From spreadsheets to self-running operations: GAF's AI-powered training infrastructure

Overview

GAF runs the largest roofing contractor training program in North America, 1,200 events per year, 51 field trainers, one coordinating operation. It used to run on a spreadsheet. Now it runs itself.

Phos AI Labs rebuilt the operation from the ground up: automation foundation first, AI layer on top. The result was a 75% reduction in training administration time and a system that enforces the right sequence, routes the right information, and gives leadership real-time visibility across every trainer and every event, without asking field trainers to behave like administrators.

The Problem

At GAF’s scale, every gap in the automation chain becomes a task for a human.

• Wrong class closure sequence corrupted LMS records, someone had to find and fix each error by hand.
• Trainers got enrollment links late, and checking rosters meant opening a different system entirely.
• Non-training travel had no submission pathway into the calendar at all.
• Travel that wasn’t tied to a training event had no place to go, it fell off the calendar completely.
• Matching the right trainer to the right job was a manual judgment call, no system could cross-check badge level, region, and availability at the same time.

The Objective

Build a system that enforces the correct operational sequence automatically, routes information to the right people before anyone has to ask, and gives leadership real-time visibility across all 51 trainers, 1,200 events, and every open action item.

The constraint: field trainers are experts in roofing, not software. The system had to work for them, not the other way around.

What we built: from spreadsheet to self-running intelligent infrastructure

At Phos AI Labs we built an operation that runs itself, three versions in eight months.

How It Works

Version 1: from spreadsheet to operational foundation:

• Trainers see only their own classes; admins see everything, from any device.
• Classes can’t be closed incorrectly anymore; the system enforces the right steps automatically.
• Rosters and enrollment links go to trainers automatically, no manual handoffs.

Version 2: from reactive to proactive:

• Trainers get notified when something needs their attention, before anyone has to chase them.
• Travel and non-training time now has a submission path and shows up on the calendar automatically.

Version 3: from operational tool to intelligent infrastructure:

• The access to edit a training automatically locks as the date gets closer.
• Admins can update system rules on their own, without touching code.
• QR codes track attendance at in-person sessions.

The AI Layer

With the automation foundation in place, the AI layer is now in build and designed to:

• Recommend the right trainer for each job based on badge level, region, and availability.
• Show which trainers perform best where, and which training types cause the most errors.
• Answer new trainer questions instantly, without pulling experienced team members away.
• Send reminders at the right moment based on what’s actually on each trainer’s calendar.
• Draft and send team communications automatically.

The Outcome

• 75% reduction in training administration time.
• 1,200 training events managed annually across 51 trainers; no spreadsheet dependency.
• 51 field trainers fully mobile; updating statuses, closing classes, and accessing rosters from job sites, airports, and training centers across North America.
• Zero manual coordination required for class closures; the system enforces the correct sequence automatically.
• Full real-time operational visibility for leadership across every trainer, every training, and every open action item.

The trainers haven’t changed. What changed is that the system no longer asks them to behave like administrators.

The GAF deployment is the training coordination use case. The pattern; automation foundation first, AI layer on top, structured data as the connective tissue; is what Phos AI Labs builds for any organization running field operations, certification programs, or distributed workflows that depend on manual coordination to hold together.

What’s next

Eight months of automation work created something more valuable than efficiency: the foundations AI actually needs to work. Every completed training now generates clean records, documentation, attendance, error flags, trainer comments.

The AI layer is in build. The operation that already runs itself is about to get smarter.



If your operation still depends on manual coordination to hold together, let’s talk.