Handling service complaints with AI

Why complaints matter

Complaints are how residents tell you a service has failed: no water for three days, waste not collected, a broken manhole cover, a road dug up and left open. Each one is small to the office and large to the person who made it. When complaints pile up, the urgent ones get lost among the routine ones and residents stop believing that complaining achieves anything.

AI can help an office keep up, mainly by sorting complaints and drafting first replies. It cannot inspect a pipe, judge whether a contractor did the work or decide who is at fault. This module shows how to use it for the first and keep people firmly in charge of the second.

Sorting complaints

Complaints arrive in many forms: letters, register entries, phone notes, messages, Urdu, English, Roman Urdu and regional languages. An AI assistant can help turn a batch of complaint texts into a consistent table with columns such as service (water, sanitation, waste, roads, street lights), area or ward, the problem in one line and any sign of urgency.

Practical rules for sorting:

• Agree the categories first. Give the tool your office's own list of services and areas. Do not let it invent new ones.
• Flag, do not filter. Ask the tool to mark complaints that mention health or safety risks (sewage in drinking water, an open manhole, a fallen electricity wire, a collapsed wall) as urgent for a person to review. Never let it close or discard a complaint.
• Check a sample every time. Read a handful of complaints against the table to see whether the categories and urgency flags are right. Errors in sorting are quiet: a misfiled complaint simply waits.
• Watch for language gaps. Tools often understand standard Urdu and English better than Roman Urdu or regional languages. Complaints in those languages may be sorted badly. Check them by hand.

The how-to guide on pulling data out of documents covers building tables from text.

Drafting replies

Many replies are routine: acknowledging receipt, giving a reference number, explaining which section will deal with it and when the resident can expect an update. AI can draft these quickly. The reply to a citizen complaint prompt is a starting point.

Before any reply goes out:

• Check every fact. Dates, timelines, fees, office addresses and phone numbers must come from your own records, not from the AI. Remove anything you cannot confirm.
• Do not promise what you cannot deliver. AI drafts tend to be reassuring. "The issue will be resolved within 24 hours" is a commitment your office may not be able to keep.
• Keep the tone respectful. Read the reply as the resident would. A polite, specific reply builds trust; a generic one reads as dismissal.
• Sign it as a person. The reply should come from a named officer or section, so the resident knows whom to follow up with.

Keeping decisions human

The decisions in complaint handling are: whether the complaint is valid, what action to take, who is responsible, whether it is resolved and whether anyone faces a penalty. These stay with officers. AI can summarise the file or draft the reasoning, but a person who knows the facts decides and records the decision. This is human oversight in practice.

Two habits help:

1. Keep an audit trail. Note in the register when AI was used to sort or draft and who checked it. If a resident later disputes how their complaint was handled, you can show what happened.
2. Make challenge easy. Every final reply should tell the resident how to escalate if they are not satisfied. Redress is part of the service.

Protecting residents' data

Complaints often name neighbours, landlords or officials and sometimes describe disputes or health problems. Do not paste raw complaints into a public AI tool. Use only tools your department has approved for this kind of data. For practice, remove names, phone numbers, CNIC numbers and exact addresses first, using the remove personal data prompt as a guide and checking the result yourself. See personal data in the glossary.

Try it

Write ten short, made-up complaints of the kind your office receives, in a mix of English and Urdu, with no real names or addresses. Ask an AI assistant to sort them into a table with service, area, a one-line summary and an urgency flag. Check every row. Note which complaints were sorted wrongly and whether any urgent case was missed.

Check yourself

1. Why should AI flag urgent complaints rather than filter out routine ones?
2. Name three things you must check before an AI-drafted reply goes to a resident.
3. Which decisions in complaint handling must stay with an officer?

Answers: (1) A filtered or closed complaint can disappear without anyone noticing; flagging keeps a person in control. (2) Any three of: facts such as dates and fees, promises about timing, tone, that it is signed by a named officer. (3) Validity, action to take, responsibility, whether it is resolved and any penalty.