Automation Belongs in Screening. So Does Judgment.

Most organizations running a background screening program already rely on automation, and increasingly on AI-enabled tools. They may not have selected those technologies directly. They often arrived inside tools already being used for record matching, name resolution, data classification, document review or adjudication logic.

The question is no longer simply whether automation is part of the process. It is whether anyone can explain where it sits, where AI is involved, and what either one is allowed to decide.

That distinction matters more than the technology itself, so it is worth being precise about the terms.

What we mean by automation, and what we mean by AI

Automation, in this context, means rules-based processing. Someone wrote the rule, the system applies it, and the same input produces the same output every time. If you want to know why a file routed the way it did, you can go read the rule.

AI-enabled tools differ in one specific respect: they infer. Illinois, which now regulates the use of AI in employment decisions, defines artificial intelligence as a machine-based system that infers how to generate outputs such as predictions, content, recommendations, or decisions. California’s employment regulations use a broader framing, covering automated-decision systems that may draw on artificial intelligence, machine learning, statistics, or other data processing techniques.

The practical distinction for a screening program is not sophistication, and it is not which label a vendor uses in a sales deck. It is whether you can trace the rule that produced a result, or whether the system derived that result from patterns in data. Both belong in a screening program. They require different kinds of oversight, and they are starting to carry different disclosure obligations.

What automation is genuinely good at

Screening produces a large volume of repetitive, structured work, and automation handles a lot of it better than people do. Parsing court records into consistent formats. Flagging discrepancies between an application and a returned record. Routing files by jurisdiction. Identifying when a report is missing a required component. Tracking turnaround against a service standard.

These are pattern tasks with clear inputs and clear outputs. Used well, automation at this layer can reduce avoidable delay and inconsistency while freeing experienced staff for the work that actually requires judgment.

Where a person still has to stand

The trouble starts when automation moves from processing information to interpreting it. Consider three points where interpretation is unavoidable.

Identity matching. A common name, a shared date of birth, a transposed digit in an identifier. Confidence scoring narrows the field, but someone has to decide whether a record belongs to this person. A false match is not a data quality issue. It is a wrong outcome attached to a real name.

Adjudication of ambiguous records. Charges that were amended, dismissed, deferred, or sealed. Sentences recorded inconsistently across jurisdictions. Older records with thin detail. A matrix can sort the clear cases; the unclear ones need judgment applied against a written standard.

Disputes. Simply rerunning the same query against the same source may reproduce the same answer without resolving the issue the candidate raised. A meaningful dispute process has to address the substance of the dispute, not simply repeat the original search.

In each case the risk is the same: an automated step produces a defensible-looking output that no one examined, and the organization treats it as a decision.

Automation Does Not Erase Accountability

This is where the operational question becomes a legal one.

The Fair Credit Reporting Act still governs employment screening regardless of how a report is assembled. When the Consumer Financial Protection Bureau withdrew several FCRA guidance documents in May 2025, including its 2024 circular addressing background dossiers and algorithmic scores, the underlying statute did not disappear. Employers and consumer reporting agencies remain responsible for their respective obligations under the law.

States have moved into that space. Illinois amended its Human Rights Act effective January 1, 2026, requiring notice when AI is used in employment decisions and prohibiting the use of zip code as a proxy for a protected class. California’s automated-decision-system employment regulations, effective October 1, 2025, extended record retention for that data to four years and define certain agents acting on an employer’s behalf as covered entities. Colorado replaced its original AI act in 2026 with a disclosure-oriented framework, effective January 1, 2027, that requires notice, an explanation after an adverse outcome, meaningful human review, and three years of records.

The specifics differ. The direction does not. Employers are increasingly expected to know how a decision was produced and to be able to say so.

What to ask your screening partner

Six questions produce most of the useful answers:

1. Where in your process is automation making or influencing a determination, rather than organizing information?
2. What triggers human review, and who performs it?
3. How are identity match confidence thresholds set, and who reviews the ones that fall below?
4. What does your audit trail capture when an automated step contributes to an outcome?
5. How does a dispute get routed to a person rather than back through the same automated path?
6. If we are asked to explain a specific outcome, what can you produce, and how quickly?

A partner with a mature program answers these plainly. Vagueness here is the finding.

The standard is explainability

The organizations handling this well are not the ones that avoided automation, or the ones that adopted the most of it. They are the ones that documented the boundary — this is automated, this requires review, this is the threshold, this is who owns the call — and can walk an auditor, a client, or a candidate through it.

That documentation is not a technology decision. It is a program design decision, and it belongs to you as much as to your screening provider.

Compliance does not require you to move slower. It requires you to be able to explain what happened.

‍

‍

Liberty Screening Services builds background screening programs designed for consistency, defensibility, and clear accountability. Contact us to review how your current program handles its automated steps.

‍