If you’re moving into cloud, Amazon launched a service on September 25th called AWS Cost Anomaly Detection within AWS Cost Management to find surges in spend. Part of the product is a machine learning algorithm that tracks your spend to ensure that spend peaks aren’t just part of a cyclical spend change and to detect anomalies. One of the interesting aspects of this product to me is the flexibility of monitoring spend based on service, account, category, or tag.
The tagging capability is the most interesting one to me, as tags are how cloud costs are effectively cross-charged to projects, cost centers, geographies, and the other financial categories that are most relevant from an IT expense and financial management perspective. Although the other spend monitoring categories are interesting from a practitioner level and obviously should be used to optimize spend, they will likely be less useful to share with your colleagues.
I’m especially interested in seeing more detail about how machine learning ends up tracking AWS service spend over time to correct its recommendations. One of the interesting aspects of this service is that you actually do not choose your parameters for which anomalies get tracked, as the algorithmic approach picks up every spike. Rather, the service focuses on when it should alert you to changes and anomalies based on the size of the spike. And then you can choose to be alerted in near-real time, daily, or weekly basis.
Given that it’s currently a beta product, I’m betting that the alerts and recommendations aren’t quite fully baked at this point. But even so, this optimization moves cloud towards the state of in-billing period monitoring and optimization that we’re used to doing in wireless and wired spend. Take a look and see how Cost Anomaly Detection starts to shape and optimize your AWS services’ spend.
Of course, this is an AWS-specific service, so there are still opportunities both for other cloud providers to provide similar services as well as for the leading third-party cloud service management providers such as Apptio Cloudability, Cloudcheckr, CloudHealth by VMware, Calero-MDSL, Flexera, Snow Software, Tangoe, and Upland Software to also develop similar capabilities for multi-cloud.
For now, Amalgam Insights recommends taking a look at the documentation and learning how the service works. We are starting to transform IT cost management from a practice of manually tracking cost data on our own to depending on algorithms and machine logic to do the hard number-crunching and swivel-chair work for use. Even if you’re not going back to school to learn the linear algebra, calculus, and neural net designs needed to do data science on your own, you need to have an idea of what can and can’t be done through algorithmic means.