Case file · Airport ground support

Knowing what breaks next — and who should fix it

Aeroworks installs and maintains the passenger boarding bridges at eight Mexican airports. Their technicians had been coordinating in a Telegram group since 2024 — every fault, every repair, every photo. None of it was data. Now it is.

  • 8Airports
  • 40Bridges at Cancún
  • 5,000+Tickets recovered
  • 24/7Live since Feb 2026

The work already had a home. It just wasn't data.

Operators, supervisors and mechanics ran the entire operation out of one Telegram group. A bridge fails, someone reports it, photos and video go up, a mechanic picks it up, the fix gets confirmed. It worked — which is exactly why it had survived two years and thousands of messages.

What it could not do was answer a question. Which component fails most often? Who has repaired this before? What will we need in stock next month? Every answer was in there, spread across the conversation, and not one of them could be counted.

Two registries, not one

In the summer of 2025 a bot joined the group, with a workflow behind it that turns what people report into structured maintenance tickets. Then two things were modelled — and the second one is what the rest of this story depends on.

Every asset All 40 bridges at Cancún, down through their auxiliary equipment to the individual components that actually fail.
Every person Each operator, supervisor and mechanic — their hard skills, and a Belbin profile describing how they work alongside everyone else.

Maintenance systems routinely model the first. Modelling both, in one place, is the difference between a maintenance log and a plan.

Eighteen months of history, read in a single pass

The workflow was then run backwards over the group's own archive, all the way to January 2024. Roughly eighteen months of conversation came out the other side as 5,000+ structured maintenance tickets.

That backfill is why forecasting worked immediately instead of a year later. A system that starts collecting on the day it is installed has nothing to reason about until it has accumulated a past. This one arrived already holding one.

Since February 2026, it doesn't stop

The bot now reads the group as people post to it, and a live dashboard reflects the state of the operation continuously — every hour of every day since February 2026. Nobody files anything twice. The technicians work the way they always have, in the group they already used.

What having both halves makes possible

Because the equipment and the people are connected inside one model, Peerforce answers questions that neither half could answer alone:

What is going to fail Component-level failure patterns drawn from thousands of historical tickets, projected forward.
What to have on the shelf Spare parts estimated statistically from those same patterns — ordered before the part is needed, not after it is missed.
Who works which shift Maintenance shifts organised around what is likely to break and who is actually qualified to fix it.
Who grows into what Team composition and professional progression built on real skills and a real repair record, not on who was available.
peerforce — chat Cancún
What should we stock for next month, and who can fit the repairs?
◆ Components ranked by failure probability from the full ticket history — each one with the technicians certified to do the work, and whether they are already on the roster for that week.

An illustration of the kind of question the model can answer, not a record of a specific one.

Scheduling a preventive repair means knowing the failure is coming, having the right technician rostered, and having the part on the shelf. Those are three different questions, usually asked of three different systems by three different people. At Aeroworks they are one question, asked in plain language, of a model that holds every bridge, every component, every technician and every skill that has ever touched them.