Reserved instances can help save 70% or higher on your instance costs (versus using on-demand instances). You have the flexibility to pay all, part or nothing as upfront fee, and they are available for 1 year or 3 years' duration at much lower hourly rates. You can also modify or sell your reserved instances, if your requirements change, subsequently. Typical breakeven on these instances vary between 5 months and 7 months depending on the duration of the contract.

Typically, you would run your systems for 3 to 4 months to understand your workloads, and tune and optimize your instances types, OS, tenancy, and so on before switching to Reserved Instances. As soon as you know your instance type and size, try to swap your on-demand instances with reserved instance.

Reserved instances are flexible, for example, they can be moved between AZs, and their sizes can be modified. In fact, you don’t have to choose exactly where you want to run your reserved instances—this essentially decouples capacity reservation from cost optimization. You can also have convertible reserved instances (available for 3 years only) that allow you to change instance families/type to leverage the latest offerings from AWS. The effective rate of 3-year convertible reserved instances is better than 1-year standard reserved instances. So, even if you run them for a year and a half to two years, it is fine.

As production instances are typically required to run 24x7x365 in a reliable manner, Reserved Instances are a good fit for enterprise applications (in production). For dev/test environments (and in startups), you might want to experiment with and spend more time evaluating spot instances because spot prices can be a fraction of the regular on-demand prices.