Sustained Hiring Trends Hold Up Better Than Spikes — And Now We Can Prove It

TalNexis flags two different shapes of hiring acceleration: a Hiring Velocity signal (a job category growing for three straight weeks) and a Category Spike (a single week that suddenly doubles). Both say "this company is hiring more here than usual" — but do either of them predict anything real, or do they just catch noise that fades right back to normal? Unlike our earlier stock-performance backtest, this question doesn't touch the market at all — it's a direct test of whether our own signal types are measuring something durable.

How we tested it

For every Hiring Velocity and Category Spike signal detected between March and October 2026, we recorded two numbers from the moment it fired: the baseline (the pre-signal weekly posting count in that category) and the signal level (the elevated count that triggered the signal). Then we looked 5–12 weeks later and measured the average weekly posting count in that same category, at that same company, during that follow-up window.

Three outcomes are possible: the elevated hiring held or grew further (follow-up ≥ signal level), it partially faded (follow-up landed between baseline and signal level), or it fully reverted (follow-up dropped back to or below baseline — the spike was noise).

One more rule: independent observations only. Some companies fire the same signal many times in a few months, with their 12-week follow-up periods overlapping. We only count a new observation for a given company + category + signal type once the previous one's 12-week window has fully closed.

And one data-quality rule: exclude sync-outage artifacts. A company whose job-board source breaks goes silent for a stretch, then dumps a backlog once it's fixed — a burst that can look exactly like a real signal but isn't. We exclude any signal within 21 days of a company's source resuming after a 10+ day silent stretch, which removed 26 of 3,533 candidate signals (mostly at Walmart and Replit).

Together, these rules took the usable sample from 2,043 candidate signals down to 871 genuinely independent, clean observations — 263 Hiring Velocity signals and 608 Category Spike signals.

What we found

Neither signal predicts continued acceleration — about a third of both types fully revert to baseline, which is expected: any signal built on crossing a threshold will partly be catching one-time noise, a statistical fact known as regression to the mean. But the two signal types are not equivalent. Hiring Velocity signals retain 59% of their elevated hiring rate on average 5–12 weeks out; Category Spike signals retain only 50%. That gap is small in absolute terms but statistically real: t = 2.95, p = 0.0034, on 871 independent observations. A three-week sustained trend is a meaningfully better predictor of durable hiring than a single strong week.

Which job categories hold up, which don't

We broke retention down by job category (categories with at least 8 independent observations). The split is striking and, honestly, a little counterintuitive: the categories people assume are "hot" right now are the least sticky, and the ones nobody writes headlines about are the most durable.

AI/ML is the headline here: it's the category our subscribers probably care most about, and it's also one of the least durable — a hiring surge in AI/ML retains only about a third of its peak level two to three months out. That's consistent with how AI/ML hiring often actually works: bursty, project-driven pushes (standing up a new model team, a research sprint) rather than steady headcount ramps. Public Sector and Supply Chain, by contrast, look like genuine structural build-outs that keep going.

What this isn't

• Not a claim that signals predict growth. They don't, on average — most elevated hiring fades. What we found is a relative difference between two signal types, not evidence that either one forecasts expansion.
• Regression to the mean is partly baked in. Any signal defined by crossing a threshold will catch some one-time noise by construction, so some decay is mathematically expected even if the underlying hiring pattern is completely real. The comparison between signal types controls for this (both are measured the same way); the absolute retention numbers don't fully.
• One time window. March–October 2026 only. Category-level results in particular (Public Sector, AI/ML, etc.) could reflect this specific period's conditions rather than a durable pattern.
• Correlation, not mechanism. We don't know why Velocity holds up better than Spike — only that it does. It could be that three-week trends are inherently more deliberate (harder to fake with one busy recruiting week), or it could be something else entirely.

What this means in practice

If you're watching a company and a Category Spike fires, treat it as a thing to note, not a thing to act on — roughly a third of these turn out to be noise, and even the ones that aren't tend to fade. A Hiring Velocity signal is worth more weight: it's measuring three consecutive weeks of real commitment, not one busy sprint, and that shows up in the data as a genuinely stickier result.

And if the category is AI/ML specifically, read any signal with extra caution regardless of type — it's the most volatile, headline-grabbing category in our data and also one of the least durable. The categories that quietly keep compounding — Supply Chain, Public Sector, Operations — rarely make anyone's hiring-trends newsletter, but they're the ones actually worth watching long-term.

We're continuing to validate our own signal taxonomy against what actually happens next — this is the first of several planned checks. If you want to see Hiring Velocity and Category Spike signals as they fire, explore TalNexis signals.