Where should automation stop? asks Alice Jeffries, Head of Marketing and Communications, Secure Screening Services.
A system can tell you that two employment dates don’t match. The harder question is whether the mismatch matters. It could be an innocent error, of course. Perhaps a payroll date differs from somebody’s last working day. Or maybe a recruitment agency and the organisation where someone was placed record the same period differently.
But the discrepancy may also point to something that genuinely needs further investigation. Recognising that two pieces of information are different is a technical task and deciding what the difference means is a judgement. As employment screening becomes more automated, this distinction is increasingly important. So, the question is no longer whether technology can carry out more of the screening process. It’s whether every part of the process should be automated in the first place.
Some decisions are easier to automate than others
There’s plenty of screening work that technology can handle extremely well. Systems can identify missing information, send reminders, track references, compare dates, monitor outstanding activity and flag cases that don’t follow the expected path.
At scale, this matters. Removing repetitive administration can make screening considerably quicker and give employers far better visibility of what’s happening. But there’s a difference between automating a task and automating a judgement, and that difference becomes obvious when a case stops being straightforward.
Real employment histories contain uncertainty. People work through agencies, become self-employed, take career breaks, work overseas and move between different forms of employment. Previous organisations disappear, records conflict, evidence arrives in different formats and from sources of varying reliability. A rules-based system may identify all of those things as exceptions, but determining their significance is harder.
This brings us to a much older idea about practical judgement: knowing a rule is not the same as knowing how to apply it in a particular set of circumstances. Rules are enormously useful when cases are predictable, but screening often becomes difficult precisely when they aren’t.
Unusual is not the same as significant
There’s a danger in assuming that because technology can identify something unusual, the unusual thing must represent risk. It might not.
An employment gap certainly isn’t evidence of any wrongdoing. Neither is an unavailable reference, a discrepancy between dates or a period of self-employment. These things may need to be examined, yes, but there’s an important distinction between something we don’t yet understand and something we’ve reason to distrust. That distinction requires judgement, along with a degree of humility about what the available evidence can actually tell us.
One of the interesting consequences of increasingly sophisticated technology is that uncertain information can start to look more certain than it really is. A score looks precise; a flag looks definitive; a recommendation looks like an answer … But precision of output isn’t the same thing as certainty of knowledge.
However sophisticated the technology becomes, the underlying evidence may still be incomplete, contradictory or open to more than one reasonable interpretation. Good screening needs to preserve that distinction.
Screening is fast when a candidate remembers everything and has the documents to hand. Most fall short, in more combinations than any single employer sees. We see them tens of thousands of times a year and technology is used to learn from them and improve the process. The hard judgements stay with people: they’re good at them, and automating judgement adds risk. – James Matlock, Chief Data Officer, Secure Screening Services
Automation doesn’t remove responsibility
There’s another question underneath all of this, who is responsible for the decision? Technology can distribute the work involved in reaching a decision. It can collect evidence, compare information, identify patterns and recommend that somebody takes a closer look. What it can’t do is make responsibility for the eventual decision disappear, and this matters particularly when screening technology comes from a third-party provider.
An employer may interact with a portal, receive a status and eventually see an outcome without having much visibility of what happened between those points. That might be convenient, but it can also create distance between the organisation making the employment decision and the process that helped produce it. This is why employers need to understand where automation sits within their screening process.
What information is the system analysing? What causes an exception to be raised? Does the technology make recommendations or decisions? At what point does somebody review the evidence? And, crucially, does that person have enough information and authority to reach a different conclusion?
Those are governance questions as much as they are technical ones.
Human involvement needs to mean something
“Human in the loop” has become a familiar phrase in discussions about automated decision-making, but the presence of a person doesn’t necessarily mean meaningful human judgement is taking place. If a system produces a recommendation and somebody routinely approves it without considering the evidence independently, little has really changed.
The Information Commissioner’s Office (ICO) has made this distinction increasingly important in recruitment. Following changes introduced by the Data (Use and Access) Act 2025, significant solely automated decisions are permitted in a wider range of circumstances, but safeguards remain. These include informing the individual about the decision, allowing them to make representations, obtain human intervention and contest the outcome.
The ICO has also specifically warned employers that automation in recruitment can improve efficiency while creating risks around unfairness, bias and transparency if the appropriate safeguards are not in place. That is important legally, but there’s a more fundamental principle underneath it.
A fair decision is not only about reaching the right outcome. It’s also about having a process through which that outcome can be understood, questioned and, where necessary, corrected. That matters a lot in employment screening, because the information being considered is about a person’s history, and that history won’t always fit neatly into standard categories.
Fairness depends on the process as well as the outcome
Imagine that a screening system identifies a discrepancy. The candidate might know immediately why it exists, but if they’re never given a meaningful opportunity to explain it, a technically accurate system can still contribute to a poor decision.
That’s why procedural fairness matters. People need appropriate opportunities to correct inaccurate information, explain discrepancies and provide additional evidence. This doesn’t mean every screening decision needs lengthy human intervention, it just means that the process should be capable of recognising when intervention is necessary. Good automation should make those moments easier to identify, not harder.
Speed matters, but certainty has limits
There’s an understandable commercial pressure to make screening faster. Employers have roles to fill, contracts to mobilise and people waiting to start work; candidates have little interest in being stuck unnecessarily in a screening process. Technology should absolutely help, but the goal shouldn’t be the fastest possible decision. It should be the fastest defensible decision.
Straightforward cases should move quickly; complex cases should become visible early enough for somebody to deal with them properly. Technology should remove unnecessary administration, make evidence easier to assess and direct human attention towards the cases where it adds most value. That’s a more useful measure of automation than simply asking how many steps have been removed from the process.
What should employers expect from screening technology?
As screening becomes more digital, employers should be asking more of technology than whether it makes the process quicker. They should understand what it makes visible, for example:
Does it show where information came from?
Can somebody understand why a case was flagged?
Are exceptions clear?
Can incorrect information be challenged and corrected?
Is there a sensible route from automated processing to human review?
Can the organisation ultimately explain the decision it has made?
If the answer to those questions is unclear, the problem isn’t necessarily that there’s too much technology. It may be that responsibility has become too difficult to see.
The aim is not maximum automation
There’s a temptation with new technology to measure progress by how much human involvement can be removed, but employment screening needs a more thoughtful measure. Technology is exceptionally useful where the task is repetitive, rules-based or involves finding patterns across large amounts of information. Human judgement becomes infinitely more valuable when evidence is incomplete, context matters or several explanations remain possible. It’s important to remember that the two are not competing approaches; the best screening processes should use each where it’s strongest.
Technology can help us establish what we know, it can also help show us what we don’t know. The difficult judgement is deciding what to do with the difference, and that responsibility should remain visible, wherever the technology sits.
