Introduction

On 16 June 2026, Research in Practice, in partnership with Somerset Council hosted a webinar about Building a single digital view for children and family services. 

Local authorities are searching for ways to join up information about children and families – particularly through the Family First Partnership programme. Many areas are committed to creating a single digital view but get stuck on the practical ‘how’.

The webinar spotlighted Somerset’s journey to build the Transform Family View (TFV), a sophisticated single digital view of the child. The team behind TFV shared their knowledge and insights about:

• Technical architecture.

• Data matching.

• Using artificial intelligence (AI) and other capabilities to generate insights.

• Implementation lessons. 

Film one: Somerset Council's Transform Family View demonstration

Key highlights and system walkthrough

• The Case for Connected Data: An exploration of how data integration reduces the need for families to retell their story, prevents missed early intervention opportunities, and aligns with the Department for Education’s Families First Partnership objectives. 
• User-Centric Accessibility: A live tour of the platform showing how it was built to accommodate varying digital skills, featuring a simplified lookup interface directly accessible via the staff intranet. 
• Vulnerability & Risk Flags: A demonstration of the dashboard's color-coded flag system, which allows practitioners to instantly distinguish between active, current vulnerabilities and historic case information. 
• Multi-Agency Insights: A look at how the portal aggregates vital context across specialized domains, including Education (attendance data, pupil premium markers, and sibling links), Family Networks (identifying influential individuals outside the household), and Timelines (tracking long-term service interactions up to 10 years). 
• Seamless Integration: A guide to how the system securely connects with key social care management platforms (LCS, EHM) and local health records (SIDER) to allow practitioners to self-serve information quickly and securely.

Film two: Data foundations, governance, and platform architecture

Key highlights and technical deep dive

• Sequential Architectural Layers: Josh Pimm (Chief Data and Analytics Officer) details the high-level architecture of the project. He maps out how individual lines of business (e.g., social care and education databases), partner data (police, health, probation), and internal enterprise records (Excel/SharePoint) are funnelled into a centralised environment to handle address-matching and unified individual indexing.
• AI Governance & Safe Data Sharing: Antje Carpenter (Information Governance Lead - South, Central and West Commissioning Support Unit) addresses a critical question for local authorities: Are data protection laws a blocker or an enabler? Moving the needle from "can we share" to "how do we share safely," she outlines a streamlined governance structure spanning three essential pillars:
  • Tool-focused: Assessing AI safety, functionality, and algorithmic bias.
  • Data-focused: Running Data Protection Impact Assessments (DPIAs) to ensure lawful, minimized data processing.
  • Principle-focused: Conducting rigorous ethics assessments regarding public trust, fairness, and discrimination risks.
• Navigating Modern Legal Gateways: The governance walkthrough highlights the statutory obligations and legal frameworks enabling multi-agency safeguarding integration, drawing upon the Data Protection Act 2018, the Digital Economy Act 2017, and updated statutory duties outlined in the Children and Wellbeing and Schools Act 2026.
• The Power of Microsoft Fabric and Medallion Architecture: Michael Antwi (Senior Data Engineer) breaks down the choice to leverage Microsoft Fabric as a unified data platform. He walks through a standard data engineering implementation path (Discovery, Planning, Setup, Implementation, Testing) using a Medallion Architecture:
  • Bronze: Landing raw data as closely aligned to source files as possible in OneLake.
  • Silver: Cleansing, validating, and regularising data with service-owner approval.
  • Gold: Building optimised, curated data models fit for predictive analytics and Power BI dashboards.
• The ROI of Platform Modernisation: A closing look at the real-world operational benefits realised by migrating away from legacy reporting models.

Film three: Data matching, fuzzy logic, and row-level security

Key highlights and technical deep dive

• The Record Linking Challenge: A breakdown of real-world data discrepancies that disrupt simple data integration, including phonetic spelling variations (e.g., Katherine vs. Cathrine), aliases, and out-of-date address histories.
• The 4-Step Hierarchical Matching Process: Somerset map out their custom compute-efficient matching logic (adapted from open-source record linkage principles):
  • Deterministic Match: A quick-win, low-compute layer to instantly catch identical records matching perfectly on name and date of birth.
  • Blocking: A standard indexing method that groups names by characteristics (such as the first letter) to stop the system from running "monstrously large" cross-checks of every record against every other record in the system.
  • Probabilistic Match: A fuzzy-matching layer that applies statistical weights to different data points (e.g., an 80% forename match + exact surname + address history match) to confirm identities without over-fitting.
  • Manual Corrector System: A critical human-in-the-loop feedback process allowing practitioners to flag and immediately break an incorrect automated match.
• Dynamic Flag Architecture: Reuben explains how granular multi-agency indicators are rolled into broader "Headline Codes" (such as financial stability or physical health) using a dedicated SQL vulnerability engine.
• Enforcing Row-Level Security (RLS): A look at how Somerset maps data views to specific professional roles within Power BI. For instance, an internal social worker may see specific child protection details, whereas an external school partner sees a sanitised, aggregated flag (e.g., "Crime") to keep data sharing proportionate and lawful.

Film four: The AI intelligence layer and future horizons

Key highlights and technical deep dive

• Shifting from Reaction to Prevention: Josh Pimm (Chief Data and Analytics Officer) introduces the core philosophy driving Version 3 of the platform: the road to prevention is through prediction. By engineering advanced machine learning and deep neural networks on top of their master dataset, the council aims to flag escalating risks earlier, allowing practitioners to triage support before crisis points are reached.
• Unlocking Unstructured Case Notes: While previous versions relied primarily on structured tables, the richest contextual information often lives within unstructured text (case notes, practitioner logs, PDFs, and emails). The team outlines how they are utilizing Generative AI and Natural Language Processing (NLP) to parse these complex narrative histories safely.
• The Strategic Principle of Augmentation: Allan Holly (Senior AI Engineer) emphasises a vital guardrail of their implementation: human judgment remains central. AI is strictly treated as a supplementary enhancement to support professional decision-making and reduce administrative friction—never to automate or replace human statutory conclusions.
• Four Core Generative AI Use Cases: Allan details how large language models (LLMs) add direct value to practitioner single-view apps:
  • Explaining the Story: Moving beyond text-heavy database fields to generate clear, concise narrative summaries of a client's history, context, and active vulnerabilities prior to worker visits.
  • Answering Questions: Integrating a natural language Q&A assistant so workers can safely interrogate record databases using standard conversational prompts.
  • Extracting Signals: Automatically reading unstructured notes to pull out and categorise specific risk metrics, strengths, and relationships, turning them into standardised data coordinates.
  • Assisting with Actions: Transforming the portal into an action hub by using AI to draft routine administrative summaries, email notifications, and placeholder referral forms (with all final approvals remaining under strict practitioner control).
• Retrieval Paths & Ingestion Timing (RAG vs. Agents): A technical blueprint of how queries are routed depending on data structure:
  • Structured data: Handled via Fabric Data Agents that dynamically translate natural language requests into structured SQL or DAX table queries.
  • Unstructured data: Handled via Vectorisation and Retrieval-Augmented Generation (RAG) to extract contextual text snippets alongside precise source citations for auditability.
  • Batch vs. Runtime processing: Explaining the trade-offs between cost-efficient, low-latency pre-generated batch pipelines (for stable timelines and history flags) and real-time execution models (required for ad-hoc, dynamic Q&A).
• The Long-Term Horizon: In the closing wrap-up, Josh shares the council's wider vision to scale this framework out of a purely child-centric architecture into an enterprise-wide Single View of the Resident—safely combining data lines across adult social care, municipal contact centres, and local infrastructure frameworks.