The Blog

How Does AWS Measure Customer Numbers, and Should It?

Calendar Icon 12.01.2021
aws-section-divider aws-section-divider

Something I noticed in the recent AWS earnings call was how AWS counts customers. Given that they’ve been public about having “millions of customers,” this does naturally lead to some interesting questions. Before we dive into it, I want to be very clear: I don’t think anyone’s being disingenuous here. These are hard problems to solve for, and no metric is going to solve for all cases.

AWS’ customer definition is as follows:

References to AWS customers mean unique AWS customer accounts, which are unique customer account IDs that are eligible to use AWS services. This includes AWS accounts in the AWS free tier. Multiple users accessing AWS services via one account ID are counted as a single account. Customers are considered active when they have had AWS usage activity during the preceding one-month period.

To unpack this a little, that means that an AWS “customer” is intended on some level to correspond to individual AWS accounts (not IAM users, individual credentials or AWS organizations).

That translates to meaning that my own company (The Duckbill Group) with 11 employees counts as roughly 13 AWS customers, whereas I’m something on the order of 40 personally (it’s been a weird and long few years).

Conversely, a client of ours who employs thousands of engineers full time counts as approximately four AWS customers, give or take.

A large company that gives every employee their own “sandbox” AWS accounts would count then as thousands of AWS customers.

And this becomes in turn completely maddening.

What’s the goal here?

Let’s be clear. Since there’s no industry wide accepted definition, there’s no way you’re going to be able to compare apples to oranges in any reasonably effective way. It’s only useful insofar as it demonstrates where you are today versus where you were at some other time, but inherent changes as to how you’re dealing with accounts mean that there are many factors weighing in besides pure growth of your customer base.

These are inherently always going to be vanity metrics that don’t actually matter all that much; they equate all customers being the same, when that’s very far from true.

What other measures could be used?

Some folks have suggested that each IAM (Identity and Access Management) user count; I don’t think that works for a variety of reasons. Does every engineer that a company hires count as a new AWS customer? I sure hope not! Let’s also not forget that it wouldn’t count effectively for single sign-on users, folks who stubbornly insist on using the root AWS account for everything (NO! Do not do this!) and engineers at companies who only interact with AWS resources via a CI/CD system or internal developer portal.

iRobot’s Ben Kehoe suggests a compromise: The count should instead be every AWS organization plus every account that isn’t attached to an organization. That would reduce me to being three customers, which is close to accurate, as I am very large post-pandemic.

I think Ben’s closest to correct when he says that every organization plus unlinked organizations should count — but that doesn’t account for resellers who own the organization’s payer account while managing a variety of independent companies’ AWS accounts.

I’d say that from where I sit, trying to calculate out the intent of the metric, the best answer is to simply give up. There’s no great way for the number to be derived in a way that accounts for all use cases, and it’s too easy for it to be misinterpreted in ways that don’t matter.

At its core, AWS is attempting to demonstrate adoption and growth. There are far better metrics to expose that. Fundamentally it all comes down to revenue and revenue growth.We’re well past the point where one or two large AWS customers constitute the bulk of AWS. Even very large customers are still a small slice of the overall pie.

And to its credit, AWS hasn’t recently (at least, anywhere that I’ve seen!) cited this metric as anything other than “millions of customers,” which I’d absolutely accept once you factor in all of the learner accounts out there in the world. The only time the metric is referred to (aside from defining the term “AWS customers” in its most recent earnings report is in the sentence “tens of thousands of AWS customers are supporting more than 10 million contact center interactions a day on Amazon Connect.” I think it’s fair to say that Amazon Connect has tens of thousands of companies using it for various things; it’s not the sort of service that’s going to turn up by accident in a bunch of test accounts like S3 or data transfer will.

I therefore suggest that this metric isn’t useful to anyone; effectively every company is an AWS customer somewhere, and continuing to focus on this metric will merely serve to chart the growth of the overall IT sector and its customers. More interesting things to be tracking are per-customer revenue, sectors that are actively using AWS, and what percentage of AWS’s incredible growth is net new accounts versus organic expansion of large customers. Unfortunately, it’s hard to see a lot of those details from the outside.

aws-section-divider