Complaint management system: logging, root cause and closing the loop with AI help

A complaint management system records every complaint as a case with an owner, a deadline, a category and a root cause. It tracks the case until the customer has an answer and the underlying problem is fixed. AI can classify incoming complaints, draft acknowledgements and summarize history. Regulated, high-value and sensitive complaints always go to a named person.

#Treat a complaint as a case

A support ticket asks for help. A complaint says something went wrong and expects an answer. They often arrive in the same inbox, which is how complaints get buried under routine questions.

If your system treats them the same, you will miss the deadline on the one that matters. Mark a message as a complaint when it expresses dissatisfaction and asks for something to be put right, even if the customer never says "complaint."

Regulated sectors set formal rules. For example, the UK Financial Conduct Authority's handbook (DISP 1.6 (opens in a new tab), as read in October 2026) requires a firm to send a prompt written acknowledgement and then a final response, or an explanation of why one is not yet possible, by the end of eight weeks. Complaints about payment services have a shorter window, 15 business days in the standard case. Rules vary by industry and country, and your compliance team owns the definition that applies to you.

#The case record

Give every case the same fields, so you can sort, report and audit.

• Received date and time, and the channel (email, phone, chat, social, letter, review site).
• Customer, account and the order, product or service involved.
• Category and severity, using a fixed list.
• Owner and a due date for acknowledgement and for a final response.
• Full history: messages, calls, internal notes and attachments.
• Remedy offered, with approver and amount where money is involved.
• Root-cause tag, filled in before closure.
• Customer outcome: accepted, rejected, escalated.
• Closure date and who closed it.

Keep the category list short. Eight to twelve beats forty, because staff pick the first plausible option on a long list.

#The complaint management process

1. Log. Capture the complaint from any channel into one case, with the original message attached.
2. Acknowledge. Send a prompt reply that confirms receipt, names the owner and states the next date.
3. Classify. Set category and severity. Flag regulated, legal-threat, safety, vulnerable-customer and high-value cases.
4. Investigate. The owner collects facts: the order, the delivery record, the call recording, the policy that applies.
5. Decide and respond. Offer a remedy where due, or explain a rejection with reasons.
6. Confirm. Check that the customer received the answer, and note whether they accept it or escalate.
7. Find the root cause and close the loop. Tag the cause, pass it to the team that can fix it, and close the case only after that handoff.

Most teams skip step 7. A complaint that was answered but never analyzed comes back next month under a different customer's name.

#Root cause tagging

A root-cause tag records why it happened. The customer's words describe the symptom, and "late delivery" is a symptom. "Carrier missed a pickup because the dispatch notice went to an old email address" is a cause.

Use a short, stable list: product defect, delivery or logistics, billing error, staff conduct, unclear policy, misleading information, system fault, third-party failure. Each case gets one primary tag plus free text.

Then report monthly by tag and volume. A team might find, for example, that 30 percent of last quarter's complaints (an illustration) trace to billing errors, and most of those to a single invoicing rule. Fixing that rule removes more complaints than any rewrite of the reply template.

#Where AI helps

An agent can take over the slow parts of a case while a person owns the decisions.

• Detect complaints inside a general queue, and set category and severity.
• Draft the acknowledgement with the owner's name and the right date.
• Summarize a long email thread or call history into five lines for the owner.
• Pull the facts: order status, delivery scans, previous contacts, warranty terms.
• Suggest a root-cause tag and a likely remedy from policy.
• Draft a response from your approved wording and policy documents, with the sources shown.
• Spot repeat contacts and signs of churn. See our use case on flagging escalation and churn risk in tickets.

Our support operations blueprint shows where this sits in a full support flow. The use case on triaging and routing support tickets covers the sorting step at the front.

Key point: Let the agent classify, gather facts and draft. A named person decides the remedy and sends anything about a regulated, high-value or sensitive complaint.

#Where complaints must go to a person

Some complaints should never be handled end to end by software.

• Regulated complaints with a statutory clock or an external ombudsman route.
• Anything mentioning legal action, a regulator, the media or a data protection request.
• Safety or health issues.
• Customers who may be vulnerable or who say they are in difficulty.
• High-value accounts and large refunds, based on a threshold you set.
• Allegations about named staff.
• Cases where the agent's confidence is low or the facts conflict.

For these, the agent can prepare the file. The owner reviews and replies. Log every AI draft and who edited it, so you can show how a response was produced.

#Choosing software

The choice turns on whether you face external deadlines. If you do, check that the system shows every open case against its clock, escalates before a breach and exports the audit trail. Make vendors prove it on your sample cases instead of a demo script.

#Measuring the system

• Time to acknowledge and time to final response, against target.
• Share of complaints answered within deadline.
• Reopen and escalation rate. A rising reopen rate usually means the first answer missed the point.
• Complaints per 1,000 orders or per 100 customers, by root cause.
• Customer satisfaction after closure, measured as CSAT on a short survey.
• Number of complaint causes fixed upstream each quarter.

The last measure shows the loop closing. Teams that fix causes see complaint volume fall over time.

#How to start

1. Define what counts as a complaint, with your compliance or legal lead.
2. Fix the case fields and the category and root-cause lists.
3. Set deadlines for acknowledgement and response by type.
4. Run a month of cases through the process by hand, to find gaps in the fields.
5. Add the agent for classification and drafts, with review on every output.
6. Report root causes monthly and give each cause an owner.

Our customer support page covers the wider picture. If complaints are scattered across inboxes and nobody can say how many are open, book a strategy call. Bring an export of last month's complaints and we'll show you what a case-based process would change.

#Frequently asked questions

#What is a complaint management system?

It is a system that logs each complaint as a case, assigns an owner and a deadline, tracks the investigation and response, and records a root cause. It also reports patterns so teams can fix the underlying issue. It can be a module in a help desk or CRM, or dedicated software.

#What is the difference between a complaint and a support ticket?

A ticket asks for help. A complaint expresses dissatisfaction and expects a remedy or an explanation. Complaints often carry deadlines, need a named owner and require a root-cause finding, so they should be flagged and tracked separately from routine tickets.

#What should be in a complaint management process?

At minimum: logging from every channel, prompt acknowledgement, classification, investigation, a decision and response, confirmation with the customer and a root-cause review before closure. Add deadlines and escalation rules for the complaint types you handle.

#Can AI handle customer complaints?

It can classify them, draft acknowledgements and replies from approved wording, summarize history and suggest root causes. It should not close regulated, high-value or sensitive complaints alone. A named person approves the remedy and sends those responses.

#How do you measure complaint resolution?

Track time to acknowledge, time to final response, share answered within deadline, reopen rate and complaints per 1,000 orders by root cause. Add post-closure satisfaction. Also count how many root causes were fixed, since that shows whether complaints are really falling.