Amazon has introduced a new cloud-based tool, Lookout for Metrics, that turns machine learning-enabled anomaly detection (AD) into Software as a Service.
Relevance
The tool eases deployment and management of cutting-edge algorithms that can learn what “normal” looks like in any data stream and alert users of significant departures from normal
The algorithms behind this technology are particularly helpful in applications where “normal” is hard to characterize, such as in IoT applications
While folks looking to solve AD problems on a specific application or data stream can benefit from this new tool, those looking to solve the problem of “too many metrics” are unlikely to find a solution in Lookout for Metrics
The good news
Lookout for Metrics will allow more organizations to test the feasibility and potential value of machine learning-enabled AD in specific applications
There are many useful applications for AD. For example, Horizon has helped clients design and apply machine learning algorithms to high-frequency data on temperature and voltage to identify departures from “normal” patterns. In these cases, “normal” is difficult to characterize, reflecting unknown patterns of operation of the underlying equipment, equipment and/or battery degradation over time, and anticipated changes (e.g., changes in shift schedules). This is where machine learning is well-positioned to help
What’s more, the tool can be used via key third-party SaaS applications like Salesforce, Servicenow, Zendesk, Marketo, and Amplitude to market, sell, oversee and nurture customer relationships, conduct FP&A, build out products, and more. When it detects anomalies, the tool can deploy alerts through multiple flexible frameworks such as Webhooks and messaging applications
But look before you leap!
Amazon markets this tool as a solution to the “too many metrics” problem. While that undoubtedly constitutes a large addressable market, we don’t think Lookout for Metrics solves the problem. See our other post on this topic.