Blog
How to compare issuer complaint counts fairly, with your own numbers
Raw complaint counts favor small issuers. A worksheet that divides by account size, adjusts the time window and product mix and reads response categories.
A list that says "Issuer X had 24,000 complaints and Issuer Y had 45" tells you which issuer is larger, and almost nothing about which one treats customers worse. The count rises with the number of customers, with the number of products and with how easy it is to file. To compare issuers, you have to divide, adjust and read, and this post shows how, using invented numbers that you replace with your own.
The method has six steps: choose a window and a product, get the count, get the size, compute a rate per million accounts, adjust for window and mix, and read what the complaints were about and how they ended. It is a worksheet, so you can follow it with a pencil. The numbers in the examples are made up, and no real issuer is named. Where this post mentions a figure about the market, it comes from the fact sheet this site uses, with its date.
Why raw counts mislead
Imagine two coffee shops. One serves 40,000 customers a day, and the other serves 150. If the first had 24 complaints this week and the second had 1, you would not conclude the first is worse. You would ask about complaints per customer. Card issuers are the same, except the range is larger. According to TransUnion, there were about 590.5 million bankcards in the U.S. in the third quarter of 2026 (as of September 2026), spread across issuers of very different sizes, from national banks to small credit unions. A count that does not account for that spread will always rank large issuers as the biggest problem.
There are other distortions, and the worksheet addresses each.
- Size of the customer base. Fixed by dividing by accounts.
- The window. A complaint database may show six months of data for one issuer and twelve for another, or a spike in a month because of an event.
- Product mix. A bank may have credit cards, deposit accounts, mortgages and student loans in one complaint file. Only the card complaints are relevant to a card comparison.
- Complaint type. A wave of complaints about a data breach or a system outage is different from steady complaints about billing disputes.
- Filing friendliness. A company that makes it easy to complain, or that tells customers how, may show more complaints than one that does not.
- Self-selection. Only people who complain are counted. They are not a random sample of customers.
Where the numbers come from
Complaints. The CFPB's public Consumer Complaint Database lists complaints about financial products, searchable by company, product, issue, state and date. You can filter to the credit card product, set a date range and export or count results. The CFPB's own capacity and update schedule have been in flux, according to press reports in 2025 and 2026, and its status after August 2026 could not be verified for this post, so check the date of the most recent data before you use it. The database shows what consumers submitted and what companies reported in reply. The CFPB does not verify every claim.
Account counts. This is the harder number. Large issuers sometimes give open or active account counts in annual reports, investor presentations or regulatory filings. The definitions differ: "open accounts," "active accounts" and "accounts with a balance" are not the same. Credit unions report member and loan counts in call reports to their regulator. If you cannot find an account count that is sourced and dated, do not use a number from a site that does not say where it came from. An alternative denominator is purchase volume or balances, but the same warning applies, and issuer-level market-share tables are mostly behind paywalls. A comparison with a weak denominator is worse than none.
Time. Use the same window for every issuer, such as the last 12 months or the last full calendar year, and write it down.
The worksheet, step by step
Step 1: choose the window and the product
Write "credit card complaints, 12 months ending [month year]." Use the same window for every issuer you compare.
Step 2: get the complaint count
For each issuer, count complaints for the credit card product only. Write the number.
Step 3: get the account count
Write the issuer's number of credit card accounts, in millions, with its source and date. If an issuer gives a range, note it.
Step 4: compute the rate
Complaints per million accounts equals complaints divided by accounts in millions. Here is an illustration with invented issuers.
| Issuer | Complaints (12 months) | Accounts | Accounts in millions | Complaints per million accounts |
|---|---|---|---|---|
| A, a very large issuer | 24,000 | 40,000,000 | 40 | 600 (24,000 / 40) |
| B, a mid-size issuer | 3,000 | 4,000,000 | 4 | 750 (3,000 / 4) |
| C, a small issuer | 45 | 150,000 | 0.15 | 300 (45 / 0.15) |
| D, a credit union | 6 | 80,000 | 0.08 | 75 (6 / 0.08) |
By raw count, A has the most complaints by a wide margin. By rate, B is highest at 750, A is at 600, and C and D are lower. That reordering is the point of the exercise. But keep reading, because the table is not yet adjusted.
Step 5: adjust for the window and the mix
The window. Suppose C's database entry covers only six months and the 45 complaints are from that period. To compare with the others on a 12-month basis you could double the figure to 90, and the rate becomes 600 per million (90 / 0.15). Annualizing assumes the second half looks like the first, which may not hold if there was a spike, so treat it as an estimate and say so.
The product mix. Suppose B's 3,000 complaints include 1,200 about deposit accounts that were filed under the same company name. The card-only count is 1,800, and the card rate is 450 per million (1,800 / 4), not 750. If you do not filter by product, you will compare unlike things.
A single event. Suppose A had 6,000 of its 24,000 complaints in one month because of a billing system outage that month. Without that month, the count is 18,000 and the rate is 450 per million (18,000 / 40). Whether to remove the spike depends on your question. If you want to know about everyday service, set it aside and say so. If you want to know how the issuer handles failures, keep it.
After these adjustments, the invented comparison looks like this.
| Issuer | Raw count | First rate | Adjusted rate | What changed |
|---|---|---|---|---|
| A | 24,000 | 600 | 450 | Outage month set aside |
| B | 3,000 | 750 | 450 | Non-card complaints removed |
| C | 45 | 300 | 600 | Six months annualized |
| D | 6 | 75 | 75 | None |
The ranking has changed again. A and B are level at 450, C is highest at an estimated 600 and D is lowest. That is why you should show your adjustments instead of a single number.
Step 6: ask how reliable the rate is
Small numbers are unstable. D's 6 complaints are 75 per million. If it had 9 the rate would be 112.5 (9 / 0.08). Three complaints changed the rate by half. For a small issuer, a rate based on a handful of complaints is a weak signal, and you should say "too few complaints to compare" instead of ranking. Large issuers have stable rates but may hide differences by product.
Read the categories
A rate tells you how often. The complaint categories tell you what for. Complaint databases usually label the issue (for example billing disputes, fraud or unauthorized charges, problems with a purchase, closing the account, credit reporting, rewards) and the sub-issue. Do the same count for each issuer and compare shares.
An invented example: of B's 1,800 card complaints, 40 percent are billing disputes (720), 20 percent are about rewards (360) and 15 percent are about account closures (270), and the rest are spread among other issues. If a friend asked "does this issuer handle disputes well?" you would look at the dispute share and the responses to it, not at the total.
Compare shares carefully. A high share of billing-dispute complaints might mean disputes are mishandled, or it might mean the issuer has many disputes because it has many online transactions. Neither reading is proved by the table.
Read the responses
A complaint database also records how the company responded. Typical response categories are "closed with explanation," "closed with monetary relief," "closed with non-monetary relief," "in progress" and "untimely response." These categories are the company's own reports to the system, and the consumer may dispute them.
Do not equate "closed with explanation" with a failure, because many complaints are misunderstandings that an explanation resolves. Look instead at the pattern: the share of complaints that ended with relief, the share where the company missed the response deadline and whether the share has changed over time.
An invented example: say that of 1,800 card complaints at B, 1,260 (70 percent) were closed with an explanation, 360 (20 percent) with monetary or non-monetary relief and 180 (10 percent) were in progress or untimely. If at A the figures are 80, 10 and 10 percent, A closed fewer with relief. Even then, a sample of narratives can show whether "explanation" was a clear reason or a form letter.
Read some of the complaints
Counts hide stories. Open a sample from each issuer, using the same filters, and read ten or twenty narratives. Look for:
- repeated patterns such as lowered limits soon after approval, payments not credited, unresolved disputes or rewards not posted;
- whether the complainants describe the same fix attempts that you would try;
- whether the company's public response addresses the problem or repeats a standard phrase.
One angry complaint proves little. The same story told by many people is a signal worth weighing.
Traps in the data
Company names. Complaints are filed under a company name, and a bank may appear under its legal name, a parent name or a brand. If you search one spelling you may miss complaints under another, and if you search a parent you may pick up products you do not want. Check the list of company names that match your search, and decide which ones belong to the card you care about.
Fintech and co-brand cards. A card sold under an app's name is issued by a partner bank, as explained in your card has a fintech logo, but which bank holds the account?. A complaint may name the bank, the brand or both, so the same problem can be split across two company entries. For a fintech brand you may need to add the partner bank's card complaints and the brand's complaints and divide by the combined accounts. The same applies to retail and airline cards.
Duplicates and re-filings. A person who files again after an unsatisfactory reply creates a second entry. That is usually a sign the first response did not settle it, which is informative, but it inflates the count.
Changing definitions. The categories of issues and sub-issues have been revised over time, so comparing a recent year with an old one can be misleading. Stay within a recent window.
Portfolio moves. When accounts move from one issuer to another, complaints and account counts do not move at the same time. Around a transfer, a rate can look strange for a period. Use a window that does not span the move, or say that it does.
A thirty-minute version
If you want a quick comparison and not a research project, this is the shortest route that still avoids the main errors.
- Pick two or three issuers you are actually considering, and one window of 12 months.
- In the complaint database, filter to the credit card product and count complaints for each.
- Find a sourced account count for each, with a date. If you cannot, write "no rate" for that issuer and rely on the narratives.
- Compute the rate per million, and note any obvious spike month.
- Read ten narratives each, and write one line on what you saw.
- Compare with other things you care about, such as fees, rewards and service channels, and treat the complaint rate as one input.
The result is a short note with three numbers and three sentences. That is more reliable than any single published ranking, and you can repeat it in a year.
What this method cannot tell you
Whether the issuer is a good fit for you. A low complaint rate means few people complained, not that you will have a good experience. Terms, rewards, fees and service channels matter too.
Causes. A complaint rate does not say whether the issuer was at fault. Customers may complain about lawful practices they do not like.
The current state. Data lags. A fix last month will not show up yet.
Regulatory history. Complaint data is not the same as an enforcement action. A consent order is a regulator's formal order against a company, and My card issuer was fined by a regulator. Does that affect my account? explains what a fine does and does not mean for your account.
A rating. This method produces a comparison for your own use, not a ranking. For why published ratings and awards are a different and weaker signal, see Is a highly rated card issuer better for me? Claims tested.
A blank worksheet
Copy this table and fill in one row per issuer.
| Issuer | Window | Card complaints | Accounts (millions) and source | Rate per million | Adjustments | Adjusted rate | Top issue share | Relief share |
|---|---|---|---|---|---|---|---|---|
Add a line below: what you read in the narratives and your confidence in the comparison (high, medium or low). If the account count is not sourced or the complaint count is small, write "low" and keep it in mind.
If you have your own complaint
A comparison helps you choose an issuer, and it does not replace using the system when you have a problem. The CFPB complaint process is free, and the steps are in How to file a CFPB complaint about a credit card, step by step. Your own complaint also becomes a data point that others will compare.
Key takeaways
- Raw complaint counts follow the size of the customer base; divide by accounts, in millions, to get a rate per million that can be compared.
- Use the same window and the card product only, and annualize or set aside a one-time spike only when you say you did.
- Rates from a handful of complaints are unstable, so label small-issuer comparisons as low confidence.
- Read the issue categories and the response categories, then sample the narratives, because the pattern matters more than the total.
- A complaint rate is not a rating or proof of fault, and it says nothing about whether a card suits you.