❓ Problem

Hiring for judgement has always been guesswork, and AI just made the guessing harder. 42.6% of US job seekers used AI the last time they updated their résumé, and 7.5% let it write most or all of it. When everyone's CV is polished by the same chatbot, "creative, strategic, exceptional problem-solver" stops meaning anything. Employers already know it: 86% of US employers report problems with résumés, and 45% say they're hard to rank.

The gap is widest for roles built on taste. A coder can point to a GitHub. A designer can point to a portfolio. But the person with an uncanny eye for which ad will win, which thumbnail gets the click or where a checkout will lose people has no equivalent proof. Their portfolio shows what they shipped, not whether their calls were right. And getting those calls wrong is expensive: the US Department of Labor's oft-cited estimate puts a bad hire at 30% or more of first-year earnings.

Meanwhile, the bar for "good judgement" just moved. In WONO's experiment, an off-the-shelf AI predicted which YouTube packaging people would click better than the large majority of humans who took the test. That cuts both ways. Most people can't beat the machine, which makes the ones who can genuinely rare, and genuinely worth finding.

✅ Solution

Humark, a human talent benchmark platform where people go head-to-head with AI on commercially valuable calls, scored against what real audiences actually did.

• Challenges built on ground truth, not opinion. Pick the winning ad, the stronger thumbnail, the subject line that got opened, or the checkout step where users dropped off. Every challenge is built from a real, statistically significant A/B test or behavioural dataset, so there's a right answer and nobody is grading on vibes.
• Every challenge has an AI opponent. The same question goes to the leading models. Your profile shows not just your score but which models you beat, on which skills, and by how much.
• Retesting separates skill from luck. Anyone can get lucky on ten questions. Humark keeps feeding fresh, unseen challenges and only awards verified status to people who hold their edge across rounds. This is exactly how the Good Judgment Project found its "superforecasters": it skimmed off the top 2% of a huge public pool, and roughly 70% of them held that status the following year.
• A verified talent pool companies pay to access. Participants opt in. Recruiters, agencies and brands search by skill ("top 1% at predicting paid social winners, beats every frontier model on DTC creative") and hire with evidence rather than adjectives.

📊 Key Numbers

Market size

• Estimates for the pre-employment assessment market range from $2.8B to $4.8B in 2025, growing around 9-10% a year. Skills testing is now mainstream: 76% of employers use skills tests.
• The US alone has 454,800 advertising, promotions and marketing managers, with marketing managers earning a median of $166,790. That's before you count copywriters, designers, growth marketers, CRO specialists and YouTube strategists.
• Agency recruiters typically charge 15-25% of first-year salary for contingency placements. On a $130k marketing hire that's ~$26k a placement, which is the price anchor Humark undercuts.
• The second buyer is AI labs. Mercor went from zero to a $2B gross revenue run rate by selling expert human judgement to labs for model training. A pool of people verified to beat frontier models at predicting human behaviour is exactly the kind of talent labs pay a premium for.

ARR potential

• Top-down: 1% of a ~$3-5B assessment market is a $30-50M business, before any recruiting or AI lab revenue.