When Amazon Kendra was announced last year at re:Invent, I was initially excited. It solved a problem that had vexed me for ages: How do I search all of the various places where data winds up to surface the thing that I care about? Maybe it’s in Confluence, maybe it’s in Asana, maybe it’s in a Slack message, maybe it’s in my Twitter DMs, or maybe it’s in the least searchable thing of all, which is somehow Google Docs.

The lost opportunity of this service is that neither I nor companies similar to mine are going to find out any time soon.

Maybe Kendra can solve my search problem, maybe it can’t. But its unsuitability for my use case makes itself clear in the branding: “Amazon Kendra: Highly accurate and easy to use enterprise search service powered by machine learning.” 

The giveaway is in the use of two terms: *enterprise* and *machine learning*, which is a company’s not particularly subtle way of saying “bring money!” 

Sure enough, the pricing on this service is (in typical AWS fashion) incredibly hard to parse. But the Developer Edition starts at $2.50 an hour with some serious constraints, and the more durable Enterprise Edition costs $7 an hour. 

This means that, as a responsible user, you’re staring at a system that starts at $60K a year before you factor in all of the other weird billing dimensions (e.g., “$0.35 per hour per connector when syncing” means that whatever you think you’re going to pay, you’re going to guess wrong). 

For the target customer that this is aimed at, the price is immaterial compared to the value it delivers. They’re probably spending a couple of million or so a year on AWS services already, so Kendra solves an expensive problem while its cost disappears into the larger noise of the plethora of AWS services they’re already using.

I can’t help but view this as a lost opportunity.

Enterprises aren’t born enterprisey

Big companies are small companies that struck it lucky and happened to grow. In other words, tomorrow’s large enterprises are today’s scrappy startups. 

AWS knows this; it’s why they offer multiple onramps to virtually every AWS service. It’s why they shower startups with uncomfortable amounts of free credits under the auspices of the Activate program. And it’s why, with remarkably few exceptions, you can get started with virtually any AWS service without a lot of up front investment.

Kendra, however, is one of the exceptions. 

What its model dictates is that there’s no “try it out for a bit and see if it meets your needs” other than the laughable 30-day free tier. This, in turn, means that Kendra will see adoption only in companies that fit a very specific profile—companies that have somehow managed to scale to a significant size without solving for the “where does our internal data live and how do we search through it” problem. 

Meanwhile, the smaller companies that are growing will see Kendra’s price point and deduce correctly that the Kendra target market is Not Them, and they’ll find another way to solve this problem.

The trouble here for Amazon is that when that small company grows, whatever other solution they find isn’t going to be ripped out and replaced just as soon as Kendra’s scale makes sense for them. Those alternate solutions will be made to scale, and now Kendra’s found itself forced into a “bake-off” scenario wherein they’re now competing with everything else in this space and an entrenched incumbent—when they could have had won the business years earlier by offering a far more straightforward onramp for small users.

Why leave out small customers intentionally?

The unstated goal of every AWS service is universal adoption. 

Most services have onramps that cost pennies and let customers experiment with them as they grow. You don’t grow to “millions and millions of customers” as AWS claims without having myriad ways for customers to leverage your offerings; not every customer starts with EC2, then plays with S3, then discovers RDS / DynamoDB / Route 53. 

There are many paths to AWS—and some lie squarely through corporate IT, struggling with data management problems that Kendra is very well-positioned to solve. 

Kendra’s lost opportunity

The problem I have with Kendra isn’t its pricing—although I am disappointed that I’m not the target market for something I’d love to use. 

My problem instead is that the use case for Kendra is defined down to a niche that only encompasses a small number of the possible adoption cases. In other words, the target market for Kendra is definitionally going to be “existing AWS customers above a certain size currently grappling with an information management problem.” 

That’s a fine target market, to be sure. But Kendra is interesting enough that it could be so much more than that; it could theoretically be a differentiated service that becomes the first one a company implemented.

This is a lost opportunity then—not to charge less but rather to experiment with alternative structures that let people “try before they buy” and reach a corporate IT buyer that historically may not ever have considered moving workloads to AWS. 

Kendra going forward

If the service works as promised, Kendra has the potential to be something transformative to enterprise search. We may very well see a future where it moves downmarket as the economics of the system change. 

I just hope it doesn’t become a niche service, relegated only to large existing AWS customers who just so happen to fit the customer profile. That would be a colossal disappointment.