.avif)
Stage 1: Identify and prioritise residents based on impact
Targeting prevention where it is most needed
For local authorities implementing targeted proactive prevention, the foundation of success lies in answering a deceptively simple question: "Who should we focus on first?" This critical decision shapes every subsequent aspect of the model - from which partners to engage, to what interventions to offer, to when measurable impact might be observed. This is not, of course, to suggest that other people would not benefit, but rather to make a strong case for prevention that can generate local confidence from the outset, by prioritising where the greatest possible impact can be made.
The most successful prevention authorities approach this question systematically, using evidence to identify cohorts where early intervention can demonstrably improve outcomes while helping respond to increasing needs.
On this page you can find out:
- the steps involved in this stage
- the ethical considerations of using AI to identify people most at risk
- case studies from local authorities already doing this in practice
- potential barriers you might face and how to overcome these
This stage involves two key steps that build the foundation for effective targeting:
Step 1: Prioritising a cohort of older people for targeted proactive prevention
The ultimate goal is enabling as many people as possible to live independently and connected to their communities. To build momentum and evidence for this vision, authorities typically begin with cohorts where prevention can demonstrate clear impact within a reasonable timeframe.
Older adults represent a strategic starting point due to demographic trends and growing patterns of need. The rapid growth in this population segment means that improved outcomes can simultaneously enhance individual wellbeing and address system pressures.
Local authorities typically identify priority cohorts through one of two approaches:
Experience across numerous local authorities has shown that when data quality or governance constraints exist, multi-disciplinary case discussions offer valuable insights. By reviewing a representative sample of recent care package starts, teams can understand the primary drivers of need and explore preventative measures that might have been beneficial.
While this method is resource-intensive and may not represent the entire population, it introduces multi-professional judgment into decision-making, including assessments of preventability.
For example, a local authority wanting to understand the biggest preventable risk factors for long-term care, brought together practitioners from across adult social care, their health partners, and the voluntary sector. This group reviewed recent long-term care starts for older people, looking through their shared case notes to understand what opportunity there could have been to intervene earlier.
Through this activity, the authority identified that falls and carer breakdown were the most common reasons why someone would go onto receive residential care, and practitioners estimated that these older people could have benefited from earlier help in approximately a third of cases.
Some local authorities have used machine learning to identify drivers of need from diverse datasets. For example, given the richest information on people sits in free text case notes, secure Large Language Models are now being used to automatically analyse case notes and extract useful insights about people’s strengths and needs, such as whether they have had a recent bereavement or whether they feel isolated living at home alone.
Authorities are using this as a means to understand why people are approaching the social care “front door” and requiring long term care, which in turn is enabling them to understand which needs to prioritise supporting proactively. While this approach requires upfront investment in technology and at times complex Information Governance conversations, it is often a more sustainable and adaptable option that allows authorities to incorporate new trends and information over time.
When partner data, such as that from district councils, VCFSE sector and NHS organisations, is included, authorities are able to identify individuals not yet known to social services and reveal opportunities for collaborative intervention.
Step 2: Identifying individuals in need of targeted proactive, preventative support
Once a priority cohort is established, authorities focus on identifying specific individuals most likely to benefit from preventative support. Starting with those at higher risk creates opportunities to demonstrate observable differences in outcomes earlier, building the case for broader prevention efforts.
Approaches to individual identification:
Applies straightforward criteria to identify individuals meeting specific risk profiles. For example, an authority might focus on unpaid carers over a certain age providing substantial care hours.
This approach offers clarity and transparency, making it relatively easy to explain to stakeholders and implement with existing systems. However, it may lack the sophistication to account for multiple interacting risk factors, potentially missing individuals who would benefit from support.
Represents a more advanced approach, analysing thousands of data points to identify combinations of characteristics that best predict future needs. This creates a nuanced risk profile that can more accurately target limited resources.
In Norfolk County Council, for instance, a falls prediction model achieved 70% accuracy in identifying residents at risk, enabling more confident resource allocation. The model safely connected data across organisational boundaries and enriched existing information by applying large language models to case notes.
A third of people over 65 fall every year, deeply impacting a person’s confidence, mobility and wellbeing, and costing health and social care over £4,000.
Norfolk County Council and its partners wanted to identify people at risk of a fall and offer them preventative support to increase independence and reduce demand for adult social care and health.
The programme centred on two key elements:
- Identifying at risk individuals: The Council used the Xantura platform which provided a single view of residents. A supporting AI model predicted with 70% accuracy which residents were most at risk of a fall. They did this by safely connecting individual data across organisational siloes to form a holistic understanding of Norfolk’s residents; automatically enriching existing data by applying Large Language Models to case notes to extract strength and risk information; and by applying machine learning models to identify people at risk.
- Intervening to mitigate the risk: Tailored interventions are provided to at-risk individuals, building on individuals’ strengths and on the assets within local communities.
What next?
The Council is now mobilising the Proactive Intervention operating model (proactively identifying and engaging with c.12,080 people at 58-99% risk of a fall). The platform and tools developed will next be used to identify subsequent cohorts of individuals to target for proactive interventions, enabling the Norfolk system to shift from a reactive to a proactive model of support. Cohorts being considered include isolation and loneliness.
As artificial intelligence (AI) and large language models (LLMs) become increasingly integrated into social care services, local authorities are carefully considering the ethical implications of their use. While AI-driven tools offer the potential to enhance prevention strategies by identifying risks early, personalising interventions, and improving accessibility, they also raise significant ethical questions. Guidance from the government highlights the importance of addressing these concerns, ensuring that AI is deployed responsibly and aligns with core social work values. Authorities have also utilised the Government's Data Ethics Framework and the Ethics, Transparency and Accountability Framework for Automated Decision-Making, to support with understanding and assessing risk and ethical concerns with utilising AI to support practice, such as supporting targeted proactive prevention efforts.
Some key considerations for local authorities include:
- Data privacy: Safeguard sensitive data by adhering to GDPR and ensuring robust security measures.
- Bias and fairness: Regularly audit AI systems to identify and address any biases, ensuring fair outcomes for all groups.
- Transparency: Clearly explain how AI decisions are made and establish accountability for its actions.
- Human oversight: AI should complement, not replace, human professionals who provide contextual and empathetic judgment.
- Informed consent: Residents must be informed about AI usage in their care and have the right to consent.
By addressing these ethical considerations proactively, local authorities aim to harness the benefits of AI and LLMs while safeguarding the rights and well-being of residents. A balanced approach—one that embraces technological innovation while maintaining ethical integrity—will be key to ensuring AI’s responsible integration into preventive social care.
Like many other areas, Kent has been experiencing a significant increase in the number of individuals going into interim and temporary accommodation. Identifying and supporting individuals at risk of homelessness using an early intervention and prevention-focused approach is a priority for forward-thinking councils.
An effective data solution is necessary to identify these at-risk individuals in need of proactive support. Authorities in Kent used Xantura’s OneView system to identify an individual’s circumstances and areas of financial risk. By identifying individuals at risk of becoming homeless, proactive support was able to be put in place to help prevent homelessness.
The project resulted in a 40% reduction in homelessness as a result of risk alerts enabling proactive support. This reduced the administrative burden for Council staff, with 61 days reinvested in working directly with vulnerable citizens, and a potential increase to 160 days with project expansion.
Barriers and mitigations to implementing this stage of the delivery model
Limited access to AI and advanced analytics
- Challenge: some authorities lack the technical infrastructure, expertise, or funding to deploy AI-driven models for identifying at-risk individuals.
- Mitigation: authorities overcoming this issue begin with rule-based prioritisation (e.g., identifying older carers with high care responsibilities), leverage multi-agency case reviews, and explore partnerships with universities, NHS, or tech providers to access AI tools.
Data availability and integration challenges
- Challenge: key indicators of risk (e.g., social isolation, recent falls, carer strain) may sit in disparate systems across health, housing, and social care, limiting the ability to build a holistic risk profile.
- Mitigation: authorities overcoming this challenge strengthen data-sharing agreements with partners, start by integrating high-impact datasets (e.g., hospital admissions, social care assessments), and use secure AI models to analyse unstructured data in case notes.
Ethical and governance concerns around AI-driven risk identification
- Challenge: using AI and machine learning to predict risk may raise concerns about transparency, fairness, and data privacy, potentially delaying implementation.
- Mitigation: authorities overcoming this issue follow national AI ethics frameworks, ensure human oversight in decision-making, and engage residents and frontline staff to validate AI predictions. They pilot models in small, controlled phases to demonstrate accuracy and fairness.
Accurately identifying the right cohort for maximum impact
- Challenge: without robust data, authorities may struggle to determine which group of older people would benefit most from targeted proactive prevention.
- Mitigation: authorities overcoming this challenge use multi-disciplinary case reviews to assess past users of services and identify preventable risk factors (e.g., falls, carer breakdown). Where AI tools are available, they use machine learning to analyse historical trends and predict future demand drivers.
Aligning partner priorities for collaborative identification
- Challenge: different organisations (e.g., NHS, VCFSE, housing) have competing priorities and data ownership concerns, limiting their ability to identify at-risk individuals collectively.
- Mitigation: authorities overcoming this issue establish joint governance structures, develop shared outcome frameworks, and use evidence from small-scale pilots to demonstrate the system-wide benefits of early intervention.
Implications of this stage for evidencing prevention
The decisions made at this stage significantly influence the ability to demonstrate prevention impact:
- Establishing clear baseline data on need and expenditure for the selected cohort
- Developing preventability estimates that inform potential benefit forecasts
- Creating comparison methodologies that will enable impact measurement
Together, these elements form the foundation of the economic and strategic case for prevention, enabling authorities to forecast potential returns on investment with appropriate confidence levels
Summary of key points
- Begin with impact in mind: Select cohorts where prevention will deliver meaningful outcomes for individuals while addressing significant system pressures.
- Use available data effectively: Make use of AI to target support at those who will benefit most from it
- Engage Information Governance colleagues early: Involve data governance specialists from the outset to design appropriate and secure data use processes.
- Balance precision with practicality: Recognise that while perfect targeting is unattainable, evidence-based approaches significantly outperform purely reactive models.
- Maintain the human element: Ensure that technological approaches complement rather than replace professional judgment and person-centred practice
