Use cases/Financial services/Complaints handling
AI complaints handling for banks, insurers and wealth
Complaints handling means every complaint reaches the complaints register as one record, whichever channel it came from. The system sorts it by product and issue, shows the response deadline under ASIC Regulatory Guide 271, and drafts the response from your policy and past outcomes. It also flags patterns that may be systemic issues for the risk team. A person approves every response.
Talk to our team →- Supports
- Complaints officer
- Reads
- Phone notes, Emails, Web forms, Branch logs, Policy documents, Past outcomes
- Runs on
- Your servers, on premises
- Connects to
- Salesforce, SharePoint
- Decides
- A person, every time
Architecture
Everything runs on your servers
We build it on an open-weight model such as Llama or Mistral, served with vLLM on NVIDIA GPUs inside your own environment. Complaints are read from email, forms and call notes, policy and past outcomes are indexed in PostgreSQL with pgvector, and drafts are tested with pytest and Ragas style evaluation before they reach the complaints team.
- SourcesPhone notes / Emails / Web forms / Branch logs / Policy documents / Past outcomes
- Your serversRead the complaint → Check the deadline → Draft the response
- ReviewComplaints officer
- OutputOne record from every channel / Response deadline tracking / Draft response from policy and past outcomes / Systemic issue flags
It takes four jobs off the complaints officer
- 01
One record from every channel
Phone notes, emails, web forms and branch logs are read into a single complaint record. The record is sorted by product and issue and routed to the owning team.
- 02
Response deadline tracking
Each complaint shows its response deadline under RG 271, and the list is sorted by the days left. A complaint about a default notice or a hardship notice is marked for its shorter deadline.
- 03
Draft response from policy and past outcomes
The response is drafted from your own policy and from how similar complaints were resolved. Each statement links to the policy clause or past case behind it, and the officer edits it.
- 04
Systemic issue flags
Complaints are grouped by product, issue and cause, and a group that grows is flagged for the risk team with the complaints behind it. The risk team decides whether it is a systemic issue.
It goes live in five steps
- Step 01
Collect closed complaints and responses
Closed complaints with the channel they came in on, the product and issue, the response sent and the outcome. This is the test set, and access to it is restricted to the people who already have it.
- Step 02
Write down how complaints are sorted and answered
We sit with the complaints officers and list how they sort a complaint, which policy they apply and how a response is worded. The rules and the approved wording are written down and encoded.
- Step 03
Test against closed complaints
The system sorts complaints and drafts responses for cases that already have an outcome. Officers compare each draft with the response that went out and mark what was wrong or missing.
- Step 04
Run in shadow
Officers handle complaints as usual while the system sorts and drafts beside them. Nothing reaches the complaints register or a customer until an officer accepts it.
- Step 05
Officer approves and the log records it
A person approves every response before it goes. The risk team reviews each systemic flag, and the log records what the system read, drafted and flagged.
Software or a custom build
Your complaints register or CRM already logs each complaint and keeps its status. A custom build does not replace it, and it reads from the register and writes drafts back. Use the case management in your CRM first if complaints arrive through one channel and the volume is low. A custom build earns its cost when complaints arrive through several channels, when responses are written from scratch against policy documents, or when complaint files cannot leave your environment.
Salesforce / SharePoint
In practice
We build this on servers you own. The model, the policy index and the logs sit inside your environment, and complaint files are not sent to a public AI service. The retrieval pattern is covered in RAG on private infrastructure, and the build falls under AI agents and workflow automation.
The deadlines come from ASIC Regulatory Guide 271. A standard complaint gets a response no later than 30 calendar days after the firm receives it. A credit related complaint involving a default notice gets 21 calendar days. So does one involving a hardship notice or a request to postpone enforcement. Limited exceptions apply when the lender needs more information or the parties reach an agreement. A superannuation trustee complaint, other than one about a death benefit distribution, gets 45 calendar days. The guide also requires firms to analyse complaint data to identify systemic issues and to escalate possible systemic issues promptly. The system shows each date and a person sends the response.
The same layer answers a complaints officer from policy documents. How the output is checked is covered in LLM evaluation and observability. The wider picture for banking, insurance and wealth management has its own pages.
Questions
What is AI complaints handling?
It is the sorting, tracking and drafting that happens before a person answers a complaint. The system reads each complaint, sets the product and issue, shows the response deadline and drafts the response from your policy and past outcomes. The complaints officer approves it.
Does the AI decide the outcome of a complaint?
No. The complaints officer decides the outcome and approves every response. The system drafts from your policy and from how similar complaints were resolved, and each statement links to its source.
How does it handle the RG 271 timeframes?
It shows the response deadline beside each complaint, set from the RG 271 timeframes for the complaint type. A person sends the response. Your compliance team confirms how the timeframes apply to each complaint.
How does it flag systemic issues?
It groups complaints by product, issue and cause and flags a group that grows, with the complaints behind the flag. The risk team investigates and decides whether the group is a systemic issue.
Industries where this comes up
Banking / Insurance / Asset and wealth management
More financial services use cases
Document search / Credit file assembly / AML alert triage / Claims triage / Month-end close / Audit evidence matching / Advice file drafting / Lease abstraction / KYC and onboarding / Regulatory reporting / Underwriting submissions / Board and management reporting
Before you call
Tell us where your hours go.
We map the work, size the hardware and price what is worth automating.
Talk to our team →