Mapping Social Enterprises at Scale

1. Background

Community Interest Companies (CICs) are a legal form designed for social enterprises that want to use their profits and assets for the public good. When a CIC is registered with Companies House, its founders must submit a CIC36 (or CIC37) incorporation form describing the company's intended activities, the communities it will benefit, how those communities will benefit, and how any surplus will be used. These four free-text fields represent a uniquely rich—and largely untapped—source of data on what social enterprises set out to do and who they set out to serve, written in their founders' own words at the point of formation.

Using AI-powered text extraction, we have processed CIC incorporation PDFs filed at Companies House since the CIC form was introduced in 2005. The chart below shows the proportion of CICs registered each year for which we have successfully extracted incorporation documents, benchmarked against the total CIC population recorded in the CSO Spine database. Coverage exceeds 80% for most years, with lower rates in the earliest years (fewer PDFs available digitally). Coverage dips in 2022 (to 51%) due to an incomplete batch of PDF downloads from Companies House—the extraction pipeline itself processed over 99% of available documents. Additional PDFs are being retrieved to close this gap.

Proportion of CICs registered each year with extracted incorporation documents, coloured by government in office. Denominator: CSO Spine.

2. Classification Methodology

The analysis pipeline has four stages. First, we download the full corpus of CIC36/37 incorporation PDFs from Companies House via their bulk data API. Second, each PDF is processed through OpenAI's GPT-4o-mini model using structured extraction prompts, which reliably separates the four free-text fields—activities, beneficiaries, community benefit, and surplus use—from the form layout, handling variations in formatting, handwriting, and scan quality. Third, we classify each extracted text against an extended version of the UK Charity Activity Taxonomy (UKCAT), a hierarchical classification system originally developed for the charity sector. We have augmented UKCAT with 14 new beneficiary categories designed specifically for CIC language (covering groups such as local communities, people with mental health needs, those in poverty, carers, and civil society organisations). Classification uses pre-compiled regular expression patterns, matching against both the beneficiary field alone (“specific” match) and a combined text of all four fields (“broad” match) to maximise coverage. Fourth, we perform statistical analysis of the classified data, examining activity sectors, beneficiary populations, and geographic and temporal patterns across two decades of CIC formation.

Classification Coverage

3. Activity Sectors

CICs span a wide range of activity sectors, with health, education, and training consistently among the most common. The charts below show the five activity sectors whose share of new CIC registrations has risen the most (left panel) and declined the most (right panel) over the past two decades, while the bar chart summarises the overall distribution.

Activity sectors with the largest increase (left) and decrease (right) in share of CIC registrations between 2005 and 2025. Dashed lines mark changes of government.

Top 10 activity sectors across all CIC registrations.

Most Distinctive Activity Sector by Region

The tile map below highlights what makes each region different. For each region, we compare the share of CICs in a given activity sector to the national average. The sector shown is the one that is most over-represented locally—i.e., the sector where CICs in that region are most disproportionately concentrated compared to the UK as a whole. The two most common national categories are excluded to surface genuinely distinctive regional patterns.

4. Beneficiary Populations

Who do CICs set out to serve? The beneficiary analysis reveals that the general public and local communities are the most commonly named beneficiaries, followed by people with mental health needs and those experiencing poverty or disadvantage. The charts below show the five beneficiary groups whose share has risen the most (left panel) and declined the most (right panel) over the past two decades.

Beneficiary groups with the largest increase (left) and decrease (right) in share of CIC registrations between 2005 and 2025. Dashed lines mark changes of government.

Top 10 beneficiary groups across all CIC registrations (specific match on beneficiaries field).

Most Distinctive Beneficiary Group by Region

Using the same approach, this map shows the beneficiary group that is most over-represented in each region relative to the national average. This highlights which populations are disproportionately served by CICs in different parts of the UK, after excluding the two most common national categories.

5. Next Steps

This analysis demonstrates what is possible when CIC incorporation data is extracted and classified at scale. Several avenues for further research and collaboration are available:

• Topic modelling: Using embedding-based topic models (BERTopic) to discover thematic clusters in CIC activities beyond the UKCAT taxonomy—revealing emerging sectors and niche specialisations that predefined categories miss.
• Financial linkage: Joining CIC incorporation data to annual accounts filed at Companies House, enabling analysis of income, expenditure, and financial sustainability by activity sector and beneficiary group.
• Survival analysis: Tracking which CICs remain active over time and identifying factors associated with longevity or dissolution—particularly by sector, region, and era of formation.
• Comparison with charities: Comparing CIC activity and beneficiary profiles with those of registered charities to understand how the two legal forms serve different (or overlapping) populations and sectors.
• Geographic deep dives: Detailed analysis of CIC formation patterns at local authority level, examining the relationship between CIC activity and local deprivation, funding landscape, and existing civil society infrastructure.
• Longitudinal text analysis: Tracking how the language CIC founders use to describe their missions has evolved over 20 years—detecting emerging terms, shifting priorities, and responses to policy changes.

Appendices

A1. Sample Narratives

Ten randomly selected examples of the activities and beneficiaries fields extracted from CIC36/37 incorporation forms. These illustrate the range of language, length, and specificity found in the dataset.

A2. Activity Tag Frequencies

All UKCAT activity tags matched in the dataset, ordered by frequency. Percentages are calculated over all 62,088 CIC incorporations.

A3. Beneficiary Tag Frequencies

All UKCAT beneficiary tags matched in the dataset, ordered by frequency. Percentages are calculated over all 62,088 CIC incorporations.