Modern Hiring Feels Like a Machine. Transparency and Human Accountability Fix It.

Most hiring teams reach for AI to move faster, and most candidates experience the result as a wall. An application goes in, nothing comes back, and the silence reads as proof that no person ever looked. The instinct is to treat that as a communication problem to patch at the edges in the form of a friendlier rejection email or a status page bolted on. The harder and more useful position is that the care a candidate feels has to be engineered into the system itself, decision by decision, or it never shows up at all. That reframes the AI question from 'How much can we automate?' into 'Who is accountable for what the automation surfaces?'

Christin Marshall is a Senior Talent Acquisition Specialist on the TA Ops team at Babylist, the digital destination for growing families. She built the company's in-house talent function largely from scratch over roughly seven years, from structured interview frameworks and ATS architecture to the AI tooling and governance that now sit on top of them. A Greenhouse Connect founding member and co-founder of a Black employee resource group, she works at the seam between the systems a company installs and the people who move through them. It's a vantage point that makes her direct about where technology belongs in hiring and where it doesn't.

"Care isn't something that you sprinkle on at the end like a finishing touch. It's 100 tiny details that are built into the system," she says. That principle only means something if it changes what gets built, and Marshall can point to exactly where it shows up.

Care is a system property

Marshall's core assertion is that the warmth candidates remember isn't a layer applied after the fact, but an accumulation of structural decisions made long before anyone applies. She points to the mechanics: service-level agreements written directly into the system, job descriptions that lead with skills so applicants can rule themselves out early, communication timed to the moment candidates find hardest. "For most candidates, their most anxious moment is definitely that period of waiting to hear back," she says. Her team holds itself to reviewing an application and either advancing or declining it within five to seven business days, on the logic that someone applying to dozens of roles deserves to know where they stand at every step rather than being left to guess.

The point is that none of these details are gestures. Each one is a commitment encoded into the workflow, which is what makes it reliable instead of dependent on whoever happens to be handling the requisition that week.

AI speeds the work but never makes the call

Where AI enters, Marshall is precise about its job. It compresses time, but doesn't make the final judgment. "It does speed things up, but it doesn't ultimately make any decisions," she shares. That line is the governing rule of her system, and she's blunt about the version of events candidates often imagine, the secret algorithm filtering people out by age or career stage. "At my company we require a human touch on everything. Every 'no' is from a human. Every 'yes' is from a human."

She's careful not to pretend AI is inert. It can surface signals a human would have missed, and her ATS uses matching and stack-ranking that runs candidate profiles against the role. But when an AI-written resume meets an AI ranking system, she notes, the two are effectively testing each other, which is exactly why a person has to sit at the end of it. The tool raises items for review. The decision to advance or reject stays with a member of her team every time.

Transparency is a design spec

Marshall treats openness as something the system does, not something the company says. Transparency in her process starts at the application itself, which discloses that interviews are recorded and explains how AI is used, and extends to a candidate portal that lays out the next steps at each stage. Candidates can opt in or out of AI tools like recording, note-taking, or voice interviews at any point. They can look back at the stages they've cleared and clearly see what comes next.

The reason she builds it this way is that the alternative subtly erodes trust. A candidate who falls in love with a role and only then discovers five hour-long panels plus a case study feels baited, and she's watched it happen to people close to her. "Radically transparent is always the best way. Sometimes the truth can be hard, but it's always the better way," she says. Told the real shape of the process up front, some people opt out, which she counts as the system working rather than failing.

Automating a broken process automates the mess

Marshall's advice to anyone building this is to resist adding more technology until the fundamentals are in place. "Don't even look at AI tools until your baseline is solid," she says. Clear job descriptions, defined process, structured interviews, and a mapped candidate journey come first, because the failure mode she and her peers in the TA Ops space see most is buying software to paper over a process problem. "AI on top of a broken process will just automate the mess, and it'll still be messy."

The deeper reason is that these systems only inherit the standard you give them. "AI only knows what you teach it, and if what you're teaching it varies, by person, by role, or by team, it's not going to work." Standardization is the precondition for automating anything without amplifying inconsistency. Get the process coherent first, and the tooling has something worth building on.

Someone has to be willing to answer for it

The last requirement is a person prepared to be held accountable for what the system produces. Because candidates will always suspect a machine made the call, Marshall believes someone must stand behind the decision and say plainly that a human reviewed the file, weighed specific factors, and owns the outcome. That accountability is also what lets her team lean on AI to flag fraud signals, duplicate applications, and inconsistencies in a landscape where bot-generated applications run high, without ceding the judgment to the tool.

It resolves back to a line Marshall keeps returning to: a tool can't provide the part of the candidate experience that's inherently human. "A tool cannot care about a candidate, ever. A person has to, no matter what."