Predicting leaks before they happened.

The brief was uncomfortable. Leaks of SF6 (a potent insulating gas used in high-voltage switchgear) were happening across the network, and nobody had a clear picture of where, how often, or why. There was data: IoT sensors, drone imagery, video from on-site inspections, paper logs. There was no team to make sense of it.

We built one.

In eight weeks, we hired a six-person discovery squad: a data architect, a data scientist, and four engineers, mostly on contract. We mapped the data landscape, worked with environmental and engineering specialists to turn their tacit knowledge into features the model could learn from, and consolidated the patchwork of sources into one usable dataset.

The hardest call came early. The senior sponsor wanted us to skip discovery and "just build the model". We held the line. A predictive model trained on data nobody trusted would have been worse than no model at all. We did the discovery work first. The modelling came second.

We also reframed the brief. The original ask was "find the leaks". We pushed it to "predict the leaks", because the funding case for prevention was stronger than the funding case for reaction.

Outcome: £1.75 million of new funding secured for the Net Zero initiative, on the back of the data and the model. The contract was taken in-house so the operator could continue to develop it. As a side benefit, we worked with on-site engineers on AR-enabled access to digitised station infrastructure using Meta Quest hardware, cutting the time to find inspection records in the field.

What this shows: building a data product from a vague brief. Hiring the team, defining the data model, navigating a senior stakeholder, delivering something that paid for itself. This is what Mission Control looks like at scale.