Cosmo Schools: Automating Admissions for a Multi-Campus School Network

Cosmo Schools runs one of Colombia's fastest-growing private school networks. Backed by one of the country's largest social enterprises, they operate over 11 campuses across the Medellín metro area with thousands of enrolled families. Growth was never their problem. Their systems couldn't keep up with it.

Every enrollment cycle, tens of thousands of families entered an admissions process held together by manual follow-ups, disconnected tools, and data nobody fully trusted. Leads fell through the cracks. Waitlists lived in spreadsheets. The commercial team couldn't tell which campuses needed attention until it was too late.

Three issues compounded each other:

Enrollment data lived across four different platforms, none of them in sync. The same student could show different statuses depending on where you looked. Leadership couldn't confidently answer basic questions like "how many students have actually paid?"

Waitlist management was handled by one person with a spreadsheet. Recovery outreach was copy-paste. When a spot opened at a campus, someone had to manually check who was next, send a message, wait for a reply, and repeat. During peak enrollment, this broke down.

Nearly half of all pre-assigned enrollment spots were being released because families never completed payment. Thousands of leads stalled between registration and their first campus visit. But without reliable data, the team was making allocation decisions on gut feeling.

We worked as an embedded extension of the Cosmo team over a 12-week engagement across six sprints. The work fell into three layers.

We built a production-grade data warehouse on Google Cloud Platform, designed to meet enterprise security requirements. BigQuery replaced dozens of manual exports and Excel transformations as the central analytical layer.

• Managed environments (dev, staging, production) through Terraform with full CI/CD
• Automated ETL pipelines consuming data from HubSpot and BotMaker APIs on a scheduled basis
• Security compliance reached 85% on CIS 4.0 benchmarks
• Everything documented and version-controlled via GitHub

For the first time, the entire admissions funnel — from first WhatsApp message to final tuition payment — was visible in one place, updated automatically, and trustworthy.

• Full-funnel visibility across every campus and grade level: lead → qualified → immersion → admitted → invoiced → enrolled, tracked daily
• Lead source attribution that connects marketing spend to families who actually enroll, not just families who click
• Automated lead prioritization through a scoring engine that flags high-opportunity leads every 2 hours based on real-time campus capacity
• Data reconciliation across 30,000+ lead records that had undermined trust in reporting for months

The dashboard changed how the team allocated effort across 11 campuses. Before, prioritization was a weekly conversation based on whoever had the freshest spreadsheet. After, it was a daily reality driven by live data.

Recovery flows. The existing follow-up system sent the same generic message to every family regardless of where they dropped off. We redesigned the entire sequence across WhatsApp and email to be stage-aware, time-sensitive, and routed to the right team member automatically.

• A family that didn't pay the immersion fee gets a different message than one that didn't show up to their visit
• Follow-up windows tightened from 24 hours to 4 hours for initial recovery
• Unresponsive leads get tagged and escalated without anyone checking a list
• Remarketing audiences are created automatically for paid social retargeting

Waitlist automation. Two workflows that previously depended on a single person were fully automated. When spots open, eligible families are notified instantly. Waitlisted families are offered spots in chronological order, with automatic advancement if there's no response within 48–72 hours.

Lead prioritization engine. An automated system that cross-references real-time campus capacity with lead data to surface the highest-value opportunities. Campuses running below 80% capacity trigger priority tagging on every matching lead from first contact. The system runs autonomously every two hours.

We didn't hand over a deliverable and walk away. We embedded with the Cosmo team, joining their weekly syncs, pairing with their engineers, and running knowledge transfer sessions throughout the engagement.

The structure was sprints, but the approach was iterative: ship something useful fast, validate it with the team, then build the next layer. Quick wins like lead prioritization went live within days. Structural work like the data warehouse was built in parallel and deployed progressively.

Every automation included runbooks. Every dashboard included training sessions. By the final sprint, the Cosmo team was running their own iterations on the systems we built together.

We used BigQuery because Cosmo needed a warehouse that could handle 30,000+ lead records with sub-second query times and integrate directly with Looker Studio for dashboards the commercial team could actually use without engineering support. Terraform gave us reproducible infrastructure across three environments without manual configuration drift. Python and Go handled the ETL pipelines and automation logic respectively, with GitHub Actions managing deployment. On the communications side, HubSpot and BotMaker via the WhatsApp Business API powered the recovery and waitlist flows where families actually respond: their phones.