AI Usage Policy

Executive Summary
For Local Authorities and Commissioners

Smart Care Intel uses a safe, responsible and transparent hybrid approach to data analysis. Our core intelligence is powered by secure in-house processing, not AI. Artificial Intelligence is only used sparingly and only on small, fully de-identified snippets of text after all sensitive processing is completed internally.

We never send personal information, care records, full reports, addresses, or any identifiable details to external AI services. The majority of the platform continues to operate normally even if external AI is unavailable. AI enhances clarity; it never generates ratings, determines outcomes or creates fictional results.

A transparent explanation of how Smart Care Intel combines in-house analytics with carefully controlled AI assistance, while maintaining full compliance with UK data protection and regulatory expectations.

Core Approach

Our Hybrid Intelligence Model

Smart Care Intel uses a hybrid intelligence approach where the majority of analysis is performed using our own secure, deterministic and verifiable systems.

The following are processed entirely in-house, within our UK infrastructure:

  • Structured extraction and cleaning of uploaded inspection documents
  • Mapping data to the 34 CQC Quality Statements
  • Comprehensive SQL-based scoring logic
  • Internal rules for NHS best practice, HSE guidance and regulatory expectations
  • Pattern recognition using Elasticsearch across 100,000+ KLOEs
  • Full-text search and phrase detection performed directly in SQL
  • Timeline analysis, cross-domain linking and quality statement modelling

This means the system relies first and foremost on deterministic, replicable and fully transparent logic that we control. AI is used only where it adds an expert-like interpretation layer on top.

Data Flows to AI

What We Send (and Do Not Send) to AI

Smart Care Intel does not send full CQC reports, complete documents, PDFs, raw text uploads, or large unprocessed data to external AI systems.

Before any information is sent to AI, the platform:

  • Extracts content internally using our secure parsing pipeline
  • Reduces it to targeted statement-level segments
  • Strips out names, identifiers, and sensitive information
  • Retains only the minimum non-identifiable text required
  • Ensures the content cannot be reconstructed into the original document

In line with UK GDPR, the Data Protection Act 2018 and CQC information governance guidance, Smart Care Intel ensures AI only ever receives de-identified, minimal data.

We never send to AI:

  • Resident information
  • Staff names or personal details
  • Care plans or clinical/medical data
  • Addresses, DOBs or unique identifiers
  • Uploaded PDFs or full report content
  • Internal organisational documents

What AI receives is strictly limited to:

  • De-identified text fragments
  • Statement-level summaries
  • High-level descriptions of practice
  • Context-free content relevant only to CQC grading characteristics

AI receives only short, sanitised, statement-relevant excerpts — never entire files or identifiable records.

Data & Analytics

Our Advanced In-House Analysis Tools

Smart Care Intel maintains one of the largest structured datasets of care quality intelligence in the UK:

  • Over 33 million+ CQC data points at our disposal

These datasets power highly accurate sector-wide statistics, trend analysis, risk indicators and predictive modelling — all run internally without any AI processing.

  • Elasticsearch for wide-scale pattern analysis across 100,000+ KLOEs
  • SQL full-text search for linguistic and phrase patterning
  • Rules engines based on NHS Best Practice, NICE, MHRA and HSE guidance
  • CQC-aligned scoring and mapping logic built on fixed criteria

These tools ensure that Smart Care Intel does not infer, imply, fabricate or invent results. All in-house outputs rely on rules, evidence and structured logic.

Organisational Intelligence
Extensive Companies House & HMRC Intelligence (Not Processed by AI)

Alongside CQC datasets, Smart Care Intel maintains substantial structured datasets from Companies House and HMRC, including:

  • 18.6 million company records
  • 52 million director and officer records
  • Financial filings and accounting periods
  • Charges, mortgages and secured lending
  • Insolvency notices and administration filings

These datasets support organisational intelligence, governance checks and due-diligence reporting. No part of these datasets is ever sent to AI. All processing is deterministic, rule-based and performed entirely within our secure environment.

AI Role

Why We Use AI

After our system has completed all core analysis, AI is used only as an interpretive assistant to improve clarity, consistency and readability.

  • Explaining how short excerpts relate to CQC grading characteristics
  • Highlighting missing context or inconsistencies
  • Proposing improvement suggestions aligned with Quality Statements
  • Performing consistency checks across multiple statements

AI does not replace our algorithms; it adds a narrative layer on top.

Decision Boundaries

AI Is Never a Decision Maker

AI does not determine ratings, override rules, or create findings that do not exist in the source data.

All final scoring, outcomes and classifications are produced by our own SQL logic, rules engines and compliance frameworks.

Clinical Boundary

No Clinical or Medical Advice Provided

Smart Care Intel is a compliance, governance and quality analysis platform. It does not provide medical advice, clinical decision-making, diagnosis, treatment guidance or any form of professional healthcare instruction.

The application is intentionally restricted to prevent the generation of clinical or medical advice. If such advice is requested, the platform will actively refuse and redirect users to appropriate professional channels.

All outputs relate solely to quality, governance, safety, regulatory compliance and organisational good practice — never to individual medical assessment or treatment.

Resilience

Operational Resilience

Smart Care Intel continues to operate normally even if external AI services are temporarily unavailable. All core intelligence — scoring, timelines, extraction, mapping, SQL analysis and Elasticsearch — remains fully functional.

Only optional enhancement features require AI to help construct the appropriate sentences:

  • Generated procedures
  • Narrative summaries
  • Deep comparison reporting
  • Statement-level interpretation checks

These are optional enhancements — not essential functions. The assessment workflow remains operational.

Governance & Ethics

AI Governance, Ethics and Responsible Use

Smart Care Intel follows strict AI governance principles aligned with Local Authority DPIA expectations, NHS DSPT standards and UK information governance best practice. Our approach to AI use is governed by:

  • Full logging of AI prompts, responses and metadata for audit purposes
  • Strict model versioning to ensure consistency and reproducibility
  • No automated decision-making — all decisions use deterministic rules
  • Internal validity checks for all AI outputs before they are stored or displayed
  • Regular monitoring for bias, drift and unintended behaviours
  • Use of UK English terminology and alignment with UK regulatory language
  • Data minimisation and privacy by design
  • Strict de-identification of all content sent to AI

AI is always used as a controlled, supervised enhancement layer — never as an unmonitored, authoritative or opaque component of the platform. If you would like a more detailed breakdown of our AI architecture, security controls or in-house analytics, we can provide a full technical briefing.

Comparison

Why Smart Care Intel Is Not the Same as Using a General AI Chat Engine

It is technically possible for someone to copy and paste a question into a public AI chatbot. However, doing so would not provide safe, accurate or reliable compliance analysis — and it would not meet information governance expectations.

General-purpose chat engines:

  • Are not bound to CQC frameworks, KLOEs or Quality Statements
  • Do not apply NHS, HSE, NICE or sector-specific rules
  • Have no access to our 33+ million structured data points
  • Cannot cross-check responses against historical CQC intelligence
  • May invent content (“hallucinations”) without referencing real evidence
  • Do not follow our de-identification and minimisation controls
  • Provide non-auditable, non-reproducible results

Smart Care Intel, by contrast, operates within a fenced, controlled and evidence-aligned environment. Every AI request is:

  • Strictly de-identified before processing
  • Bounded by the correct CQC Quality Statement
  • Cross-referenced with SQL scoring, timelines and rule engines
  • Validated against structured data, not free-form AI inference
  • Audited, logged and version-controlled for governance purposes

This ensures that Smart Care Intel's outputs are accurate, repeatable, safe and compliant — something that cannot be achieved by manually asking questions to an unrestricted public AI service.