Megan Skiendel: David, I want to start with a confession — I've been annoyed about something for three days and you're going to help me figure out if I'm right to be.
David Sterling: That's a high bar. What's the number?
Megan Skiendel: Eight days. The claim is that AI interviews cut time-to-hire by eight days on average. It is everywhere. Recruiting conferences, LinkedIn posts, vendor decks — everyone is quoting it like it's settled.
David Sterling: And you can't source it.
Megan Skiendel: I cannot. The closest I got was a reference to Ollie Forsyth — which is already not a household name in enterprise HR-tech — and even that trace flags 2016 as the origin. No current context, no update, no confirmation it applies to anything happening now.
David Sterling: And Art Papas? Bullhorn gets pulled into this conversation constantly — he's genuinely one of the most prominent AI advocates in the staffing and recruiting software space.
Megan Skiendel: Papas is real, Bullhorn is real — major global provider, legitimate player. But direct attribution of the eight-day claim to him? Also unconfirmed in the sources I looked at. He champions AI in recruiting broadly. The specific number just doesn't trace back to him either. So what you have is a figure that is load-bearing for the entire industry narrative, and it is, honestly, floating in midair.
David Sterling: Right — but the part that doesn't fit is that there are real numbers out there. Specific ones. Eightfold AI, Chipotle. So why does the industry need a phantom statistic when the actual case studies exist?
Megan Skiendel: That's exactly the tell — because the real numbers don't need the phantom. Eightfold AI, actual org deployments, six weeks compressed to one week. Chipotle's Ava Cado chatbot, twelve days to four days, that's a 75% cut for frontline workers. A Berlin software startup, 27 days to 7. LinkedIn's own hiring team — Tracy He and Shang Liu — published a 60% reduction in time-to-interview. All independent. All landing in the same 50-to-75% range.
David Sterling: That convergence is actually the story. So what's doing the work?
Megan Skiendel: It's not AI scoring your answers. It's asynchronous interviewing — candidates complete a structured video or text interview on their own schedule, no recruiter on the other end, no calendar puzzle. That's the mechanism. Remove the calendar dependency, the process moves forward.
David Sterling: Okay, paint it. No jargon.
Megan Skiendel: A nurse finishes a double shift at 11pm. She applies to a hospital job. Under the old model, she waits — honestly, ten days, sometimes two weeks — for a recruiter to find a slot that works for both of them. Under async interviewing, she records her answers on her phone that night. Done. The hiring process advances before she wakes up. That's the whole trick.
David Sterling: That's just scheduling software. Calendly solves that.
Megan Skiendel: Which is actually — wait, that's the follow-up question hiding right inside this. If it's purely a scheduling fix, what does the AI layer add?
David Sterling: Because Eightfold's number isn't 30%. It's 99% on time-to-interview. That's not a scheduling app.
Megan Skiendel: Right — but here's what I'd push back on. The 99% is the scheduling friction removal. The AI scoring layer, the candidate intelligence claims — that part isn't driving those numbers. Not yet. And that gap matters enormously because the bias risk isn't in the calendar. It's in the scoring.
David Sterling: So the metric that makes the tool look impressive and the metric that actually flags the liability are measuring completely different things.
Megan Skiendel: And that gap is exactly where I want to press. Because look at the actual mechanism behind Eightfold AI's 99% figure — it's not that the AI scored candidates more accurately. It's that you eliminated the calendar-alignment dependency entirely. The interview happens when the candidate is ready. That's it. That's the whole 99%.
David Sterling: I don't dispute the mechanism. But that doesn't make the AI layer worthless.
Megan Skiendel: No — but here's the test. Chipotle's 75% reduction, the Berlin startup's 74% reduction — both consistent with exactly that same explanation. Scheduling friction removed, number collapses. Neither case study traces the savings to the accuracy of AI-scored assessments. Neither. And HireStream and Bullhorn both differentiate on their NLP scoring layers, their candidate intelligence claims — but when you look at what the published numbers actually measure? It's the async format doing the lifting.
David Sterling: So your claim is — the intelligence layer is decorative.
Megan Skiendel: Not decorative, I mean — it adds something at volume. If you're screening 200 candidates, NLP keyword matching genuinely helps. But that's a filtering efficiency argument, not an intelligence argument. And vendors are selling it as intelligence.
David Sterling: Which raises the question — would a simpler async video tool, no AI scoring at all, produce roughly the same time reductions?
Megan Skiendel: Probably yes. And nobody wants to fund that study because the answer kills a premium pricing tier.
David Sterling: That's actually the load-bearing assumption the whole market is built on — that the scoring layer justifies the cost. If it doesn't, you're paying enterprise rates for a scheduling app.
Megan Skiendel: And the part that makes this significantly worse — which we'll get to — is that employers are scaling these tools on time-to-hire KPIs that have no way of capturing whether the candidate pool being fed through the scoring layer is being narrowed along demographic lines before anyone even looks at a résumé.
David Sterling: That's the structural trap. Employers are optimizing on time-to-hire — a metric that is completely blind to who got filtered out before the count even started. None of the case studies — not Eightfold AI, not Chipotle, not the Berlin startup — publish a bias-auditing methodology alongside those numbers.
David Sterling: So the question is — what's the actual mechanism? Like, where does the demographic narrowing happen?
Megan Skiendel: Picture a distribution center opening in Phoenix. Three hundred applicants in a week. The AI runs keyword-matching NLP against job descriptions — honestly, trained on whoever got hired at similar facilities in the past five years. If those prior hires were demographically homogenous, which they often were, the model just replicates that. The candidates it surfaces as 'strong fits' were never scored on merit. They matched a historical pattern. And the employer's dashboard shows four days to hire. That's the win they report.
David Sterling: And GINOTECH 2025, IRJAEM 2025 — those academic conferences are already presenting the next layer. AI-scored virtual interviews, deepfake detection. The technical frontier is accelerating faster than any governance structure can follow.
Megan Skiendel: Which means the liability gap just gets wider. By the time a disparate impact lawsuit surfaces — and I mean actually surfaces, gets certified, gets to discovery — you're talking three years minimum. In that window, Eightfold AI, Ava Cado, LinkedIn's system, whatever HireStream ships next — they will have processed and rejected tens of thousands of candidates from protected groups. Invisibly.
David Sterling: That's — the incentive structure is completely inverted. Speed shows up in a quarterly report. Discriminatory filtering shows up in a lawsuit.
Megan Skiendel: And once this is standard — once every ATS just has an AI screening layer by default — there's no control group left. Nobody's comparing AI-on versus AI-off. The bias is just the baseline. That's the real consequence of scaling on a KPI that was never designed to see it.
David Sterling: The number that makes the tool look essential is the exact number that hides the liability. That's the load-bearing problem.
Megan Skiendel: And that's the question. Bullhorn, Eightfold AI, Chipotle — they all have the speed numbers. None of them published a bias audit. Not one. So what I genuinely cannot answer is: do employers build the governance infrastructure before the first class-action filing arrives, or is that lawsuit the moment this industry decides it has a fairness problem?
David Sterling: I don't have a good answer to that. The incentive to audit shows up after the liability. Not before.
Megan Skiendel: No. And by then the 8-day figure — wherever it actually came from, whether it's Ollie Forsyth in 2016 or something nobody can trace — that number will be the least of it.
