Part of Kibbo Research →

A wall of identical consumer complaint narratives, representing the CFPB Complaint Vault research project
Kibbo Research · Financial & Banking

The CFPB Complaint Vault

An open research project on the US Consumer Financial Protection Bureau's complaint database: the data, the method, and everything we have published from it.

Kibbo Research · October 2026

What This Is

The Consumer Financial Protection Bureau (CFPB) has collected consumer complaints about banks, lenders, credit bureaus and debt collectors since December 2011. The public database holds more than 18 million of them. In 2026 we downloaded it, together with the archive of complaint narratives published through the CFPB's FOIA Electronic Reading Room (complaints received up to 14 August 2026), and measured what is in it.

The Vault is the home of that work. Everything is open: the dataset has a DOI, every analytical decision is logged with its date and reasoning, and every figure on the site traces back to a table you can download.

A complaint in this database is a consumer's allegation. It is not a finding that a company broke the law, and company responses are self-reported and not verified by the CFPB.


Start Here

Main investigation

The Copy-Paste Complaint: How Templates Took Over America's Financial Complaint Database

The main investigation. How complaints doubled in 2025, how credit reporting went from 1 in 5 complaints to 9 in 10, how many narratives are shared templates, and what happens to your chances when you use one.

Read the investigation →

The Numbers in Brief


The Data

Kibbo original dataset

The CFPB Complaint Vault Dataset (Version 1.0)

Complaint-level file, 12 summary tables, data dictionary, README, scripts and checksums. Narrative text and filer keys are not included. License: CC BY 4.0.

Zenodo record (DOI) →

Kibbo's extracts, data dictionary and README →


How We Did It

  1. Two sources, joined by complaint ID (99.9% match on a 5,000-ID random sample).
  2. Products harmonised into 11 families; credit reporting treated separately because it is 88.4% of 2025 complaints.
  3. Template detection at three levels: exact duplicates, normalised duplicates, and fuzzy matches (MinHash with word 5-shingles, 128 permutations, Jaccard similarity of 0.8 or higher). Fuzzy groups are built within each calendar year.
  4. Shared template versus same person: a text group counts as a shared template only if it spans at least two distinct locations. Otherwise it is classified as the same filer repeating.
  5. Company comparisons use only complaints that are not known templates, within the same product family. Bank comparisons are stated per 100,000 deposit accounts of $250,000 or less, because three reasonable denominators gave three different rankings.

Read the full research log, with every decision and what we rejected →
Methodology section of the main investigation →


What We Publish From This Data

Investigation

Real cases

The stories behind the numbers — six individual complaints, and three about shared templates.

Individual complaints

Shared templates

Teaching materials

Tools to use in your own complaint


Code and Reproducibility

The analysis scripts (Python and DuckDB) are distributed inside the Zenodo archive (scripts.zip), with a requirements file and SHA-256 checksums. To reproduce a result, download the archive from the DOI and follow the README; the scripts read the CFPB files from a local folder.


How to Cite

Dataset: Kibbo Research (2026). The CFPB Complaint Vault Dataset (Version 1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23188842

Investigation: Kibbo Research (2026). The Copy-Paste Complaint: How Templates Took Over America's Financial Complaint Database. getkibbo.com. https://www.getkibbo.com/investigations/cfpb-copy-paste-complaints


Limits and Corrections

Corrections log: No corrections to date.