A Swiss startup takes on AI at the sensor edge

At CES, I ran into a startup at Avnet’s booth called Octonion. It was showing off a little brick comprised of more than half a dozen sensors that ran software for predictive maintenance and anomaly detection. Avnet was selling the brick, which was designed to help industrial customers get a device connected to the internet and sending data for analysis.

The partnership with Avnet and Octonion also includes STMicroelectronics, whose chip is inside the brick, and a deal with Microsoft to send data to the Azure cloud. While I was at the booth I watched at least two potential customers walk up, hear the pitch, and then try to order development kits. It was pretty compelling.

So when I got back I spoke with Cedric Mangaud, the CEO of Octonion, to find out what his company is about. He said it was created in 2014 in Switzerland to make it easy to build intelligence into sensors. It began with a demonstration project, for sports, called the PiQ sensor. The hardware is an amalgamation of sensors and software that can handle AI algorithms that could be placed on a variety of objects, ranging from kite-surfing gear to a tennis racquet.

So far, Octonion is managing data from 100,000 PiQ sensors out in the field. These devices generate 1.5 million points of data a day for analysis, says Mangoud. The goal of the PiQ pilot was to show investors and customers that the technology would work, and that it would work at scale. Now it is doing so.

Mangoud says that the Avnet sensor is the next stage of growth for the company. The sports use cases are cool and a good selling point, but they don’t pay Octonion’s full engineering or operations costs. That’s where the industrial internet comes into play. Avnet wanted a way to marry the hardware it resells to some kind of value-added service or differentiator.

When it found Octonion, it found its point of differentiation. Together with STMicro, Microsoft, and Octonion, the firms took the sensor tech and created the Brainium platform. The platform consists of the brick of sensors, software optimized for running AI on the brick, and a cloud-based monitoring and management service. The software also provides the AI used to train and tweak the algorithms used at the edge for a customer’s specific purpose.

So far, the Brainium platform can handle anomaly detection used to detect changes in machines’ states as a crucial component of predictive maintenance. It can currently detect anomalies in vibration and motion, and will soon add sound to its repertoire.

Its business model is a bit different from many other IoT platforms. Customers pay for the brick and a one-time maintenance fee if they want to receive support and updates for the product over time. Most IoT platforms are using monthly SaaS revenue models. Mangoud says the goal is to create a new category of hardware that he calls “meta sensing.” It’s a combination of hardware and software tightly tuned together.

To date, he’s raised €16 million ($18.2 million). Unlike other IoT companies that provide easy-to-use AI, Octonion is strictly focused on running the algorithms at the edge. A company like C3, for example, offers a similar service, but it all takes place in the cloud.

Mangoud and apparently Avnet are hoping this idea of meta sensing at the edge takes off and can help turn sensors — which are currently somewhat of a commodity — into something with a higher value, and a higher margin.