No black boxes — every step inspectableOn-premise and air-gapped deployment availableDocumentation ready for scrutiny and transparency recordsFixed fee, priced against a budget cycle

AI systems for the public sector and charities

A public body has to be able to explain itself. Not just to a resident who disagrees with a decision, but to a scrutiny committee, an ombudsman, a tribunal and occasionally a journalist with a Freedom of Information request. That requirement shapes what we build here more than any technical constraint does.

01The problem

Statutory duties, fixed deadlines, and no more staff

The pressure in public bodies and charities is rarely the decision itself. It is the volume of evidence-gathering and correspondence required before anyone can make it.

01

Information requests with a statutory clock

Twenty working days, requests arriving unstructured, material scattered across systems and inboxes, then redaction and exemption assessment before anything goes out. Backlogs build quietly and then become a published compliance statistic.

02

Casework evidence-gathering

Statutory assessments require collating advice and evidence from several parties before a case officer can even start drafting. The bottleneck is almost never the professional judgement — it is the chasing and collation that precedes it.

03

Repairs and hazard reporting against prescribed timescales

Awaab's Law has applied to social landlords since 27 October 2025, covering all emergency hazards and all damp and mould hazards, with more hazard categories phased in during 2026 and the remainder in 2027. Triage against those categories, and evidencing the response, is a new and growing manual load.

Phase 2 expands the categories during 2026
04

Charities running on goodwill and no headroom

Sector research suggests around four in five charities now use AI in some form, while a much smaller proportion of trustees report their charity using it at all. That gap — real usage without board oversight — is a governance problem waiting to surface.

02What we'd build

What we'd build for a public body or charity

Each of these accelerates the work that leads up to a decision. None of them make the decision.

Build 01

Information request triage

Classifies incoming requests, detects duplicates and repeat requesters, locates candidate material across systems, and proposes redactions for an officer to review. The exemption decision and the sign-off stay with a qualified person.

Proposes redactions. Never applies them.
Build 02

Casework evidence collation

Assembles the evidence bundle a case officer needs — chasing contributors, tracking what's outstanding against the statutory clock, and summarising what's arrived. The officer writes the assessment.

Build 03

Repairs and hazard triage

Classifies inbound reports against hazard categories and the prescribed timescales, drafts the tenant acknowledgement, raises the work order and records the evidence trail — escalating anything urgent to a person immediately.

Build 04

Committee and meeting documentation

Drafts reports and minutes from transcripts and source material in your house format, with the officer recommendation left blank for the officer to write.

Build 05

Correspondence and complaints triage

Routes inbound correspondence to the right service, drafts acknowledgements within your response standards, and flags anything that looks like a statutory complaint or an ombudsman risk.

Build 06

Grant applications and funder reporting

For charities: assembles funder reports from data you already hold, and drafts applications against a funder's criteria for a human to shape. The case for support still has to be yours.

03Governance

Transparency, procurement and the rules that actually apply

Public sector AI has a specific accountability overlay that private sector buyers don't face. Getting this wrong is a scrutiny problem, not just a compliance one.

Full governance, data residency and EU AI Act position →

04Hard limits

What we will not build for a public body

If a resident could appeal it, a person decides it. That principle covers almost everything below.

✕ 01

Benefits, welfare or financial eligibility decisions

High-risk under the EU AI Act, recordable under transparency standards where they apply, and the area where automated decision-making has caused the most public harm internationally. We build the evidence-gathering that gets an officer to the decision faster. We don't build the decision.

✕ 02

Housing allocation and priority banding

These determine where people live and in what order. They are appealable, contested and frequently litigated. A person decides, on a record they can explain.

✕ 03

Statutory assessments and safeguarding decisions

Care assessments, children's and adults' safeguarding, special educational needs determinations. Statutory, appealable, and non-delegable. AI may collate the evidence; the professional judgement is the job.

✕ 04

Applying exemptions or signing off redactions

A system can propose redactions and flag likely exemptions. A qualified officer applies the exemption and takes responsibility for what's released — because the consequence of getting it wrong is a data breach or an unlawful withholding.

✕ 05

Enforcement, licensing and regulatory decisions

Anything that removes a permission, imposes a penalty or restricts an activity is a decision a named officer makes and can defend.

05Questions

Common questions.

Do councils have to publish their AI use?

+

Not under the Algorithmic Transparency Recording Standard — that mandate covers central government departments and arm's-length bodies delivering public or frontline services, and is recommended rather than required for the wider public sector. English local authorities sit outside it, though a number publish records voluntarily and it is generally treated as good practice. Separately, if an AI system makes a significant decision about a person without meaningful human involvement, data protection law requires that the person is told and can contest it — which is a different obligation and does apply to councils.

Can AI be used to make benefits or eligibility decisions?

+

It shouldn't be, and we won't build it. Evaluating eligibility for essential public assistance benefits and services is classified as high-risk under Annex III of the EU AI Act, and it is the category of automated decision-making that has caused the most documented public harm internationally. What AI can legitimately do is compress the evidence-gathering that precedes the decision — collating what's been submitted, flagging what's missing, and getting a complete case to an officer faster. The determination stays with the officer.

Can an AI system run entirely on our own infrastructure?

+

Yes. On-premise and fully air-gapped deployment — with no outbound connection at all — is available and is what we'd usually recommend where a body handles sensitive casework or is subject to heightened scrutiny. It costs more and takes longer than a cloud build, typically 8–12 weeks rather than 4–6, and we'll tell you plainly if your use case doesn't warrant it.

How does this fit public procurement?

+

Through the existing frameworks. Crown Commercial Service became the Government Commercial Agency in April 2026, with frameworks and call-offs carrying over unchanged; G-Cloud 14 runs to 28 October 2026 and G-Cloud 15 was awarded in August 2026. We price builds as a fixed fee against a defined scope, which tends to fit a budget cycle and a business case better than a per-seat subscription that grows unpredictably.

Do charity trustees need an AI policy?

+

There is no AI-specific statutory duty on charities, but existing trustee duties — acting with reasonable care and skill, in the charity's best interests, and complying with the law — already apply to AI use. The Charity Governance Code was refreshed in November 2025 and includes a recommended practice that charities have a policy covering the use of technology and AI tools. Given that sector research shows widespread staff-level AI use with much lower board awareness, a short written policy is worth having regardless of what the Code requires.

Related sectors

Start here

Start with the work before the decision.

Tell us which statutory process is running late and why. If the bottleneck is professional judgement rather than administration, we'll say so — that's not a problem AI should be solving.