Lesson Learned: Some Services Are Not Worth Delivering

Here’s one of the secrets to AWS’s unprecedented scale and financial success: they quickly figured out that some services are not worth delivering. Most everyone else believes in building snowflake single-customer solutions to solve imaginary problems, effectively losing money while doing so.

You’ll need a Free ipSpace.net Subscription to watch the video.


  1. This works great if you are a SaaS and have a selection of customers. For any internal IT organisation, the world is a lot more complex. We tried not offering certain (legacy) services to our internal customers, which made a lot of sense from financial, architecture and design point of view, as in the past these "snowflake" services put a massive burden on the infra team.

    But in the end the business always wins, no matter how great your pitch is to the C-people. The business has to run. So we built a snowflake again :(

    1. Pushing the risk acceptance up the stack sometimes works - document what risks (and increased costs) you'll be exposed to due to the snowflake service and get a C-level person to sign off on that.

      Until you get the sign off, you cannot proceed ;)

      Alternatively, I was told that compliance (or risk management) team could be of great help if you could properly present the risk of deploying snowflake services to them.

  2. If you're an ISP, you always have the option to increase the cost of their legacy service upon the customers next contract renewal. There are always indirect ways nudging customers off legacy infrastructure.

    After all, if you were still on a frame relay connection, what cost and speed benefit would the customer have these days? Practically none. But I understand cloud providers need a common denominator to deliver services in a more unified fashion.

    It's also another reason why they only allow you to have a couple hundred routes to manage. They do not and cannot allow for past sloppy practices to be reintroduced into their cloud.

Add comment