Anyone in Customer Support knows that complaints rarely give you the full story of a customer's journey on their own.
I built Signal to turn a large export of customer complaints into a brief that helps Product & Support decide what to look at next. It groups complaints into themes, proposes what could have gone wrong and suggests a team that might own the issue category.
I used publicly available TransUnion complaints from the CFPB Consumer Complaint Database.
The saved demonstration analysed 2,000 complaint records from December 2025 through March 2026. Five displayed themes covered 1,531 records. One category, “Investigation Did Not Fix Error,” contained 556 records.
Give Signal a support problem to look for, such as customers struggling to dispute incorrect information on a credit report. It selects relevant complaints, groups them into themes and writes a brief containing:
-
What customers report experiencing.
-
Complaint counts and references to the source records.
-
Possible explanations for each theme.
-
What to check next—for example, the customer's support case, documents received, replies sent or records of how the request was processed.
-
A suggested team to begin that review.
The brief is meant to give the team a specific investigation to start, with the complaints available to check its reasoning.
In the latest targeted check, two customers described submitting documents and following up, but said their credit-report errors remained unresolved.
Signal kept several explanations open: a problem with the submitted documents, a processing delay, or a gap in communication. It suggested checking the dispute records, document receipts and replies sent to the customers. It did not assume that an investigation had finished just because the complaint category implied one had.
This check covered signal synthesis and brief writing. The category and classification were supplied; it was not a new run of the full 2,000-record pipeline or a measurement of reliability across all complaints.
Signal proposes investigations. Complaint text alone cannot prove the cause, a regulatory breach or which team owns a fault. The volume-based score helps sort themes; it does not measure customer harm.
This demonstration did not include the company's account records, support histories or product logs. A person should compare the brief with its source complaints and the company's records before making a decision.