AI for the public sector that is transparent by default
Citizen service assistants grounded in published policy, caseworker copilots, correspondence and FOI triage, form and grant processing and multilingual access for central government, local authorities, agencies, health bodies and public-sector suppliers.
Public bodies answer the same questions thousands of times, process forms by hand and hold policy in documents nobody can search. Each use case below is measurable in waiting time and staff hours.
Citizen service assistant
Bin collections, permits, benefits eligibility, planning status, opening hours: answered from published policy and service pages, with the source linked, in plain language and multiple languages. Touches the website, the CRM and the contact centre. Anything personal requires verification; anything complex goes to a person.
Caseworker copilot
Case history summarised, the relevant policy and guidance retrieved and cited, next steps suggested, all inside the case system. Touches case management and the policy library. The caseworker decides; the copilot saves the reading. See internal copilot.
Correspondence and FOI triage
Incoming letters, emails and information requests classified, routed to the right team with a draft acknowledgement and deadline tracking. Touches the correspondence system and the records store. Freedom-of-information requests get the clock started on day one.
Form, application and grant processing
Applications and supporting documents extracted, checked for completeness and validated against eligibility rules, with officers reviewing the exceptions and making the decision. Touches the forms platform and the case system. See document processing.
Plain language, translation and accessibility
Letters and guidance rewritten to a reading-age target, translated into community languages and offered by voice, with an officer approving the template. Touches the content management system and the contact centre. See voice AI.
Procurement and contract document analysis
Tender responses summarised against the evaluation criteria for evaluators to score, and supplier contracts checked for obligations and renewal dates. Touches the e-procurement platform and contract register. See contract analysis.
Compliance and data
Procurement, transparency, accessibility and equality
What we design for in the public sector
Public bodies are expected to be transparent about algorithmic tools, to avoid automated decisions with significant effects on individuals, to meet accessibility standards for digital services and to consider equality impacts. Citizen data is personal data under GDPR and UK GDPR, sometimes with security classifications that dictate where it may be processed. Freedom-of-information and records duties mean prompts, outputs and logs may be disclosable and must be retained properly. Procurement rules shape how engagements are structured. This is engineering guidance, not legal advice; your information governance, legal and procurement teams own the policy.
We publish a transparency description with every deployment, keep a person in every decision that affects a citizen, test with assistive technology, log in a form that can be disclosed and deploy inside your tenancy where classification requires. The AI governance and ISO 42001 services give you the management system behind it.
Reference architecture
Flagship workflow: a citizen assistant with a published explanation
Citizens reach the assistant on the website, by voice on the phone line or through a front-desk kiosk. General questions are answered only from published policy and service content, versioned so an answer can be traced to the page as it stood that day. Personal questions require verification and read case status from the case system without changing it.
When the question needs judgement, when the citizen asks for a person, or when a vulnerability indicator appears, the assistant hands over to a caseworker with the transcript. No eligibility or enforcement decision is made by the assistant.
Underneath sits the transparency layer: the published description of the system, an accessible interface tested with assistive technology, and logs retained under records rules. The pattern is a public-service form of our customer support AI with the governance from AI governance made visible.
The assistant answers from what is already public and routes everything else to a person; the explanation is published alongside it.
Track record
What we have built that applies
Grounded assistants with hand-off
Our production support bot answers only from approved content, cites it and escalates to a person, the exact behaviour a citizen assistant needs.
Eventify runs registration, apps and on-site operations for conferences and civic events, with multi-tenant isolation and accessibility in the attendee app.
Can a public body explain to citizens how the AI works?
It has to, and we design for it. Each deployment ships with a plain-language description of what the system does, what data it uses, where a person is involved and how to challenge an outcome, suitable for publication in an algorithmic transparency record. No adverse decision about a citizen is made automatically.
How do you meet accessibility requirements?
Public sector digital services must meet accessibility standards, so assistants are built to work with screen readers, keyboard navigation and plain language, and are tested with assistive technology before launch. Voice and text channels are offered side by side where possible. See voice AI.
Where does the data live?
Inside your tenancy or on-premise where security classification requires, with hosted models only where the data classification and your procurement terms allow. Retrieval is limited to published or appropriately classified material. See secure RAG architecture.
How does this fit public procurement?
We scope fixed-price discovery and pilot phases with clear deliverables, which fit framework call-offs and below-threshold procurements, and we document outputs so a later competitive procurement can build on them. Procurement rules are yours to apply; we make the engagement easy to describe and evaluate.
Ready to make AI work in production?
Tell us about your systems, your data and your constraints. We reply within one business day with a straight answer and a suggested next step.