AlphaBiz/cfpb-complaints-normalised

CFPB Consumer Complaints, normalised by company size

Every complaint filed with the US Consumer Financial Protection Bureau since December 2011 — 17,520,935 of them against 8,081 companies — aggregated by company, product, issue and state, with complaint rates normalised against company size and peer group.

Data as of 2026-09-15. Rebuilt from the CFPB archive nightly; this mirror is refreshed periodically, and the live figures are always at complaintdigest.com.

Read this first

1. A complaint is an allegation, not a finding. The CFPB does not verify the substance of a complaint before publishing it. A high complaint count is not evidence that a company did anything wrong.

2. Complaint volume mostly tracks customer count. The largest firms lead almost every raw table by being the largest firms. Use the size-adjusted columns (cri, peer_complaint_ratio) for comparison, and peer_benchmarks.csv to see what they were normalised against.

3. The scores are ComplaintDigest's, not a regulator's. Counts, dates and response categories are the CFPB's records. cri, rqs, risk_score, risk_band and the trend columns are computed by this project and are analysis. Attribute them to ComplaintDigest, never to the CFPB.

These three are not boilerplate. They are the difference between a dataset that can be published on and one that produces a correction.

What is in here

File	Rows
companies.csv	8,081
company_quarterly.csv	107,741
company_product.csv	57,779
company_state.csv	82,910
issues.csv	121
issue_company.csv	26,285
products.csv	21
states.csv	59
peer_benchmarks.csv	41
score_weights.csv	1
enforcement_actions.csv	386
enforcement_links.csv	133

283,558 rows in total. See DATA_DICTIONARY.md for every column.

What is deliberately not in here

• The complaints themselves. 17.5M rows of the CFPB's own public-domain data, which they publish better than we could: consumerfinance.gov/data-research/consumer-complaints and a bulk CSV. What is here is the aggregate layer computed on top of it, which is the part that does not already exist.
• Consumer narratives. People's own accounts of their financial trouble. The CFPB publishes them; re-hosting them is their call, not ours.
• Machine-generated theme summaries. They exist on the site, labelled as model output. As unlabelled rows in a research dataset they would invite being analysed as measurements, which they are not.

Live access

The same data is available without downloading anything, and always current:

• JSON API, no key required — https://api.complaintdigest.com/api/v1
• MCP server for AI assistants — https://mcp.complaintdigest.com/mcp
• Methodology, including which inputs are estimates — https://complaintdigest.com/methodology
• For journalists and researchers — https://complaintdigest.com/press

Licence and attribution

cc-by-4.0 for the bundle, which is a mixture and is labelled conservatively.

The CFPB's columns — counts, dates, response categories — are a US Government work and in the public domain. The computed columns (cri, rqs, risk_score, risk_band, and the trend columns) are ComplaintDigest's analysis. Use them freely, including commercially, on one condition: attribute them to ComplaintDigest and describe them as our analysis rather than as CFPB findings. A link back is appreciated and not required.

Suggested citation:

ComplaintDigest, "CFPB Consumer Complaints, normalised by company size", https://complaintdigest.com. Data as of 2026-09-15. Complaint counts from the CFPB Consumer Complaint Database; scores computed by ComplaintDigest.

Known quirks

• There is no consumer-dispute column anywhere in this bundle, and that is deliberate. The CFPB stopped collecting the "Consumer disputed?" field in April 2017, and the field is empty for every row of every year in the source we build from — not merely for the years after it was retired. Three columns that would otherwise exist here (dispute_rate_pct, disputed_count, avg_dispute_rate_pct) were dropped rather than shipped full of zeros, because a zero reads as a measurement and an absent column does not. The flag nominally carries weight in the response-quality score; score_weights.csv shows dispute_data_available = 0 and rqs_w_non_dispute = 0, with that weight redistributed across the remaining components rather than awarded to everyone for free.
• estimated_customers is an estimate, and for most companies it is not a measurement at all. The CFPB does not publish customer counts, so there is no authoritative denominator for a complaint rate. Where a company can be matched exactly and unambiguously to an FDIC or NCUA registry entry, that figure is used; where it cannot — which is most of them — the value is a midpoint inferred from its size tier, and a size tier is itself derived from complaint volume. cri inherits that, so treat it as an ordering within a peer group rather than as a rate with units. This is the softest number in the bundle and the one most worth reading /methodology about.
• Product and issue categories were renamed in 2017 and 2023. Long series crossing those boundaries compare differently-named things.
• Companies below 10 complaints in 12 months are included but not scored. Too little data to support a rate or a trend. The website keeps them out of its index for the same reason.