How it works
Where the data comes from
Every number on this site comes from the CFPB's public Consumer Complaint Database, pulled from its keyless, public search API. No scraping, no API key, nothing proprietary. This site is not affiliated with or endorsed by the CFPB.
What "spike ratio" means
Raw complaint counts mostly just measure company size — Bank of America has more complaints than a small credit union because it has vastly more customers, not because it's worse. Instead, every company here is ranked by this week's complaint volume divided by its own trailing 8-week baseline average. A ratio of 3.0x means this week ran 3x that company's own recent normal — the same number regardless of whether the company is huge or small.
| Status | Ratio | Meaning |
|---|---|---|
| ● Near baseline | < 1.2x | Roughly normal week |
| △ Elevated | 1.2x – 2x | Noticeably above normal |
| ▲ Spiking | 2x – 4x | Sustained, unusual jump |
| ▲▲ Major spike | ≥ 4x | Large, sudden jump |
Why "this week" isn't the last 14 calendar days
The CFPB's index is still being filled in for the most recent complaints — a complaint received yesterday often hasn't been indexed yet. We measured this directly: complaint counts 6–8 days after being received were still running 9–14% below the same weekday's normal volume from prior weeks, which is a large enough gap to look like a real decline when it isn't. By 13–14 days out, that gap narrows to about 4% — close to the day-to-day variation clean, fully-indexed days already show against each other. So "this week" here always means the most recent complete week, ending 14 days before the data was pulled, not the literal last 7 or 14 calendar days.
Narrative text (the free-text complaint description) lags much further still — CFPB has to review and redact it before publishing, and it isn't stable until roughly 9–10 weeks after the complaint was filed. Word clouds and complaint-flow numbers use a 70-day buffer for this reason; a shorter buffer would make narrative volume look like it's declining when it's really just not published yet.
Personal information
Complaint narratives are published by the CFPB with obvious personal information (names,
dates, account numbers, etc.) already replaced with XXXX placeholders by the agency
before release — this site doesn't do its own redaction and doesn't collect or store anything
from site visitors. Word clouds strip those placeholder tokens out before counting words.
Why the homepage leaderboard doesn't include everyone
The three nationwide credit bureaus (TransUnion, Equifax, Experian) touch nearly every consumer's credit file and outrank every actual bank by roughly 25x on raw complaint volume — blending them into one ranking would mean the same three names occupy the entire leaderboard, every week, forever. They're tracked, just ranked separately. The same logic applies to debt collectors, loan servicers, and consumer-report resellers — very different businesses with very different normal volumes, so ranking them against banks wouldn't mean much. Every one of those categories has its own ranking on the Categories page; the homepage leaderboard stays scoped to banks, credit unions, and fintechs specifically.
Company coverage
This is a fixed, categorized list of real companies, not every company in the CFPB database — see Categories for the full roster by type. Word clouds and complaint-flow breakdowns are currently built for a smaller subset of the highest-volume banks — those two features need enough narrative volume in a single month to be meaningful, which doesn't hold for every company yet. Extending either list is mostly a matter of API budget and reviewing new companies' data quality, not new code.
Update cadence
The leaderboard, category pages, company pages, word clouds, and complaint-flow tables are rebuilt from live CFPB data on a schedule (see this project's deploy notes). The narrative-clustering page ("Narrative insights") updates less often and on a separate, manual cadence — it needs a heavier local-model step this site's automated rebuild deliberately doesn't run on every update.