top of page

A digital twin for one job: live weather and site planning for a roofing company

6 days ago
3 min read

Reliance Roof and Clad is a roofing company in Northland. Whether a crew can safely work on a roof depends on the weather at that site, and the people deciding that also need to know how each job is tracking. That information was spread across job software, accounts, email and weather forecasts, so we built them a live digital twin of the business, called the Command Centre.


Manager in a roofing company office looking at a wall screen showing a weather map and project cards
Sample with info randomised for privacy. AI-generated image.

What's on the screen

It's a screen on the office wall. The middle of it is a map of the region with each open project on it, with live weather over the top: rain, wind, temperature and estimated roof temperature, which you can step through for the next seven days. Project cards and alerts sit around the map, and they're connected to the job data, the quote reader and the financials.


Site-by-site weather calls

The weather side does the most specific work. For each site, the forecast is checked against shut-down and caution limits for gusts, sustained wind, rain, roof temperature and air temperature. The result is a go, caution or shut-down call for that site. Those limits started from WorkSafe NZ guidance and the roofing code of practice, and the company can change them on an alerts page.


Estimating roof temperature

We also estimate roof surface temperature, which isn't something a normal forecast gives you. A dark steel roof in full sun and light wind can run about 30 to 35 degrees above the air temperature, so the crew gets a heat alert and a finish-by time. It's labelled as an estimate on the screen, because it is one.


Checking forecasts against an ensemble

A forecast can also be confidently wrong, so a shut-down call is checked against a 51-member weather ensemble. If fewer than a quarter of the members agree, it's downgraded to caution. Crews shouldn't be stood down because of one model run.


Making the numbers trustworthy

We made two rules before anything went on the screen, because a dashboard like this gets ignored the first time someone finds a wrong number. Every figure has to be checked against the other sources first, and it carries a mark showing whether it's confirmed, probable or possible. If it can't be verified, it's marked as possible or it doesn't appear. Each source also has to keep refreshing, and if one stops, a notice goes in the header. Silent old data is worse than a visible gap. When we checked the map, one site turned out to be 2 km from where the records put it, which is the kind of mistake you only find by looking.


Read-only connections

We also kept the system from acting on its own. Its connections to mail, calendars and files are read-only. It can draft an email for someone to read, but it can't send one.


Why it was built for one job

This is why we built something specific. A generic dashboard would have shown less and been trusted less. Most of the value came from choosing what it had to get right: is it safe to be on that roof, and what's the state of each job? The same weather modelling comes from our work on LandSure, where we've built 3D models of land and its live conditions.


Source

Source: AVANT project documentation for the Reliance Command Centre. Published with Reliance's agreement. Details in the image are randomised for privacy.


Comments


bottom of page