Launching Filters for Application Screening

Not everything in an application should be scored. Filters give screening agents a better way to reason about the difference.

When we built AI Application Screening, we designed criteria around a simple idea: evaluate the parts of a candidate’s application where degree matters.

How relevant is their experience? How strong are their achievements? How well does their career trajectory match the role? How deep are their skills?

These questions need judgment. A candidate can be a low, medium, or high match. The difference matters, and it should contribute to how the application is scored.

But as customers started building increasingly sophisticated screening agents, we noticed another class of criteria appearing.

Is this candidate located in Germany?

Are they at least 18 years old?

Are their salary expectations within range?

These look like screening criteria, but they are fundamentally different. There is no meaningful spectrum to evaluate. A candidate isn’t a “high match” for being over 18.

The agent simply needs to determine whether something is true, false, or unclear.

Today, we’re introducing Filters for AI Application Screening to make that distinction explicit.

Scoring and filtering are different jobs

This distinction might sound small. We think it matters quite a lot.

A good screening system needs to answer two fundamentally different questions:

How strong is this candidate?

And:

Does this application meet a particular condition?

Until now, customers could use criteria to answer both. But doing so has two consequences.

First, screening agents accumulate criteria that don’t actually require evaluation. A simple eligibility check gets represented with the same machinery as evaluating years of relevant experience or the quality of someone’s achievements.

More importantly, criteria contribute to the candidate’s overall assessment. That makes sense for things that describe candidate quality. It makes much less sense for things like location or salary expectations.

We believe a screening score should have a clear meaning. It should represent the strength of an application, not become an aggregate of every fact an employer wants to know about it.

Filters let us keep those concepts separate.

How Filters work

When configuring an Application Screening agent, teams can now add Filters alongside screening criteria.

Criteria evaluate something where strength matters. For example, 7+ years of experience working on LLM research. The agent evaluates the evidence and determines the strength of the match.

Filters evaluate conditions. For example, 18 years or older. The agent determines whether the condition is met, not met, or cannot be determined from the available information.

That distinction also makes the output easier to reason about. Instead of turning every piece of information into another score, the agent uses the right evaluation method for the question being asked.

Filters can also be configured as disqualifiers. If an applicant clearly doesn’t meet a required condition, they can be moved directly from In Review to Disqualified, leaving teams with a cleaner pipeline of eligible talent.

Some filters require more care than others

Building Filters also forced us to think carefully about which decisions screening software should enable.

Location is a good example.

Location and country can be used in discriminatory ways. But they can also represent real constraints. A company may only employ people in countries where it has a legal entity. A role may require work authorization. An employer may not offer visa sponsorship.

Removing the ability to express these constraints doesn’t remove them from hiring. It just pushes the decision into a later, more manual part of the process.

We therefore decided to support these filters, including the ability to use them as disqualifiers. That flexibility comes with a responsibility to use them in accordance with applicable employment and anti-discrimination laws.

‍Disability status is different.

At first glance, it may seem surprising to include it as a filter at all. The reason is not to evaluate disability as a qualification.

There are cases where employers need to identify that a candidate has disclosed a disability so the information is handled appropriately, accommodations can be considered where relevant, and candidates are not inadvertently overlooked because of their status.

Separating this information from scoring is therefore an important part of the architecture, not just a UI decision.

A more precise model for screening

The broader direction behind Filters is straightforward: screening agents should model hiring decisions with more precision, rather than reducing every decision to a score.

Some information requires judgment > Criteria

Some requires a check > Filter

Some should influence candidate ranking > Criteria

Some should remain explicitly outside of it > Filter

As screening agents take on more of the work that previously required humans to read and structure thousands of applications, these distinctions become increasingly important. The goal isn't to automate a spreadsheet of criteria. It's to give the agent the right primitives to reason about an application correctly.

Filters are one of those primitives.

‍

They are live for all HiPeople customers today, including jobs already in progress. Filters are free to use, don’t consume screening credits, and adding them does not trigger a rescreen.

We’ll add more predefined filters over the coming weeks, including university and average time at previous jobs. Filters will also come to our AI Interviewer agent.

‍

And in early Q4, we plan to introduce AI Filters: instead of choosing only from predefined options, teams will be able to describe what they want to filter for, and HiPeople will create reusable filters for their organization.

Criteria were the first step toward letting agents understand what makes a candidate strong.

Filters are the second step: understanding what simply needs to be true.

‍