Where AI fits in local services

Why local government is different

Local government is where the state meets people every day. A resident does not experience policy; they experience whether the water comes, whether the waste is collected, whether the street light is fixed and whether their application moves. That makes local offices a tempting place for AI: the volume of routine paperwork is high, staff are stretched and residents want faster answers.

It also makes the risks sharper. A mistake in a ministry brief is usually caught by someone further up the file. A mistake in a reply to a resident, a notice about a water shutdown or a decision on a complaint lands directly on a family. This module sets out where AI can help, where it should not be used and the questions to ask before you start.

Where AI can help

Most useful applications are about handling text and information, not making decisions. A municipal office or tehsil administration might use an AI assistant to:

• draft routine correspondence, such as acknowledgements, reminders and replies to common queries, which an officer then checks and signs
• summarise long complaint registers, inspection notes or meeting minutes so that patterns are easier to see
• sort incoming complaints by service and area so they reach the right section sooner (module 2)
• prepare notices in plain Urdu and regional languages for checking by fluent speakers (module 4)
• tidy and describe data you already hold, such as counting complaints by ward or month (module 3)
• answer frequently asked questions about how to apply for a service, provided the answers come from your own approved information

Each of these saves time on the first draft or the first sort. None of them removes the need for an officer who knows the area and takes responsibility for the result.

Where AI should not decide

Some local decisions affect people's rights, money or property. Building approvals, encroachment action, licence decisions, penalties, connection or disconnection of services and the allocation of scarce resources between neighbourhoods are examples. AI can help you read the file or draft the reasoning, but the decision must be made by a person with the authority to make it, on the facts, under your department's rules.

Be especially careful with tools that score or rank people or places. A system that predicts which areas will "cause problems" or which residents are likely to default can repeat old patterns of neglect. This is bias and it is hard to see from inside the office. If anyone proposes predictive analytics for local services, ask what data it learned from and who checked it for fairness.

Risks close to citizens

• Invented answers. An AI tool can give a confident but wrong answer about a fee, a deadline or a procedure. This is a hallucination. If a resident acts on it, they pay the price.
• Personal data. Complaints and applications contain names, phone numbers, addresses and sometimes details of family disputes. Pasting these into a public AI tool can expose them. Read about data protection and use the remove personal data prompt on practice text only after checking the result yourself.
• Unequal access. Residents who cannot read, who have no smartphone or who speak a language the tool handles poorly can be left behind if AI becomes the main way to reach your office. Keep a human route open.
• Lost accountability. When something goes wrong, the resident needs to know who decided and how to challenge it. "The computer did it" is never an acceptable answer. Accountability stays with your office.

Questions to ask before you start

1. What problem are we solving and would a simpler fix (a clearer form, a better register) solve it?
2. What information will go into the tool and is any of it personal or confidential?
3. Who checks the output before it reaches a resident?
4. How will a resident know AI was involved and how can they seek redress if it goes wrong?
5. Does our department have instructions on AI use? Read the National AI Policy 2025 explainer and follow your department's own rules.

If you plan to try a tool with a small team first, the guide on piloting an AI tool sets out the steps.

Try it

List five recurring tasks in your office. For each, write one line on whether AI could help with the first draft or first sort, one line on what could go wrong for a resident and one line on who would check the result. Mark any task where AI should not be used at all.

Check yourself

1. Give two examples of local decisions that AI should not make.
2. Why are prediction or ranking tools a particular risk in local services?
3. What should a resident always be able to find out when AI has been involved?

Answers: (1) Any two of: building approvals, encroachment action, licence decisions, penalties, service disconnection, allocating resources between areas. (2) They can repeat old patterns of neglect or unfairness hidden in the data they learned from. (3) Who made the decision and how to challenge it.