AI in the public sector: what local councils can automate

In a town hall, a good part of the day goes on the same questions: which documents are needed for a planning certificate, when local taxes are due, which counter a request is submitted at, what the registry's opening hours are. At the same time, requests and petitions have to be registered, assigned and resolved on time. Artificial intelligence does not replace officials, but it can take over a significant share of the repetitive work.

The guide below is for mayors, chief executives, heads of IT and councillors who want to understand what AI can concretely do in local councils, what it should not do and how to start a project without turning it into a purchase nobody ever touches again.

What AI can automate in a local council

Before the list, one clarification: "AI" here does not mean a robot running the council. It means a language model that understands questions and documents, written or spoken in Romanian, connected to the institution's systems through automation. The AI part understands and phrases, and the automation moves data from one place to another. Together they take over the repetitive work.

1. Informing citizens, 24/7

A chatbot on the website or WhatsApp, or a voice agent on the phone line, can answer frequently asked questions: required documents, fees, deadlines, opening hours, locations. The answers come from a knowledge base approved by the council, not from "whatever the AI knows". We cover this scenario in detail in our article on chatbots and voice agents for local councils.

2. Triaging requests and petitions

Requests arrive by email, through online forms and at the registry. AI can read a request, identify its type (planning, taxes, waste collection, social care), extract the essential details and suggest the department it belongs to. The official validates rather than typing. This matters because petitions have legal deadlines: under Government Ordinance (OG) no. 27/2002, a reply must be given within 30 days of registration, with a possible extension of no more than 15 days in complex cases.

3. Tracking deadlines

An automation can send reminders to departments as a reply deadline approaches, plus a weekly report of overdue requests. Nothing spectacular, but exactly the kind of thing that slips when volume grows.

4. Drafting replies

For standard replies (acknowledgements of receipt, requests to complete a file, notices), AI can prepare a draft based on an approved template. The official checks it, adjusts it and signs it. Responsibility stays with the person.

5. Searching internal documents

Council decisions, regulations, internal procedures: an internal assistant can quickly find where something is written, with a reference to the source document. Especially useful for new staff and for small departments where one person covers several areas.

6. Counter appointments

A voice agent or chatbot can book appointments for civil registration, taxes or planning directly in the department's calendar and send a confirmation by text or WhatsApp. Queues shrink without anyone having to sit on the phone.

Examples by department

You can see a pattern: AI takes on information and sorting, while everything involving assessment and decisions stays with people.

What AI should not do in a local council

• Make administrative decisions. Issuing a document, approving or rejecting a request stays with the competent official.
• Make up answers. A well-built system answers only from approved sources and says "I don't know, I'll put you in touch with a colleague" when it does not have the information.
• Work without a log. Every conversation and every action must be recorded so it can be checked.
• Hide. Citizens must know they are talking to an AI system. From 2 August 2026, Article 50(1) of the EU Artificial Intelligence Regulation (AI Act) requires providers of systems that interact directly with people to inform them that they are talking to an AI system.

Personal data and security

A local council processes personal data in almost every interaction. Before any AI project, be clear on:

• where the data is stored (preferably on EU servers) and who has access to it;
• what data reaches the AI model and what data has no business being there;
• how long conversations are kept and how they are deleted;
• how the data protection impact assessment is carried out, together with the institution's data protection officer.

At Neurotech, agents run on Claude, the model developed by Anthropic, via the API, and data is stored in Supabase on EU servers. You will find details on our data security page.

How it fits with digitalisation projects already under way

Many councils already have, or will soon have, new management systems. In Romania's North-West Region, for example, the Cloud Regional Nord-Vest project, coordinated by the Special Telecommunications Service together with the county councils and ADR Nord-Vest, is building shared software for local administrations in the six counties, prioritising local taxes, planning and the agricultural register, with completion expected in 2029.

In parallel, the North-West Regional Programme has a call open from 1 September to 30 October 2026, "Improving the quality of services provided by local public administrations through innovative digital solutions". Exactly who can apply and which costs are eligible is set out in the call guide, so check the current call guide.

The practical conclusion: an AI assistant should not duplicate the systems on their way, but connect to them. When you choose a solution, ask how it integrates with the electronic registry, the tax platform and whatever arrives through the regional projects.

How to start: a 5-step plan

1. Count. How many calls does the switchboard get a day? What are the top 20 questions? How many requests come in by email? Without numbers, you cannot measure the result either.
2. Choose a single use case. Informing citizens is usually the safest start: low risk, high volume, visible results.
3. Build the knowledge base with the departments. The correct answers come from the people working at the counter, not from the supplier.
4. Test internally, then publicly. A few weeks with staff acting as "test citizens", then a launch with clear notice that it is an AI assistant.
5. Measure and expand. Calls handled, unanswered questions, time saved. Only then move on to triaging requests or drafting replies.

Mistakes to avoid

• An "AI for the whole council" project instead of a concrete use case.
• A knowledge base written once and never updated after fees or opening hours change.
• No route to a person: citizens must always be able to reach an official.
• A system that cannot export the data if the institution changes supplier.

Questions to ask any supplier

• Which sources does the system answer from, and who approves them?
• What happens when it does not know the answer?
• Where are conversations stored, and for how long?
• How are citizens told they are talking to an AI system?
• How is information updated when a fee or opening hours change?
• Can we export all the data if we change supplier?

If you get a vague answer to any of these questions, hold off before signing.

Where we fit in

Neurotech AI builds AI voice agents, chatbots, automation and custom applications. We work mainly with SMEs, but the same components fit the needs of a local council: answering calls, informing citizens, appointments, triaging requests. We do not write funding applications; if the project has EU funding, we work alongside the consultant or team managing it.

If you want to see what could concretely be automated in your council, book a free 30-minute call. We start from the institution's real volumes and questions.