all case studies
boosting employment through AI-directed job search
Algorithmic, skills-based assessments can improve hiring rates for entry level job seekers while reducing search time.
project type
measurement & evaluation
impact
12%
increase in employment outcomes
30%
reduction in time spent seeking work
1,117
participants across seven U.S. metro areas
how might we help low-income jobseekers access high-opportunity, good-fit jobs where they are likely to succeed?
Many entry-level frontline roles build highly transferable skills, yet jobseekers often search narrowly within familiar occupations because it’s hard to judge how their skills (especially soft skills) translate to roles in new sectors, and job search is time-consuming and costly.
Our experiment tests whether AI-based (predicts performance) and/or preference-based vacancy rankings help jobseekers find jobs faster and move across occupations, without sacrificing wages or satisfaction.
scope
Scale: 1,117 enrolled participants
Target population: Entry-level job seekers
Geography: Seven U.S. metro areas: Atlanta, Chicago, Cleveland, Las Vegas, Phoenix–Tucson, Bay Area (Sacramento–San Francisco–Oakland–San Jose), Seattle.
Occupations/sectors surfaced: Cashier, Childcare, Customer Service, Food Preparation, Orderly/Personal Care, Receptionist/Administrative, Sales Associate, Waiter, Warehousing.
approach
We built a job-search platform populated with real-time vacancies (sourced via Monster.com and Aspen Tech Labs). Participants completed a baseline survey capturing psychometrics, cognitive ability, and job preferences. The platform then displayed vacancies – including in sectors where participants had no prior experience – ranked in one of four ways depending on their treatment arm:
by an algorithmic fit score that predicts performance;
by stated preferences;
by both fit and preferences;
in a standard job-board format.
We then studied impacts on likelihood of employment, time to employment, and durability of employment.
Our AI fit score combines psychometric measures with performance on simulated tasks designed to reflect real job requirements, recommending best-fit vacancies without relying on prior titles or sector experience.
methodology
Design: Randomized Controlled Trial with 4 arms (control, preference-only, AI-only, combined).
Data sources: Baseline survey, follow-up surveys (28 and 91 days), and passive platform usage data (clicks, saves, applications via platform).
Analysis approach (as described): Longitudinal ANCOVA with strata and location fixed effects; survival analysis planned for time-to-employment.
outcomes
Employment: AI- and preferences-based rankings led to higher employment; jobseekers were 12% more likely to be employed than control.
Search effort: All treatments reduced search intensity; jobseekers found jobs 30% faster.
In all treatment groups, the number of applications submitted declined, indicating that jobseekers were able to focus on quality of fit rather than volume; interviews and offers stayed stable.
potential
This improved matching can increase wages, employment duration, career growth, and mobility for low-income workers. By introducing an objective skills signal into the job-search process – rather than relying on prior experience and titles – we can direct jobseekers to higher-opportunity, better-fit roles and build a more equitable, efficient labor market.
We are seeking:
partners to expand to other underserved populations (e.g., STARs) and into additional occupations;
funders to support further pilots;
job platforms and workforce development organizations to strengthen matching;
employers to pilot fit scoring within hiring workflows.
