Quality Control Standards

Reviewing every report by hand is impractical at volume, and reviewing none leaves the consumer to find the error, so a CRA should define in writing which reports receive a second examination.

• Review every report containing adverse information. An inaccurate adverse record can cost the consumer the position, and it is the error most likely to produce a dispute or a claim. This is where a second reading returns the most.
• Review every record matched on the minimum identifiers. A match resting on name and date of birth alone should be examined again before release, particularly when the jurisdiction offered nothing further to work with.
• Review files where the search returned records belonging to more than one person. The reviewer should be able to see which records were reported, which were set aside, and what distinguished them.
• Review anything a researcher flagged. The person who performed the work knows which files were difficult, and a flag is the least expensive signal a quality program can collect.
• Put the criteria in writing. Criteria applied differently from one file to the next are not a procedure and cannot be described as one in a dispute or an audit.

Pre-release review asks whether the report accurately reflects what the sources contain, which is a different question from whether the research was thorough.

• Confirm the identifiers on every record. The reviewer should be able to see what established the match and whether any element of the record conflicts with what the consumer supplied.
• Confirm the disposition against the source. What the court currently shows is what belongs in the report, and a disposition carried forward from an earlier collection may no longer be current.
• Check for duplicates. A case located through a database, a county search, and a state repository should appear in the report once.
• Confirm the dates. The filing date, the disposition date, and the date each item was verified should all be present, since reporting limits cannot be evaluated without them.
• Confirm that each record falls within the applicable limits. Federal and state reporting limits both apply, and the stricter of the two governs what may be reported.
• Confirm that unable-to-verify results appear in the report. An item that could not be verified should be reported with its reason rather than dropped from the file.

Pre-release review finds errors in the files it examines, and sampling released reports establishes how many errors that review is missing.

• Draw samples on a defined schedule. Sampling performed when capacity allows produces results that cannot be compared from one period to the next.
• Include clear reports in the sample. A record that was never reported generates no dispute, which makes an incomplete clear report the error nobody reports.
• Re-run part of the search rather than only re-reading the file. Re-reading finds reporting errors. Running the search again is what finds research errors, including a jurisdiction that was never searched.
• Sample by researcher, by client, and by jurisdiction. Errors cluster, and an aggregate error rate hides where the cluster sits.
• Record what each sample found, including the samples that found nothing. A clean sample is evidence that the program operates, and it is evidence only if someone wrote it down.

A dispute is a consumer telling the CRA that something went wrong. Handling one dispute corrects one report, while analyzing disputes together shows where the procedure itself is failing.

• Record the basis for every dispute, not only its outcome. The outcome shows whether the item survived reinvestigation. The basis shows what the consumer said was wrong, which is the part that carries diagnostic value.
• Categorize disputes by cause. An incorrect match, a stale disposition, a duplicate entry, and a record past the reporting limit are four different problems calling for four different corrections.
• Categorize by jurisdiction, source, and researcher. A cluster in one county usually points to a source problem, while a cluster under one researcher usually points to training or workload.
• Track the deletion rate. Items deleted because they could not be verified on reinvestigation raise a question about how they were verified the first time.
• Review dispute data on a defined schedule. Data examined only when volume rises identifies the problem after it has already grown.
• Account for disputes that arrive indirectly. A consumer who raises the issue with the employer or through a reseller may never appear in the CRA’s dispute log, which makes the internal rate look better than it is.

Correcting a disputed report satisfies the reinvestigation obligation under FCRA § 1681i and does nothing about the next report carrying the same defect.

• Establish what permitted the error rather than who made it. A standard that allows one person to report a weak match allows everyone to report one, and replacing the person leaves the standard in place.
• Revise the standard when the standard was the defect. Where two identifiers were treated as sufficient and should not have been, what changes is the definition of a match rather than the handling of the one file.
• Retrain when the standard was adequate and was not followed. These are different failures, and applying the remedy for one to the other leaves the actual cause untouched.
• Discontinue a source that repeatedly produces errors, or verify everything it returns. A source that has been wrong several times will be wrong again, and continuing to rely on it without added verification is a decision the CRA may have to defend.
• Measure whether the correction worked. A change made without a follow-up measurement leaves the CRA unable to say whether the problem was solved or simply stopped being noticed.

Court researchers, international partners, and data providers produce work that is furnished under the CRA’s name, so their errors reach the consumer as the CRA’s inaccuracies.

• Sample vendor work against the same standard as internal work. A vendor reviewed more loosely than staff becomes the least examined part of the process while carrying the same exposure.
• Track results by vendor. A single deficient source disappears inside an aggregate figure and will keep producing errors until it is measured on its own.
• Compare vendor results against the source where the jurisdiction allows a direct check. This is the only reliable way to detect a vendor returning clear results without having searched.
• Set turnaround and accuracy expectations in writing, and measure against them. Expectations that were never written down will not support ending the relationship later.
• Watch for coverage gaps. A vendor that has quietly dropped a county will continue returning clear results, and those results establish nothing about that county.

An auditor, a client, or a court will ask what the quality control program consists of, and what the CRA can produce is the answer to that question.

Maintain the following:

• The written review criteria and what triggers a review
• The sampling schedule, the sample sizes, and the results of each sample
• Dispute data categorized by cause, jurisdiction, source, and outcome
• Corrective actions taken, the dates they were taken, and what changed afterward
• Vendor review results, kept by vendor
• Training records identifying who was trained on what, and when

Reviews that identified no errors belong in the record alongside the ones that did. They establish that the review took place rather than that nothing was ever wrong.