Five hundred schools. Thirty-nine states. Millions of industrial workers who’d never taken an online course. One product team.
When I joined SENAI’s digital education initiative, the brief was something like: “Bring Brazil’s industrial training system into the digital era.” I initially assumed this meant migrating content. It meant, in fact, rebuilding infrastructure, culture, and process for an organization that had been doing in-person vocational education since 1942.
Here’s what I learned.
Scale Changes the Problem, Not Just the Numbers
In product management, we talk a lot about scaling. Usually it means adding capacity — more servers, more support staff, more content. At SENAI, scale changed the nature of the problem entirely.
In a single school, you can run a pilot, talk to teachers, iterate. You have feedback loops measured in days. At 500 schools, the feedback loops stretch to months. A miscommunication in a training document doesn’t affect one cohort — it affects tens of thousands of students before anyone notices and reports back to the team.
This forced a discipline I hadn’t expected: product decisions at SENAI were infrastructure decisions. Choosing a standard for how content was packaged (we chose SCORM, and later pushed hard toward xAPI) wasn’t a content format choice — it was a choice that would affect how 500 schools produced, tracked, and reported on learning for the next decade.
When I recommend SCORM to a client now, I’m making a bet with a long tail. At SENAI, that was made explicit in a way few product environments ever are.
The Four Layers of a Learning Ecosystem
It took us a while to name what we were building, but eventually we landed on a framework: a learning ecosystem isn’t a platform. It’s layers.
Layer 1 — Infrastructure: The LMS, the content standards, the authentication, the data pipeline. The plumbing. Boring, invisible, essential. Ours ran on a national cloud that had to work in schools with intermittent bandwidth — which meant offline-first design wasn’t a feature, it was a requirement.
Layer 2 — Content: Courses, assessments, simulations, multimedia. At SENAI, this included AR experiences for industrial equipment that couldn’t be shipped to every school (imagine 500 nuclear reactor maintenance manuals when there are only 3 reactors in the country — the 3D simulation was cheaper and safer than the alternative).
Layer 3 — People: Teachers, coordinators, regional administrators, IT staff. The human layer is the most important and the most underinvested in EdTech. Technology doesn’t transform education. Teachers who trust and understand the technology do. We spent as much time on change management and training as we did on product features.
Layer 4 — Partnerships: SENAI’s reach extended through Google (Classroom integration that was the first of its kind globally), Atos (the AI engine for skills gap analysis), content providers, and national employers who defined what skills actually mattered.
Most EdTech products live entirely in Layer 2. They wonder why adoption struggles. Layers 1, 3, and 4 are usually the answer.
The Insight That Changed Everything: Skills Are Forecasts
The moment that shifted our AI work from interesting to essential was a realization about what employability data actually is.
Traditional vocational education says: “Take this course, get this certificate, get this job.” It’s a static mapping from curriculum to credential to career.
The Brazilian industrial economy doesn’t work like that. Skills required for a CNC operator in the automotive sector in São Paulo are different from those needed in Porto Alegre, and both are changing as automation evolves. A national training system that doesn’t respond to those differences is training people for jobs that are disappearing.
Our skills gap AI engine — built with Atos using machine learning on job market data, employer surveys, and SENAI’s curriculum taxonomy — turned this static mapping into a dynamic one. We could say: “In this region, for this sector, employability is 75-82% given this set of credentials — but adding this module raises it to 89%.”
That’s not a feature. That’s a reason for a 39-million-person organization to exist in its current form.
What “Digital School” Actually Means When There’s No Building
One of our products was called SENAI Digital School. The name implied a metaphor I had to actively resist: it is not a school on a screen.
The students were adults — industrial workers with jobs, families, commutes. Some were studying by phone, on lunch breaks, in factory parking lots. The idea of “a school” with a schedule, a course catalog, a learning path, a teacher who knows your name — none of that applied.
What they needed was:
- Learning in modules short enough to complete in a single coffee break
- Progress that persisted and was visible (“I am 60% through this certification”)
- Evidence of learning that their employer would recognize
- Community, because isolation kills motivation even when the content is good
We built all of this. The hardest was community — not technically, but culturally. Industrial workers in Brazil aren’t used to being learners in public. Making it safe to ask “dumb questions” in an online forum took more effort than the forum itself.
The Mistake I Made Early On
I treated adoption as a metric and not as a signal.
Early in the rollout, adoption numbers looked reasonable — schools were technically using the platform, courses were being assigned, completions were being logged. Success, right?
When I dug into which schools were adopting and which were going through the motions, I found a pattern: the schools with enthusiastic coordinators (who’d been trained, who understood why this mattered, who saw it as their project and not a mandated tool) had 3-4x the completion rates and dramatically higher teacher satisfaction scores.
The schools in the middle — technically compliant but culturally unconvinced — were producing ghost students. Completions logged, nothing learned.
This is the EdTech trap. Platform adoption and learning adoption are different things, and the latter requires human investment that most product teams don’t budget for.
What EdTech PM Work Actually Requires
After years building learning products at SENAI, Voxy, and Skillshare, here’s what I’ve come to believe is different about PM work in education:
Outcomes are long-tailed. You can’t A/B test learning. Whether someone is better at welding or English or data analysis takes months or years to surface. You build proxy metrics — engagement, retention, assessment scores — and trust the research.
Content quality is a product variable. In most product domains, content is someone else’s problem. In EdTech, the quality of what’s being taught directly affects your DAU, your NPS, and your employer partnerships. You’re not just building the container; you’re responsible for what’s inside.
The student is usually not the buyer. SENAI’s buyers were employers and regional departments. Voxy’s were enterprise HR teams. Skillshare’s were individuals — but even there, gifting and team accounts muddied the signal. Understanding who writes the check and who actually learns are two different discovery tracks.
Motivation is a first-class design concern. People stop learning when it’s hard, boring, or irrelevant. Only one of those is a content problem. The other two are product problems.
The Thing That Still Surprises Me
The Google Summit.
SENAI was recognized by Google as a global education partner for our Classroom integration — the first LMS-Classroom integration of its type in the world. We stood on a stage in front of 500 educators from across Brazil, showing a demo of something we’d built because a technical constraint (schools already had Google accounts) became a design decision (use what’s already there).
Sometimes the most elegant product choices happen when you’re solving the wrong problem.
This article draws from my work building SENAI’s national digital learning ecosystem across 500+ schools. For the full case study, see SENAI Digital School and SENAI Educational Resources Repository.