Think Family Database and Insight Bristol's Risk-Scoring Models
This overview was generated with AI from the public sources listed below and checked against them before publication. It has not had a complete human fact-check — treat it as a starting point and follow the sources.
How this atlas is made · What “last verified” means · What “source-checked” means · The full record behind this page
Bristol City Council set up a data team, Insight Bristol, in 2015, staffed jointly with Avon and Somerset Police. From 2016 it ran the Think Family Database: a single view of a family drawn from around 35 council, police, health, benefits and school datasets, including data on ethnic origin and criminal offences, processed under legal gateways rather than consent. Machine-learning models scored children for risk of sexual exploitation, criminal exploitation, and being out of education, employment or training. Two of the three were switched off in 2023. The database itself is still running.
what does this mean?
The strongest source is an independent non-academic evaluation: an audit office, a government or EU review, or an NGO. It grades who did the looking — not how well the project went. All four grades →📊 Impact
The council's own page puts the database at around 55,000 families. In 2021 it answered a freedom-of-information request saying around 450 children were flagged as at risk of sexual exploitation by the model, at a precision of around 80 per cent; the campaign group Big Brother Watch said about 90 of them may have been wrongly flagged. The one surviving model, for risk of not being in education, employment or training, is entered on the UK government's Algorithmic Transparency Recording Standard: around 150 safeguarding professionals use the front-end report each month, the score sits among more than 70 other indicators of vulnerability, and the record states there is no automated decision making based on it.
🎓 Lesson
Council staff judged the two exploitation models not fit for operational use and stopped them in 2023. An independent evaluation by the non-profit Social Finance, commissioned by Somerset Council in partnership with Bristol and published in June 2025, called the risk-scoring models the weakest element of the project and said a lack of accuracy had undermined them; staff reported that children who should have been identified as vulnerable were not listed in the results. Precision was the number the council could quote; recall was the failure that mattered. What stopped was the scoring, not the collection: the same feeds still assemble under the same legal gateways, and records responses suggest neither the council nor the police kept a record of the decision to stop.
Sources
Show 5 more sourcesShow fewer
📅 How old is the evidence?
Two clocks, kept apart: when this entry’s sources were published, and what has happened to its record in the atlas. Documented span: since 2016. 8 of 10 sources carry a publication date, from 8 April 2021 to 25 June 2026.
When the sources were published
- GOV.UK Supporting Families programme blog — reducing youth crime and exploitation in Bristol (national government)
- The Bristol Cable — how a police and council database is predicting if your child is at risk of harm (media)
- IICSA, Child sexual exploitation by organised networks, H.7 Predictive analytics (statutory inquiry)
- GOV.UK Algorithmic Transparency Recording Standard — Bristol City Council NEET model (public register)
- Social Finance evaluation of Bristol's Think Family Database, via Data to Insight (independent evaluation)
- The Bristol Cable — Think Family and the fight for transparency over data profiling (media)
- Liberty Investigates, with WIRED, Lighthouse Reports and the Bristol Cable (media investigation)
- The Bristol Cable — Children's Commissioner 'deeply concerned' over Bristol data tools (media)
Show your work What this page rests on and what it can’t tell you · all 22 claims with their sources · the forensic record
Evidence Passport · public beta 22 addressable claims, 10 source dossiers, one permanent baseline. The passport is unreviewed — an AI-assisted structural migration, not a human or expert endorsement. Every revision keeps a permanent copy under revision history.
Inspect the claims, one by one — verbatim text, sources and caveats for all 22
The text below is copied verbatim from the published record. Each claim names the source dossiers that carry it; caveats stay attached to the claim they qualify.
clm-a2de878acab015ee
quantitative
Bristol City Council set up a data team, Insight Bristol, in 2015, staffed jointly with Avon and Somerset Police
Sources carrying this claim
clm-a9864c655d6ace23
quantitative
From 2016 it ran the Think Family Database
Sources carrying this claim
clm-0378ad87af736d60
quantitative
a single view of a family drawn from around 35 council, police, health, benefits and school datasets
Sources carrying this claim
clm-7bc8a0b6af2ae7f6
descriptive
including data on ethnic origin and criminal offences
Sources carrying this claim
clm-54b025581a43c408
descriptive
processed under legal gateways rather than consent
Sources carrying this claim
clm-cab712a9538665c4
descriptive
Machine-learning models scored children for risk of sexual exploitation, criminal exploitation, and being out of education, employment or training
Sources carrying this claim
clm-842f8383e37405ad
quantitative
Two of the three were switched off in 2023
Sources carrying this claim
clm-fd234828bef30c4e
descriptive
The database itself is still running
Sources carrying this claim
clm-18dd42cd3a286566
quantitative
The council's own page puts the database at around 55,000 families
Sources carrying this claim
clm-00efab65aff2af6e
quantitative
In 2021 it answered a freedom-of-information request saying around 450 children were flagged as at risk of sexual exploitation by the model
Sources carrying this claim
clm-1bcce680a48fd33c
quantitative
at a precision of around 80 per cent
Sources carrying this claim
clm-2243dda409a448d3
quantitative
the campaign group Big Brother Watch said about 90 of them may have been wrongly flagged
Sources carrying this claim
clm-fcbcd26e90c67335
descriptive
The one surviving model, for risk of not being in education, employment or training, is entered on the UK government's Algorithmic Transparency Recording Standard
Sources carrying this claim
clm-23e19ee2c7700e19
quantitative
around 150 safeguarding professionals use the front-end report each month
Sources carrying this claim
clm-d01dac650bf8496d
quantitative
the score sits among more than 70 other indicators of vulnerability
Sources carrying this claim
clm-59f111a7a18c6bc3
descriptive
the record states there is no automated decision making based on it
Sources carrying this claim
clm-aceb4a17173e29ed
quantitative
Council staff judged the two exploitation models not fit for operational use and stopped them in 2023
Sources carrying this claim
clm-6fa8c0f9893000d5
quantitative
An independent evaluation by the non-profit Social Finance, commissioned by Somerset Council in partnership with Bristol and published in June 2025
Sources carrying this claim
clm-6f0b7be8d6e38346
descriptive
called the risk-scoring models the weakest element of the project and said a lack of accuracy had undermined them
Sources carrying this claim
clm-ff8c2a58b9aa30ff
descriptive
staff reported that children who should have been identified as vulnerable were not listed in the results
Sources carrying this claim
clm-c6346e995f7d555e
descriptive
the same feeds still assemble under the same legal gateways
Sources carrying this claim
clm-32742e0695d170eb
descriptive
records responses suggest neither the council nor the police kept a record of the decision to stop
Sources carrying this claim
Forensic record — source dossiers, gaps, hashes and the exact revision this page renders
This layer records what is known about each source, what is still missing, and the exact snapshot this page is rendering. A missing date, archive or hash stays visible as missing.
Awaiting operator and domain-expert review; publication of this revision implies neither.
sha256:ca9092d7e85a4cfbd065e14a46a77b485b8f71fecdca21dc56d1e5feb4a001ae
- 22
- of 22 claims mapped
- 10
- source dossiers
- 7
- publication dates recorded
- 0
- archives linked
- 0
- exact locators
- 0
- content captures hashed
Governance facts carried by the record
- Operator
- Bristol City Council (Insight Bristol, a data team staffed jointly with Avon and Somerset Police)
- Actor types
- city-government, regional-government
- Funding model
- public
- Funding description
- Bristol City Council and Avon and Somerset Police; the team was created under the national Troubled Families, now Supporting Families, Programme
- Cost
- Not recorded
- Ownership
- Not recorded
- Decision rights
- Not recorded
- Exit conditions
- Not recorded
10 source dossiers
Bristol City Council — Insight Bristol and the Think Family Database (operator)
- Stable source ID
src-21cc8ef43151fb5c- Publisher
- Name not classified · bristol.gov.uk
- Type / language
- web-page · language not classified
- Published
- Not recorded
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
Operator's own service page. States Insight Bristol is a Bristol City Council data team set up in 2015 in response to the national Troubled Families Programme; that the Think Family Database holds information from approximately 55,000 families; lists the contributing bodies (Bristol City Council, Avon and Somerset Police, Department for Education, Department for Work and Pensions, SCWCSU, Barnardo's, NHS); and describes a surviving 'Risk of NEET data model'.
Bristol City Council — Insight Bristol privacy notice (operator, PDF)
- Stable source ID
src-da72b54af61289ad- Publisher
- Name not classified · bristol.gov.uk
- Type / language
- web-page · language not classified
- Published
- Not recorded
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
Statutory privacy notice for the Insight Bristol products. Lists the data categories held, including personal data revealing racial or ethnic origin and data relating to criminal offences and incidents; names the legal bases as legal obligation, official authority, social protection and substantial public interest, and enumerates the 'legal gateways' relied on; describes the Think Family Database, Children and Families View, Think Family Education app and the NEET modelling; and records the onward sharing to Avon and Somerset Police's Lighthouse safeguarding unit. Text extracted locally from the PDF; the hosted file is a scanned-style PDF that summarising fetchers cannot read.
GOV.UK Algorithmic Transparency Recording Standard — Bristol City Council NEET model (public register)
- Stable source ID
src-561498af9ff7b169- Publisher
- Name not classified · gov.uk
- Type / language
- web-page · language not classified
- Published
- 2025-01-28
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
The UK government's public algorithm register entry for the one model still in use. Carries the ~150 safeguarding professionals a month figure, the 'over 70 other indicators of vulnerabilities' framing, the statement that there is no automated decision making based on the model, the precision-and-recall review commitment, and a Data Protection Impact Assessment created 01/11/2018 and last updated July 2024.
IICSA, Child sexual exploitation by organised networks, H.7 Predictive analytics (statutory inquiry)
- Stable source ID
src-34debd022745f267- Publisher
- Name not classified · iicsa.org.uk
- Type / language
- web-page · language not classified
- Published
- Not recorded
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
Statutory public inquiry. Bristol was one of six case study areas. Section H.7 describes the Think Family Database bringing together 35 different datasets and the Insight team, funded jointly with the police, producing a weekly automated list of children at heightened risk of sexual exploitation, supplied to Avon and Somerset Police's Operation Topaz. The host returns HTTP 403 to every automated client tried (browser user agent, curl, and the UK Government Web Archive mirror); existence and wording established by exact-phrase search only.
- Claims linked
Social Finance evaluation of Bristol's Think Family Database, via Data to Insight (independent evaluation)
- Stable source ID
src-617fc9f5322b5974- Publisher
- Name not classified · datatoinsight.org
- Type / language
- web-page · language not classified
- Published
- 2025-06-23
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
Publication page for the independent evaluation: 'Somerset Council, in partnership with Bristol Council, commissioned Social Finance to deliver an independent evaluation of the implementation and impact of single view products being used in Bristol'. The linked PDF is glyph-subset encoded with no ToUnicode map, so its text could not be extracted; its findings are quoted here from the journalism that obtained and read it.
- Claims linked
Liberty Investigates, with WIRED, Lighthouse Reports and the Bristol Cable (media investigation)
- Stable source ID
src-1e8b7205f4563aff- Publisher
- Name not classified · libertyinvestigates.org.uk
- Type / language
- web-page · language not classified
- Published
- 2026-06-25
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
Joint investigation with WIRED, Lighthouse Reports and the Bristol Cable, built on hundreds of pages obtained under public records requests. Carries the Social Finance review's 'weakest element' and 'lack of accuracy' wording, the staff report that children who should have been identified 'were not listed', the 2016 launch by Bristol City Council and Avon and Somerset Police, and the finding that neither authority kept records of the decision to stop the CSE and CCE models by June 2023. Most of the article concerns Avon and Somerset Police's own force-wide models, which are a separate system and are not described in this entry.
The Bristol Cable — Children's Commissioner 'deeply concerned' over Bristol data tools (media)
- Stable source ID
src-0ea2b67e6e8c6784- Publisher
- Name not classified · thebristolcable.org
- Type / language
- web-page · language not classified
- Published
- 2026-06-25
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
Carries the 2016 launch and that the database 'is still in operation today'; the Child Sexual Exploitation and Child Criminal Exploitation models and the risk factors feeding them; the council's 'not fit for operational use' conclusion and 2023 shutdown; the Children's Commissioner's response; and the council statement that it now uses no predictive analytics other than the NEET model.
The Bristol Cable — Think Family and the fight for transparency over data profiling (media)
- Stable source ID
src-9d5c846edd944188- Publisher
- Name not classified · thebristolcable.org
- Type / language
- web-page · language not classified
- Published
- 2026-01-16
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
Carries the 55,000 families and approximately 35 datasets figures, the May 2023 shutdown date for the two models, the Centre for Data Ethics and Innovation's 2021 findings on bias and on 'legal gateways', and Cllr Christine Townsend's statement that the authority uses no predictive analytics beyond NEET identification.
The Bristol Cable — how a police and council database is predicting if your child is at risk of harm (media)
- Stable source ID
src-dc8840e820011cae- Publisher
- Name not classified · thebristolcable.org
- Type / language
- web-page · language not classified
- Published
- 2021-07-20
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
The original exposure, built on Big Brother Watch freedom-of-information requests. Carries the three models, the roughly 450 children flagged as at risk of sexual exploitation as of June 2021, the council's stated precision of around 80 per cent, Big Brother Watch's estimate that about 90 may have been wrongly flagged, and the council's response that race is not an input.
GOV.UK Supporting Families programme blog — reducing youth crime and exploitation in Bristol (national government)
- Stable source ID
src-848eb13bdf87e9cb- Publisher
- Name not classified · supportingfamilies.blog.gov.uk
- Type / language
- web-page · language not classified
- Published
- 2021-04-08
- Updated
- Not recorded
- Source last checked
- No source-level check date
- Record last verified
- 2026-08-25
- Dossier created
- 2026-08-25
- Retrieved copy
- No local capture recorded
- Archive
- No archived copy linked
- Content hash
- No captured content hash
- Locator
- Descriptive only — not page/paragraph exact
Open the recorded source note
National government blog confirming Bristol's Insight Analytical Hub and the pre-existing Think Family Database, and a further targeted analytical product going live in November 2019 under the Safer Options programme.
- Claims linked
- Background source; no baseline claim points here
Known limits of this passport
- Publisher names and source languages have not yet been classified; domains are recorded without guessing.
- Existing source notes are descriptive locators, not page- or paragraph-exact citations.
- A missing archive, publication date or content hash is shown as null rather than inferred.
- The project-level evidence grade ranks the strongest source in the entry; it is not a truth score for every claim.
Continue exploring
Compare this project
Explore the wider context
📎 Cite this project
1001 Smart Cities (2026). “Think Family Database and Insight Bristol's Risk-Scoring Models” — Bristol, United Kingdom. The Smart City Atlas. https://1001smartcities.org/projects/bristol-think-family-database/ (last verified 2026-08-25). Data: CC BY 4.0.
The underlying data is free to reuse with attribution — see the open data page.
