Work / Merit America
From scattered signals to one view of the service.
Merit America's career program is a service. Learners move through it, coaches guide it, and a dozen systems record every step. I treated it that way: working alongside the coaches, I read the whole journey across 300+ learners and pulled outcomes, engagement, and coaching into one picture leadership could act on.
The problem
Merit America was making high-stakes calls about learner success without a map of the service delivering it.
The signals that mattered sat in different places, Salesforce records, learner surveys, and coaches' own observations, so no one could see where the journey lost people, or step in before it did.
300+ learners across multiple cohortsMy role
I led the research and synthesis. With the coaches as frontstage partners, I built the interview guides and surveys, ran the qualitative interviews, and pulled the quantitative data out of Salesforce.
Then I stitched the lived experience and the system data into a single narrative, the kind leadership could read in one sitting and act on the next morning.
The process
No single source told the truth on its own, so I read them against each other.
Stakeholder interviews to agree on the questions that actually mattered to the program, before gathering a single data point.
Structured surveys with learners, 1:1 interviews with coaches across cohorts, and cleaned quantitative data from Salesforce.
Read side by side, the three streams surfaced patterns no single source could: where the journey held, and where it slipped.
Key findings
The hard parts
I was briefed with a single sentence: what predicts whether a learner gets placed, and how quickly. Everything after that, the method, the definitions, the limits - I had to figure out.
With no templates, I was left to determine what would even count as an answer.
My moveI designed the study myself, starting with the inputs a placement program can actually influence: applications sent, interviews landed, and coach touchpoints. Then I deliberately left the analysis open to surface other factors, which is the only reason prior IT experience, education, and certifications ever entered the picture.
At that point, Merit America was a start up in the process of centralizing information. So I had to create order and organize the scattered inputs I found. I pulled from Salesforce records, self-reported learner data, job outcome surveys, coach calendars, individual job trackers, resumes, and LinkedIn, across three cohorts and months of work. I had a hard time filling the gaps and a harder one reconciling the contradictions such as self-reported application counts that did not match portal submissions, and a cohort tracker with missing information for half its learners.
My moveI reconciled it source by source and built the analysis skills the job needed as I went, learning enough Excel to chart distributions and set the intervals for comparing placed against non-placed learners. Where a source was too thin to trust, I left the comparison out rather than publish a number I could not defend.
"Time to hire" and "coach touchpoint" sound self-evident until you try to count them. Does a recruiter call count as an interview? Is a mass job blast worth the same as a 1:1? Without fixed definitions, you can't compare across cohorts to make solid findings.
My moveI defined every input before measuring anything, and published the definitions with the findings. A job is a learner's first accepted outcome. Touchpoints are weighted, with personal contact counting double. Time to hire runs from graduation to an accepted offer, in months rather than days, because exact dates could not always be verified. Then I stated the limits up front: which learners and cohorts fell outside the window, and which placements were excluded from the application findings because volume had not driven their outcome.
The intuitive hypothesis is that learners who send more applications get hired. It is the assumption a placement program runs on, and it was the one I expected to confirm.
My moveAccepting that the data disproved it. Placed and non-placed learners submitted roughly the same volume, averaging around 48 applications, so I followed the question further into the interview data instead. The signal pointed toward application quality and knowing which jobs to target, not volume, which sent the recommendation somewhere entirely different: after the two-month mark, stop optimizing for quantity and coach toward fit.
A report like this can turn people into numbers if you're not careful. Learners who took longer than six months sat outside my scope, but outside scope is not the same as unimportant. I did not want the exclusion of these outlier users to read as a dismissal.
My moveI accounted for them: treated exclusions as an adjacent scope, never as judgment, and kept asking what each pattern said about the person inside it. For example I discovered that the learners who succeed the most usually started college but were not able to finish. That one fact reframed recruitment around something true about a real person, and it went on to shape how the program talks about itself.
This is the project where I realized I was good at the full cycle of user research, and that I genuinely enjoyed it. It is also why I think of this as learning design rather than reporting. Focusing on the learners themselves: their lives, motivations, limitations changed the stakes and drove the entire process.
Outcomes
Leadership took the findings straight into the service. The early weeks got a new intervention protocol, aimed at the exact touchpoints where the journey was quietly failing people.
Coaches were retrained around the check-in rituals that worked. Cohort outcomes measurably improved, and the work fed Merit America's wider organizational growth strategy.
Next project
A B2B vendor-sourcing strategy and one-stop system for language services across the organization.