About
How Credence Compass works
Credence Compass gives a plain-English read on a financial company's complaint history. It is free, needs no login, and every figure comes from one public source so you can check it yourself.
Where the data comes from
All complaint data comes from the Consumer Financial Protection Bureau's (CFPB) Consumer Complaint Database, which is public. Nothing else is used: no reviews, no ratings from other sites, no user submissions.
- The CFPB publishes a complaint after the company responds or after 15 days, whichever comes first. Very recent complaints may not appear yet. (Source: the CFPB's own description.)
- Every complaint in the list on each company page links to its original CFPB record.
- Figures are fetched from the CFPB when you open a page, so they reflect the database at that moment.
How Compass Check decides a pattern
Compass Check is plain arithmetic, not AI. It compares the company's most recent six months of complaints with the six months before that, rates three measures, and combines the ratings into one of eight patterns. The same data always gives the same pattern.
The three measures
| Measure | What it is | Green | Yellow | Red |
|---|
The eight patterns
| Pattern | When it applies |
|---|
If more than one measure is red, the pattern names the first of these that is red: timely response, then complaint volume, then top issue.
What the AI does, and doesn't do
- The AI never decides a pattern, a rating, or a color. Those come from the fixed rules above.
- The AI only writes wording. The AI Summary paragraph and bullets, and the "Why?" explanation, are written by Claude (from Anthropic) using only the figures shown on the page.
- Its words are checked. Before a summary is shown, software verifies that every number in it appears in the data and that direction words such as "up" or "down" match what actually happened. A summary that fails is discarded, and if it can't be fixed it isn't shown.
- The "Why?" explanation is only written when you ask for it. Companies with too little data get a fixed sentence instead of AI text.
AI text can still be imperfect. That is why every figure links back to the CFPB.
Limits of complaint data
- A complaint is an account from a consumer, not a finding. A pattern here is not proof of wrongdoing, and a clean pattern is not a guarantee.
- Counts are not adjusted for company size. Large companies have more customers and so more complaints. "Complaint volume" measures change over time, not how a company compares with another.
- "Timely" is as recorded by the CFPB. It reflects whether the CFPB recorded the company's response as timely, nothing more.
- Some patterns reflect the type of business. A debt collector will naturally have complaints concentrated in debt collection, for example.
- Small companies get "Limited Data". With too few complaints, percentages aren't reliable, so no pattern is assigned.
- Company names must match the CFPB's spelling. The same business can appear under several legal names.
- This is information, not financial, legal, or investment advice.