Main content

Connected Homes data brings British Gas closer to customers

Phil Wainewright Profile picture for user pwainewright December 8, 2014
Most people interact with their energy supplier no more than 10 minutes per year. At British Gas, the Connected Homes team aims for an everyday relationship

British Gas traces its history back to the world's first public utility company, the Gas Light and Coke Company, founded in 1812. Later nationalized, the brand was spun off and is now owned by Centrica.

How does such a long-established organization reinvent itself for a digital future? For British Gas, the answer was Connected Homes, a new business unit set up in 2012 tasked with bringing the company into the digital age. Inevitably, this is a story where the Internet of Things — initially in the shape of smart thermostats and heating boilers — looms large.

The Connected Homes team is best known for its Hive service, launched last September, which allows consumers to remotely control the heating systems in their homes from their mobile phones. A year after launch, 130,000 UK homes are already using the Hive system.

The team also produces smart energy reports for consumers based on data collected from smart energy meters, and is currently trialling connected boilers in 700 homes.

Collecting more data

The aim is to change British Gas from a remote utility to one that is an everyday companion in its customers' lives. Most people interact with their energy supplier for just 10 minutes per year. The Connected Homes team can change that tradition, says Jim Anning, head of data and analytics.

It's all very, this was your bill last quarter - that's where the industry has been. There's a real opportunity to turn that into, this is what's happening in your home now.

To achieve that aim means stepping up the rate at which data is collected. Smart meter customers currently have their data collected daily or hourly. Annings' team then runs algorithms across the data to break it down into heating, lighting, entertainment, hot water, cooking and so on. But having such a long interval between data points limits the usefulness of the analysis. It's not granular enough, for example, to detect when a kettle or toaster are used.

One of the things we're looking to do next year is an experiment with data every ten seconds. At that level you can provide an awful lot more insight.

I think this is the way we see it going. Liveness and real-time - we all expect everything now.

Scaling data analysis

To handle this volume of time series data, it was essential to move to a more capable platform, says Anning, who I met at last week's Cassandra Summit in London.

All of these sensors splurge out masses of data. So you've got a lot of immutable, time series data and for us there's two things: where and how do you store that — how do you protect your ability to ingest that data — while at the same time wanting to run some computationally intensive data science algorithms across that, in order to be able to turn that data into something valuable and meaningful.

So we have this real-time problem with lots of data we want to do some fairly heavy lifting across. Cassandra and Spark together seemed to fit that sweet spot.

Jim Anning British Gas
Jim Anning, British Gas

The Cassandra-Spark combination will go live at the beginning of next year, and was a prerequisite for contemplating collecting data as frequently as six times a minute, says Anning.

The 10-second stuff we just went straight to Cassandra for that. It was clear that you wouldn't be able to scale that using traditional technologies.

Like a lot of people in the industry, I get frustrated by the 'big data' thing — what the hell's that? — but I've come to the realization that what big data means is, you have got so much of it that the current capabilities and technologies you have to deal with it are starting to creak. That's the point.

The realization of dealing with more and more granular data and more and more real-time data — if we built something on the traditional technology, it would probably work today, but we would be building a pretty poor legacy to hand on and to future proof.

Predictive boiler service

The platform also provides the opportunity to try out new ideas, such as the connected boilers project.

Modern boilers are much more complicated than they used to be. Largely that's because of various regulations about how efficient they need to be. When they break down, if people are left cold and without hot water it's a pretty painful, gruesome experience.

Today we know the minute the boiler has failed. Often the customer hasn't noticed for hours or days. We've had situations where people have been on holiday and they've come back to a voicemail saying, 'Your boiler's failed, are you going to be around tomorrow, we'll get an engineer out to you with the right part.'

The sexy thing from my perspective is the data science side of it. The data scientist I've got working on this stuff used to do something very similar for NASA. They wanted to rank aircraft flights. It turns out that aircraft taking off and flying and landing isn't too dissimilar from boilers heating up and staying hot and then cooling down. So this guy is applying some of the very similar principles to what are used in aircraft for boilers.

I love the idea of a Minority Report style engineer turning up just long enough to say Hello before your boiler fails.

British Gas is visiting 50,000 homes a day. There's a lot of opportunity, if we can develop the data science — it's still an if, it's something we're working on — to be able to fix something for someone when you were coming round for the annual service is a great thing.

We're building iPhone apps for the engineers themselves so that when they go out to a customer with a connected boiler, they can call up the history of that. That's good because quite often it's an intermittent fault. For the engineer to be able to call that up and see and not have to go away and come back again, from a customer perspective is a massive advantage.

A grey colored placeholder image