When a town hall buys software that scores benefits claims, schedules housing inspections, or routes call-centre queues, that purchase is not “IT housekeeping”—it is governance. We should treat every consequential algorithm contract like any other piece of public infrastructure: published, searchable, and challengeable before it touches residents’ lives.
The claim
Local authorities must list algorithmic and automated decision tools in public procurement registers with enough technical detail for journalists, councillors, and community groups to understand what changed. Secrecy wrapped in vendor NDAs is incompatible with democratic oversight, even when models are hosted in “private” clouds.
What reporting already shows
Our technology desk’s coverage of government chatbot renewals and AI invoice matching in accountancy firms shows how quickly automation spreads from Whitehall to supplier chains. Councils are not far behind: homelessness prevention pilots, damp-and-mould sensor analytics, and benefits triage tools are entering frameworks advertised as “digital transformation.”
Yet Contracts Finder entries often describe these buys as generic “case management upgrades,” burying the fact that a model ranks vulnerability or suggests enforcement action. Without a plain-language model card—training data boundaries, override rights, appeal paths—residents discover automation only when a letter arrives with an unexplained score.
The UK’s Algorithmic Transparency Recording Standard exists for central government; local adoption is patchy. Procurement law already requires publication above thresholds; the gap is semantic honesty about what is being purchased.
The objection
Vendors argue that detailed disclosure aids competitors and exposes security-sensitive architectures. Some council lawyers fear publishing performance metrics invites judicial review before pilots finish. Both concerns are manageable: redact credentials, publish evaluation metrics and human override rates, and time-box pilots with sunset clauses in the contract.
Openness does not mean dumping source code on GitHub; it means telling a tenant why an inspection was prioritised and which officer can overturn a model suggestion. That is baseline accountability, not anti-innovation Luddism.
What should happen
MHCLG should extend transparency standards to local government with a mandatory procurement tag for “automated decision support,” triggering a short public summary at award. Councils should require vendors to accept freedom-of-information clauses that cover model documentation, not just licence fees. Audit committees should hear an annual algorithm register alongside treasury outturns.
Residents deserve the same visibility into scoring tools that they expect for planning applications on noticeboards. If a model cannot survive public description, it should not be deployed on benefits or housing queues.
We are not asking councils to halt useful automation—we are asking them to stop pretending it is a back-office spreadsheet. Put the contracts where citizens already look: the register, not the slide deck.
Councillors who approve these frameworks should demand a one-page resident summary at award—what data fields enter the model, who signs off on overrides, and how long logs are kept. Without that, “digital transformation” becomes a black box with a council crest on the invoice.
Transparency is cheaper than litigation after a wrongful enforcement letter. The register is already there; use it honestly.
Journalists and tenant unions should not need freedom-of-information fights to learn which postcode ranks a mould inspection. Publish the award; name the model class; defend the policy in daylight.
