AI in banking

Complaint management in banking: how AI agents catch, route, and resolve complaints

11 November 2026
6
mins read

Complaints hide inside everyday service conversations, and AI agents can find them, start the right clock, and hand each case to the person who owns it.

TL;DR

  • Complaint management in banking is how a bank captures, logs, investigates, resolves, and learns from every complaint, and AI agents can now help with each stage except the final decision.
  • In our experience, most complaints never arrive labeled as complaints: they surface halfway through a chat about a fee or a call about a late transfer, which is where manual logging misses them.
  • The clocks differ by regime: companies generally answer complaints the CFPB forwards within 15 days, UK firms get up to 8 weeks for most complaints, and a Regulation E error notice starts its own 10-business-day investigation.
  • AI agents are strongest at detection, classification, evidence gathering, and routing, while people keep redress, final responses, and exceptions.
  • Start by agreeing with compliance on what counts as a complaint, then tag every conversation against that rule before you automate anything downstream.

A customer who types "this is the third time I've asked about this fee" into a chat about something else has made a complaint, whether or not anyone logs it. In our conversations with leading banks across the globe, the complaints that cause trouble with regulators are rarely the ones that came in through the complaints form. They're the ones that sat inside a service conversation and missed their clock.

Below is what the process involves, the deadlines it has to meet, where AI agents help, and how to set them up so compliance can sign off.

What is complaint management in banking?

Complaint management in banking is how a bank receives, records, investigates, resolves, and learns from customer complaints across every channel. A complaint program runs from intake and logging through triage, investigation, and a response, and it ends with root-cause analysis that feeds back into products, fees, and processes.

The regulator's view reaches past the bank's inbox. US customers can take a complaint to the Consumer Financial Protection Bureau, which routes it to the company and asks for a response, and national bank customers can turn to the OCC's Customer Assistance Group. Every complaint the bank's own process misses can end up in front of someone else.

Why do banks miss complaints?

Banks miss complaints because the complaint process and the service conversation usually run on different systems. Three gaps come up again and again.

  • Complaints arrive unlabeled. Customers rarely say "I'd like to file a complaint." They show frustration in a chat, on a call, or in a branch, and a busy employee answers the question without logging it.
  • Channels don't share a case. The chatbot, the contact center, and the complaints team often keep their own records, so a customer who complains in two channels becomes two unrelated tickets.
  • Containment counts the wrong thing. A bot can mark a conversation as contained while the customer leaves angrier, which is how high containment rates hide frustrated customers.

How long does a bank have to respond to a complaint?

The deadline depends on where the complaint lands and what it's about.

RegimeWhat it applies toClock
CFPB complaint process (US)Complaints the CFPB sends to a companyCompanies generally respond in 15 days, with a final response in 60 days in some cases
Financial Ombudsman Service (UK)Complaints to UK financial businessesUp to 8 weeks for most complaints, and 15 days for fraud and scams, payment services, and e-money
Regulation E error resolution (US)A consumer's notice of an electronic fund transfer error10 business days to investigate, or up to 45 days with provisional credit

The UK clocks come from the Financial Ombudsman Service, and the Regulation E clock can stretch to 90 days for point-of-sale debit card transactions and some other transfers. That last row is why classification matters: a customer who complains about a card payment taken twice is also raising an error notice, so one message starts two clocks with two owners. The bar is high, too, because 99% of complaints the CFPB sends to companies get timely responses.

Where do AI agents help in complaint handling?

AI agents help most at the start of the process, where volume is high and the work is reading, sorting, and gathering. In our experience, that's also where manual processes lose the most complaints.

Detection in every conversation

An agent working across chat, voice, and messaging can flag expressions of dissatisfaction against rules that compliance has written, and log a complaint even when the original request was resolved. That turns complaint capture from a judgment call into a consistent check on every conversation.

Classification and the right clock

The agent tags each complaint by product, issue, and regime, separating regulatory complaints from service requests and error notices from fee complaints, so the right clock starts with the first message.

Evidence before the case reaches a person

Agents pull the transaction history, the transcript, prior contacts, and the policy that applied, so the investigator starts with a complete file. It's the same pattern banks use for dispute resolution automation, where the agent assembles the case and a person decides.

Routing, tracking, and the draft response

The agent routes the case to the team that owns it, watches the clock, escalates before a deadline slips, and drafts a response for a person to review and send.

Root causes across all complaints

With every complaint tagged the same way, the bank can see which fee, product, or process step drives the most of them and fix it at the source.

What should stay with people?

Redress, final responses, and anything outside policy should stay with people. An agent can recommend a refund and show its reasoning, and a person approves it, because that decision carries the bank's accountability. Keep a human in the loop for vulnerable customers, complaints that allege discrimination or fraud, and any case where the evidence conflicts.

That's the split we build into Backbase Customer Operations: agents gather evidence, apply policy, and advance cases across systems, while people keep the decisions, approvals, and exceptions. Classic and agentic workflows run under one deterministic orchestration engine with full auditability and role-based permissions, so every step of a complaint has an owner and a record. In the conversation itself, Backbase Conversational Banking runs dedicated security, risk, fraud, and compliance agents that check every output before it reaches a customer, and it keeps a full audit trail of each exchange.

How to set up AI complaint handling in your bank

Most banks start with one product line, such as cards or deposits, and extend once the definitions hold up.

  1. Agree on the definition with compliance. Write down what counts as a complaint in each market you serve, including complaints made in passing, and turn it into detection rules.
  2. Tag every conversation. Run detection across chat, voice, messaging, email, and branch notes, including the conversations your bot contained.
  3. Start the clock at the first expression. The deadline runs from the moment the customer complains, so timestamp the detection.
  4. Give agents evidence authority and keep redress with people. Let agents read systems and assemble files, and send every refund, fee reversal, and final response to a named approver.
  5. Keep one case across channels. Link every contact from the same customer about the same issue to one case, so the second complaint updates the first.
  6. Review root causes every month. Take the top complaint drivers to the product and operations owners, and track whether each fix reduced volume.

Frequently asked questions

What is complaint management in banking?

Complaint management in banking is how a bank captures, logs, investigates, resolves, and learns from customer complaints in every channel. AI agents now help with detection, classification, evidence, and routing, while people keep the decisions.

How can banks use AI agents to investigate and route customer complaints?

Agents detect complaints in conversations, tag them by product and regime, collect the evidence, and route each case to its owner with the deadline attached. A person reviews the evidence, decides on redress, and approves the response.

How do banks tell a regulatory complaint from a service issue?

They write the distinction into rules that compliance owns and apply those rules to every conversation. In our experience, the safest setup logs any expression of dissatisfaction as a complaint and lets a person downgrade it, because a missed complaint costs far more than a reviewed one.

How long does a bank have to respond to a complaint?

It depends on the regime. Companies generally respond to CFPB complaints within 15 days, UK firms have up to 8 weeks for most complaints, and Regulation E allows 10 business days to investigate an error.

Can a banking chatbot escalate complaints to staff with a full audit trail?

Yes, when the chatbot runs on the same case and workflow layer as the complaints team. The conversation, tags, evidence, and every handoff then travel with the case, so the person who picks it up sees what the agent saw.

About the author
Table of contents