We’re decreasing the price of metrics by 200x

I screwed up — and I want to apologize.

A bit of background, first:

The team here at Better Stack built a state-of-the-art data processing pipeline. Storing logs in object storage while keeping derived metrics on local NVMe SSD drives brings the best of both worlds:

• low-latency dashboards at a massive scale

• and extremely fast high-throughput log search with horizontally scalable ingestion.

So when it came time to start charging for metrics in the summer of 2024, I looked for inspiration across our competitors.

I knew we had superior tech and could charge less than our competitors. But how much less should we charge?

I was blown away by the lack of transparency when it came to pricing metrics

Imagine you have a small app generating 200 GB of uncompressed logs per month along with derived metrics on the side: how much will the datadogs of the world charge you for metrics? I dare you to try to figure that out from Datadog pricing page.

"Is it a standard metric? Or a custom metric. Or a custom event? Ok, so how many custom metrics do I have when I exclude the included hosted metrics… (5mins later) Alright, I give up"

Oh, I see — that’s how Datadog makes $1B per year from metrics alone:

Aiming to be transparent, we designed a simple formula for charging for metrics based on the idea that more data means higher cost:

Metrics data points = active time series × aggregations × retention in days

To set the price per data point right, I spun a small app and configured the Datadog agent to send data to Better Stack to measure how much metrics volume it produced and priced it to be 10x less expensive than Datadog.

Turns out 10x cheaper than Datadog is still super expensive 🤦🏻‍♂️

In the meantime, I asked our team at Better Stack to reach out to existing customers who have been using metrics in ‘beta’, while they were completely free.

And I open Twitter. Boom.

Seeing tweets like Bruno’s triggered my inner engineer who wanted to argue: “Just optimize your metrics cardinality and the cost will drop by 100x!”

But our customers shouldn’t have to be experts in cardinality reduction —

I bet most of you have no idea what I’m talking about in the first place. It’s my responsibility to design a product that makes it easy to control costs. And to make it cheap for basic use cases by default.

Bruno, I’m sorry for the emotional roller-coaster of getting a surprise $35,000/month bill. That’s just not right and it was never my intention.

Going forward, we’ll be adjusting the price for 1 billion of retained metrics data points from $1,000 to $5. That’s 200x cheaper.

There’s no good reason to self-host Prometheus anymore.

And yes, that’s a lot cheaper than Datadog.

The industry needs more visibility into usage & pricing

We still have a long way to go, but here are things we’re doing to make sure our customers don’t get surprise bills.

Granular day-by-day usage reporting in a single place

No need to guess how many metrics you’re generating anymore. Simply send us 1 day of data and we’ll extrapolate the estimated usage for you. All your usage in a single screen. Available in Settings → Usage.

Recommendations on reducing the cardinality of your metrics

Go to Sources → Your source → Metrics to see the cardinality breakdown per column for all your metrics.

Consider generalizing your high-cardinality metrics to save on costs.

Simple to configure billing alerts

Get alerted when your usage spikes. Configure a billing alert in Settings → Billing → Advanced settings.

Set a spending limit

Configure a maximum dollar amount you're willing to spend per month. Please be careful: we will completely stop processing your data once you hit the limit. Soon to be available in Settings → Billing → Advanced settings.

Single straightforward pricing page

No need for 30 pricing sub-pages. All of Better Stack pricing is explained in a single page, made for humans:

And we have a lot more in the works. Stay tuned!

Let’s build! 💪

Juraj

Got an article suggestion? Let us know

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.