Augury

Scheduled maintenance wastes and misses

A pump that burns out in the middle of a shift at a factory stops an entire production line, and the damage is not the price of the part but the hours of downtime. The traditional way of working is scheduled maintenance, replacing components every X hours. This method both replaces parts that are still fine and misses failures that come early. Augury, founded in 2011 by Saar Yoskovitz and Gal Shaul, takes the third approach: listening to the machine.

The sensors: listening to the machine

Its sensors are installed on the machine and continuously pick up vibration, sound, heat and electromagnetic radiation. A model trained on a large database of failures recognizes the acoustic signature of a developing fault, such as a worn bearing, a misaligned shaft or an imbalance, sometimes weeks before a human ear could hear it. The alert also states the type of fault and how urgent it is, so that repairs can be scheduled at a planned time rather than in an emergency.

Business model and clients

The business model is a subscription for each monitored machine, not equipment sales. Reported clients include Colgate-Palmolive, PepsiCo and Nestlé, and in Israel ICL, Tnuva, the Israel Electric Corporation and Mekorot. Its development centers are in Haifa and Tel Aviv. In May 2022 Augury signed an agreement to acquire the Israeli industrial AI company Seebo, which analyzed the production process itself and not just the health of the machines, in a cash-and-stock deal whose size was not officially disclosed.

Funding: from unicorn to Series F

In October 2021 the company raised $180 million at a valuation of more than $1 billion and became a unicorn. In February 2025 it raised another $75 million in a Series F round led by Lightrock, again at a valuation above $1 billion, according to the company's official announcement. "With this new funding we will continue that track record, breaking new ground in the introduction of Agentic-AI capabilities," CEO Saar Yoskovitz said in the announcement. Predictive maintenance is one of the oldest and most economically proven uses of AI in industry: its return is measured in hours of downtime saved, a measure that is simpler to prove than many others in the field.

The trust challenge: false alarms

The real difficulty in this field is trust. A false alarm that leads to stopping a healthy machine costs money just like a failure that was not predicted, so success is measured not only by the share of failures detected but also by the share of correct alerts among all alerts. That is why systems of this kind are usually rolled out gradually, on a small group of machines, until the team on the floor is convinced.

The competitive advantage: data that builds up over time

The data gathered over years is also the main competitive advantage: the acoustic signature of a rare failure can only be learned if you have seen it enough times, and that is hard to achieve without a broad customer base monitoring similar machines in different factories. So the advantage in this field builds up over time and cannot be bought all at once.