With the pressures of high-volume hiring mounting, employers now lean on AI to expedite background verification, a traditionally slow, manual process. But while the benefits are undeniable – faster turnaround, reduced cost, and scalability – there’s a growing need to ensure that automation doesn’t come at the expense of data privacy or fairness.

Chris Harper, CEO of education and employment verification specialist ZippedScript, gave us an inside look into how their technology operates behind the scenes – and what safeguards are in place to protect candidate data and uphold ethical hiring practices.

A three-pronged approach to privacy

Data privacy, Harper explained, is embedded in the system’s design and guided by three core principles: data minimisation, self-hosting, and human oversight.

“We look to minimise the amount of data at every step,” he said, outlining how the system is engineered to collect only the essentials – from what is gathered from the candidate, to what is input into the AI, and what is retained after verification. Thus, less is more when it comes to personal information.

Unlike many companies that rely on cloud-based AI providers, the firm opts to self-host its models. The move isn’t just about technical control but also accountability.

“We were never able to get comfort around the control of end-user data [with cloud-hosted solutions],” Harper explained. “Our customers have strict requirements around data sharing.”

Releasing information to a third party just wasn’t viable. However, by hosting the models internally, they ensure data never leaves their secure environment. Once the model completes a task, it doesn’t retain any personal or confidential data. It’s as if the model has a short memory.

To ensure accuracy and catch any hiccups the AI might miss, human reviewers double-check each verification result. Harper said:

We make a conscious decision to have human eyes review every result in a manner akin to a QA function.

Consent and compliance: Staying in step with the law

While AI is often seen as a mysterious black box, the company follows a transparent approach when it comes to candidate consent. Harper likens their process to traditional background checks, with AI simply taking on the heavy lifting previously done via phone calls and emails. The process just replaces manual labour with machine efficiency.

“The consent required is therefore the same as a traditional background check, which already contains these types of disclosures,” Harper said.

The system also aligns with major privacy regulations, such as GDPR and HIPAA, thanks in part to their internal hosting setup.

Self-training and hosting models in our own environment really limits data privacy concerns.

“Having the model and its data sit within our network allows us to treat the data in a similar manner to that of simpler technology that is used by peers including client dashboards and APIs,” Harper said.

“If anything, AI models have simplified compliance because it means we have less eyes seeing datapoints and less staff endpoints to secure.”

However, don’t expect to peek under the hood anytime soon. The AI models remain proprietary.

Bias, fairness and the power of binary outcomes

One of the major concerns in AI adoption is the risk of algorithmic bias, a subtle but serious threat that can unintentionally skew hiring decisions.

But Harper said their system was designed to avoid that trap entirely.

“Our AI produces a binary and factual outcome, such as a name or graduation status … we return results that are devoid of interpretations; we return what we see,” Harper said. “AI tools are purposely trained to avoid interpretation which could later lead to bias, to altered results.”

No false positives, no false promises

While the AI is robust, it isn’t left to run unchecked. Human reviewers don’t just act as a formality; they’re crucial for maintaining accuracy. Automated rules can catch obvious errors, like numbers in a name field.

For the small percentage of errors the AI might miss, human review is still the gold standard.

Candidates also have recourse to dispute or correct their information. The process is documented and communicated to clients, and follows the same policy as before AI was introduced. Data retention, too, remains unchanged. They don’t retain data in the AI – only in the backend database, and only for as long as necessary.

Speed meets ethics

The value proposition of AI in hiring isn’t just about doing things faster but also about doing them smarter. For companies under pressure to fill roles quickly and cost-effectively, AI offers a competitive edge.

Speed and repeatability are the biggest gains, according to Harper, who highlighted downstream improvements to their turnaround times.

Yet, they are careful not to let automation overshadow ethics. With every AI output reviewed by a person and privacy controls baked in from the ground up, the company aims to strike the right balance between technological efficiency and human responsibility.

In short, while the background verification process may now be powered by lines of code and machine learning models, the guiding principles remain very human: protect privacy, act fairly, and get the facts right. In an age of algorithms, it’s reassuring to know that some things – like ethics and accuracy – aren’t being left to chance.