How we turned decades of study advice expertise into an intelligence system that thinks with the team

The academy behind every real estate agent in the Netherlands

AVV, the Academie voor Vastgoed, is the education arm of NVM, the Dutch Association of Real Estate Agents. If you've ever bought or sold a home in the Netherlands, chances are your agent was trained here. Tens of thousands of real estate professionals rely on AVV for their certifications, continuing education, and professional development across multiple registries, each with its own rules and requirements.

The team at AVV knows this world inside out. They understand which courses count toward which registrations, how points expire, what the exceptions are, and how to guide someone through a maze of options to find exactly the right path. That knowledge is deep, specific, and built over years of doing this work every single day.

"I can't look at another study advice"

Every personalized study recommendation took about 30 minutes. The counselor would pull up a student's profile, check which registries they were enrolled in, look at their current point balances, cross-reference dozens of rules about what counts where, and then put together a tailored recommendation. It required real expertise, and it required full attention.

Then December would come around. Deadlines approaching, points expiring, and suddenly 350 requests would pile up. The same careful, expert process, repeated hundreds of times. The team was drowning in their own competence.

“At some point you thought: I can't look at another study advice. Now someone calls, and before we hang up, it's done.”

The problem wasn't a lack of knowledge. The problem was that all of that knowledge lived only in the heads of a few people, and there was no way to scale it without burning them out.

A new colleague, not a new tool

We didn't build AVV a recommendation engine. We built them a colleague. The team actually talks about the system like it's part of the group, which is exactly how we wanted it to feel.

We spent time with the counselors learning how they actually think through a recommendation. Not the clean version you'd draw on a whiteboard, but the messy real one: the edge cases, the judgment calls, the "technically the rule says this but we always handle it differently." We took all of that and encoded it into a system that checks live certification data from the registries, calculates point balances and gaps, matches courses to requirements across overlapping rule sets, and generates a branded PDF recommendation ready to send.

It started at about 80% accuracy. Not perfect, and nobody expected it to be on day one. The team reviewed the output, flagged what was off, and we improved it together. Like onboarding any new colleague, it took a bit of time to get right.

The important part is that the team manages the rules themselves. When a registry changes its requirements, they update the system directly. No developer needed, no ticket in a backlog. The people who understand the domain are the ones in control of it.

"I already have something ready for you"

A student calls in. They've been waiting weeks for their study advice. Before, the counselor would take their details, promise to get back to them, and add them to the queue. Now, they generate the recommendation while still on the phone. Two minutes. The student gets their advice before they hang up.

That was the moment it clicked for the team. The conversation went from "we'll get back to you" to "actually, let me walk you through what I'd recommend right now." Counselors went from processing a backlog to having actual conversations with the people they're trying to help. From 30 minutes of manual work to two minutes of review, and the rest of the time spent on the part that actually needs a human.