Why complaint counts need denominators

A company with sixty million customers will receive more complaints than a company with sixty thousand, even if it treats every customer better. A list ordered by raw count is therefore mostly a list ordered by size. The CFPB makes the point itself and asks that complaint volume be read in the context of company size or market share.

A worked example from the current data

CAPITAL ONE FINANCIAL CORPORATION received 33,328 complaints in the last 12 months and its insured banks hold $662 billion in assets: 50.3 complaints per $1 billion. SIMMONS FIRST NATIONAL CORPORATION received 67 complaints against $25 billion: 2.71 per $1 billion. The raw counts differ by a factor of 497; the per-size figures tell a different story. Neither number is a verdict. They are simply on the same scale.

The ideal denominator, and the ones that exist

The best denominator would be customers or accounts for the product in question. No public file reports that for every company. Three public files come close enough to be useful:

• FDIC total assets for insured banks and their holding companies (91 companies matched).
• NCUA total assets for credit unions (19 matched).
• HMDA filings, the count of mortgage applications and loans a lender reported, for non-bank mortgage companies (40 matched).

Limits worth knowing

• Assets are not customers. A card-focused bank has many customers per dollar of assets; a commercial bank has few. Compare like with like.
• Holding companies do more than banking. Where a group's main business sits outside its insured bank, bank assets understate its size. Each company page names the institutions counted.
• HMDA counts new lending. It does not measure a servicer's book, so servicers get no HMDA-based rate on this site.
• Most companies have no public denominator. Credit bureaus, debt collectors, payment apps and most non-bank lenders could not be matched by name to a single institution in the FDIC, NCUA or HMDA files. Their pages say “no public size denominator” and tables mark them “no public measure”. That label describes the data, not the company.

Full matching rules are in the methodology.