Piloting curriculum-driven AI career coaching with Per Scholas
When AI career coaching was woven into the program instead of layered on top, learners earned their certifications at higher rates.
Per Scholas is a national workforce accelerator that provides no-cost, intensive training for high-growth careers: IT support, cloud computing, cybersecurity, and more. Its courses run 12 to 16 weeks and prepare learners, many of them adults reentering the workforce or switching fields, to earn industry-recognized certifications and move into quality tech roles. Over the past 30 years, Per Scholas has graduated more than 35,000 program participants, with an average 2.5x increase in post-training income.
What sets Per Scholas apart is that technical instruction is only 80% of the model. Alongside the technical curriculum, learners work on career navigation skills like resume writing, elevator pitches, networking, and job-search prep, because a certification only changes a life if the learner can turn it into a job. The challenge is that career navigation support is expensive to staff and uneven to deliver. Every learner needs something slightly different, and there is never enough advisor time to go around. Per Scholas and CareerVillage wanted to know whether Coach, CareerVillage's AI career coaching tool, could help close that gap. And just as importantly, how it would need to be implemented to actually move outcomes.
Testing Coach two ways
Over two pilots in 2024, the teams studied Coach with 337 learners across 14 classrooms. Classrooms either used Coach or continued with Per Scholas's standard programming.
The two pilots were run differently on purpose:
In the first pilot, Coach was available but loosely integrated. Instructors decided when and how to use the tool, and learners were asked to complete a couple of activities partway through the program.
In the second pilot, Coach was built in. Working from the first cohort's instructor feedback, the teams co-designed a sequence of Coach activities mapped to each week's career-readiness topics and introduced Coach at program kickoff rather than midway through the program.
Structured rollout, stronger credential rates
Cybersecurity is the track with the strongest matched cohort, which makes it the cleanest comparison available. In the loosely integrated pilot, Coach learners showed no clear credential advantage over their matched peers. In the structured pilot, Cybersecurity learners who used Coach earned their certification at 89%, compared with 67% in a matched comparison group drawn from Per Scholas National/Remote Training Cybersecurity cohorts.
These credential results are best read as early, directional evidence rather than settled proof, but the pattern held across statistical methods, and it lined up with everything else the teams saw: the same tool did more when it was part of the program than when it sat beside it. (Full methodology, including the more conservative robustness checks, is in the white paper.)
Part of what's working is trust
The credential numbers tell you something happened, and looking more closely at the conversations helps us understand why. Across both pilots, learners said they were more willing to admit career doubts to an AI coach than to a person.
I hate networking… I can't share those thoughts with human career coaches, but I could do it with [Coach].
In workforce training, where many learners arrive carrying real uncertainty about whether a tech career is open to them, that ability to open up matters. A low-stakes place to think things through seems to be part of what helps people follow through: somewhere to ask the question you are embarrassed to ask, or work out what to say before it counts.
Facilitators felt the difference too
A tool that creates work for instructors will not survive contact with a busy classroom. In the flexible pilot, only one of three instructors said Coach reduced their workload. In the structured pilot, where learner adoption reached 92%, all three said it did. When the tool has a clear place in the curriculum and a reason to be there, both sides lean in.
I found that it cut the time significantly when teaching elevator pitches to learners and helped provide a solid foundation for their resume writing drafts.
Job outcomes is still an open question
The teams tracked early job placement in both pilots, but the follow-up windows were short and the samples were small, so the data cannot yet say whether higher certification rates translate into better jobs and pay. The certification gains are real; whether they translate is the next question. Longer-term outcomes are near the top of the list for a future study.
Takeaways for workforce programs
While it’s encouraging to see Coach having a positive impact, the clearest lesson here is that the design of the rollout is part of the intervention. For any program weighing an AI career coaching or navigation tool, a few questions are worth asking:
How will the tool be built into the program? The structured pilot's edge came from sequencing Coach into the weekly curriculum and introducing it at kickoff. Making a tool available and hoping learners find it may not be enough to drive meaningful impact.
Are facilitators brought in early? Instructor buy-in and co-design were what turned Coach from extra work into a workload reducer.
Are you set up to learn whether it worked? Decide up front which outcomes matter to you, such as certification, completion, or placement, and track them against a baseline like last year's cohort, so you can tell a real signal from wishful thinking.
The Per Scholas pilots are one data point, not the final word. But they suggest the interesting question for the field is not only whether to use AI for career navigation: it is how to implement it so it actually helps.
Join the conversation
We are continuing to study how AI career coaching can be built into workforce programs in ways that move real outcomes. If you are a workforce leader, educator, or funder thinking about the same questions, we would love to connect.
