months ago saying this tool wasn’t supported…but maybe that’s changed. I don’t know. No one’s told me.’

The pace is uneven; some staff groups and organisations are leaping ahead with AI transformation. Leaders are wondering how much this matters. It feels messy and creates uncertainty. And because of the speed of transformation, leaders are sometimes grappling with complex tensions in isolation. It can feel lonely. Many of the critical leadership questions remain ‘privatised’ within individual organisations, creating a false sense that challenges are unique to place rather than more systemic. Peer learning networks to share and test thinking would be welcome, along with ready access to technical updates on how AI and related technology is evolving.

2. What do we mean by productivity?

Many clinicians have embraced ambient AI tools, prompting demand for broader, quicker roll out. Yet this raises questions about the intent. Is the goal to drive (potentially quite narrowly defined) productivity, for example, to increase the number of diagnostic tests that can be carried out? Or is the aim to create more space to think and reflect, and ensure an optimised experience for patients and staff?

Many leaders and institutions lack a framework for judging success in productivity terms. It is easier to favour increased flow and through-put at the expense of reflection time, but is this sustainable longer-term? As one leader put it, we are in danger of just adding ‘more widgets’. And what does it do to job satisfaction and the risk of burn-out among health care professionals? Leaders would welcome frameworks to help them think through these questions, alongside real-world evidence from places experimenting with different approaches.

3. Balancing risks and opportunities

A number of conversations focused on the opportunities AI offers. Automating administrative tasks was frequently mentioned as a way to give clinicians and managers more time. Another example cited was 24-7 AI information support and therapy. Both staff and patients are already accessing AI support around the clock, helping to interpret test results out of hours, or providing support to stressed health care professionals. While this can offer clear benefits, leaders are considering the optimal balance between AI and human input. This includes considering where it is important to have a human supplement AI-generated support or therapy, as well as risks including inaccurate information, or errors in medical notes not checked by humans. The need to understand and then balance opportunities and risks is at the heart of the leadership challenge for many we spoke to.

4. Differences across health and care settings

Participants reflected on different opportunities and challenges in distinct parts of the health and care sector. A social care leader predicted faster roll out of AI tools in social care (some of which is already happening) due to fewer professional interests ‘getting in the way’ and the widely agreed benefits of digitisation, at least for some forms of communication and record sharing.

A voluntary and community sector (VCSE) leader talked about the disintermediation of information provision and the challenge this presents for how charities build and maintain relationships with individuals. Charities have historically played a leading role in supporting beneficiaries’ information needs, but are increasingly finding this link broken, as citizens rely on AI-generated summaries of conditions at the point of diagnosis, treatment options, side effects and so on. This has implications for the business model of many organisations. People often seek information as a first step, before supporting charities in different ways; volunteering, fundraising or campaigning. This journey is becoming more fragmented and if charities have less contact with beneficiaries, they may find it harder to represent their views or advocate on their behalf. VCSE leaders want all beneficiaries to receive tailored, accurate information and expressed concerns that digital exclusion could prevent this, and that mistrust of technology could get in the way too.

Across NHS services, well-resourced organisations working at scale are better placed to forge ahead with AI implementation than those with less available infrastructure to support service transformation, for example, some primary and community trusts. This is true of any transformation, but in such a rapidly changing landscape, leaders reflected that uneven adoption could have a significant impact on how staff and patients experience organisations, with implications for where they want to work potentially reinforcing the status of certain organisations. This can work both ways, as the leader who felt social care can ‘leap forward’ reflected, in part because there is a burning platform for change that AI can help deliver.

5. Patient trust

As recent research shows, patients prefer AI outputs to be checked by a human over speed of results. They want AI to have strong evidence requirements even if this slows the rollout of tools. This is showing up in different ways in the system. One leader gave an example of AI-driven diagnostic testing that gives instant results. Clinicians delay sharing the results until the following day due to levels of patient mistrust if the findings are given too quickly. Some patients want to hold AI and other digital technology to a higher standard and will currently accept a higher rate of human error than AI error.

There is a risk of not adopting AI faster, which is less well explored or discussed. When millions of people are sitting on a waiting list, significant risk is already being held in the system. If you deploy AI to work through your waiting list more quickly, what is the error rate you are willing to tolerate? This needs more attention and national level discussion – and is in part what the National Commission into the Regulation of AI in health care is trying to address – to help leaders work through the risk-benefit analysis and to give organisations ‘cover’ to act.

6. Widening inequalities

With transformation on this scale, the introduction of AI is bound to be uneven, and some people have voiced worries about the effects on those who might get left behind. A hospital CEO noted that this could create a ‘three-tier health service’: the wealthy using private care for faster treatment, others relying on tools like ChatGPT for most of their health information needs – receiving ‘80% of the way health advice’ – and a group of people at risk of falling behind because they lack access, skills or confidence, or chose not to use digital technology. Leaders are considering what they can do to mitigate risks such as widening inequalities in their organisations or sphere of influence. 

7. Workforce displacement

Many health and care organisations take their responsibilities as ‘anchor institutions’ in their communities very seriously, understanding the benefits they offer for local employment, training, professional development, buildings and land use. They are therefore worried about the impact of AI on their workforce. Leaders talked about the fact that entry level roles are disappearing, that the ‘first seven years’ of career ladders are shrinking in some professions and that this, whilst not unique to health and care, has ramifications for the development of managers and senior leaders in future.

Several leaders reflected on significant job reductions anticipated in the coming years. One Trust CEO based in an area of significant deprivation reflected that this would affect clerical and administration staff in his organisation, mainly female, often in their fifties and the main breadwinner in families. The tension between organisations taking their role as anchor institutions seriously and the role displacement they are beginning to predict or experience risks ‘cutting the soul’ out of places, and some described this as an unquantified risk. There are no easy answers here, but grappling openly with this trade-off is an act of leadership.

8. Workforce polarisation

Another reality expressed is the tension between early AI adopters among staff experimenting ‘broadly and loudly’ with AI versus anxious colleagues avoiding AI use so far.

There is divergence in many organisations between organisational culture and the appetite for innovation and risk shown by some leaders or staff groups. Where organisations are trying to accelerate AI adoption, some are facing blocks and delay from wider cultural forces that are less adaptive and accepting of risk. Navigating this takes time and helping lead staff on an implementation journey is an important task for system leaders.

9. Governance and strategy

‘We don’t have a strategy for electricity, so why do we need one for AI?’

Leaders discussed how boards (trustees, non-executive directors, selected senior executive leaders) often want an AI strategy for assurance and to manage organisational risks. Some are resisting this as they do not want to waste time developing quickly outdated policy documents. Standalone strategies risk siloed ownership, low engagement, rapid obsolescence, and a fear of defaulting to ‘documents not action’, meaning that behaviour change in organisations is sluggish and uneven.

Instead, leaders identified a clear need for a framework for using AI and other digital technology, but with a preference to integrate into existing governance. They want to prioritise embedding AI across their workforce, encouraging experimentation, creating a learning environment where adapting to an AI-enabled world becomes everyone’s business, and is aligned to strategic planning and the overarching mission of an organisation.

At the same time, AI accountability was highlighted as a challenge, and something that many organisations are working through. There is often a desire to mirror medical governance; who is accountable for errors? Where does the buck stop? This was also linked to benefits realisation in one conversation. Organisations often ‘follow the next implementation’ without robust outcomes tracking. This raises an important question. Who is responsible for taking a view on whether AI is doing the job required and the stated objectives are being met?

10. National digital leadership was described as sub-optimal

‘AI adoption is dependent on a few culture carriers and culture warriors holding it in organisations, and they’re very often not at the top.’

AI adoption is increasingly led by (younger) staff groups and patients, rather than strategically. AI champions notice frontline problems, investigate if AI can help, test with small groups, then scale. A problem-first approach is often beating a strategy-first approach.

Given the accelerated pace of change, this outcome may be unavoidable. Leaders acknowledged that while this is not inherently negative, it underscores the necessity for enhanced national leadership supported by appropriate inputs. Historically, digital leadership competencies have not been championed in the NHS, nor has successful digital transformation been particularly rewarded. This needs to shift quickly. Against a backdrop of integrated care board cuts and broader NHS restructure, there is a risk of eroding these capabilities further. And bottom-up leadership in AI could be supported and accelerated by tackling a culture that is often hierarchical and does not champion these leadership behaviours at more junior levels.

The gap between frontline AI adoption and strategic national support represents a genuine leadership void that is leaving leaders exposed. What do leaders need to navigate this? They would welcome stronger signals and guidance about best-in-class tools, along with resource to implement where possible. But crucially, they need permission to hold complexity, peer networks to test thinking, and frameworks that acknowledge the trade-offs honestly.

Do these insights resonate? Are there other opportunities or challenges you are considering? Share your thoughts in the comments below.