Hacker News new | past | comments | ask | show | jobs | submit login
[dupe] Show HN: I built a service to help companies reduce AWS spend by 50%
71 points by kavehkhorram 6 days ago | hide | past | favorite | 49 comments
Hey HN: Kaveh here, the founder of https://www.usage.ai/

We help companies drive down AWS EC2 spend. Why? Because the way it's done now is a pain. DevOps engineers end up becoming cloud accountants trying to figure out what commitments are expiring soon and how much they're saving.

Previous to founding Usage, I worked on high-performance computing research at JP Morgan Chase and as a software engineer at a number of medium-sized startups.

Here's how it works: We are typically brought in by a DevOps manager to cut AWS EC2 costs. The app is entirely self-service and the savings are generated automatically, typically we do this live on a call. On average, we reduce AWS EC2 spend by 57% for 5 minutes of work.

To reduce by ~57%, we don't touch the instances, require any code change, or change the performance of your instances. We buy Reserved Instances on your behalf (a billing layer change only) and bundle them with guaranteed buyback. So you get the steep 57% savings of 3-year no-upfront RIs with none of the commitment (you can sell them back to us anytime after 30 days).

We make money off a 20% Savings Fee. Happy to chat directly kaveh@usage.ai

Have you experienced any issues with managing your company or organization's AWS expenses? We'd love to hear your feedback and ideas!

Three Show HNs of the same project in 3 months is excessive.

Show HN: Usage, Cut your AWS Bill by 50%+ in 5 Minutes - https://news.ycombinator.com/item?id=31015171 - April 2022 (17 comments)

Show HN: I built a service to help companies reduce AWS spend by 50% - https://news.ycombinator.com/item?id=30183465 - Feb 2022 (60 comments)

The rule on HN is that if a story has had significant attention in the last year or so, we mark reposts as dupes (https://news.ycombinator.com/newsfaq.html). For Show HNs it's sometimes ok to post a bit more frequently as long as the Show HN focuses on new work since the last thread. Not this frequently, though.

Please see also this rule from https://news.ycombinator.com/newsguidelines.html:

Please don't use HN primarily for promotion. It's ok to post your own stuff occasionally, but the primary use of the site should be for curiosity.

Thanks for keeping an eye on things dang.

Interesting business model. You're essentially loaning companies money for the RIs, right? So if all your customers sold them back to you you'd be in trouble, like a loan company?

Either way, an interesting idea. Since you're fronting the money for the RIs, I assume you have a max spend that you won't go over (like, you won't take Netflix as a customer)?

It's closer to a leasing arrangement than a loan. They buy the RI with a traditional loan, then finance that by leasing it out at a higher price.

Actually we aren't loaning any money since we use no-upfront RIs only.

We have some customers that spend $10M+ on AWS, so we can work with any size of company.

Right but if someone sells you back their RIs, aren't you on the hook for paying for them if you can't resell them? So that's like a loan.

If we can't find a buyer then yes, we buy them back until we find a buyer.

To limit this risk, we only offer buy-back on particular instance families and regions that are most popular (and easy for us to sell to other users).

With 100s of users -- we have network effects that make it easy for us to sell.

Do you have any interesting problems lined up that you would consider solving if you found yourself with a bunch of reserved instances that you're unable to sell?

Not at the moment. Any suggestions? :-)

Use the compute to solve your own allocation problems using mathematical optimization.

Neat idea. Why gate it to companies that already have existing AWS setups though? You are basically a secondary EC2 marketplace. Couldn't people just get an instance from you directly? I'm sure lots of people would love to get reserved instance pricing without the commitment. And in that case what would you price it at (since there is no comparison point for savings)?

The cloud reseller motion is something we briefly explored but not currently on the roadmap.

Our roadmap for the next 2 Quarters is Full Multi Cloud (Azure, GCP, and AWS) support.

I assume it's because AWS doesn't let you buy no-upfront RIs without an established billing history.

(Since a recent conversation on HN) I think this is what e.g. Heroku does. Repackage AWS products that they get at a crazy discount and tweak pricing to maximise profit.

> So you get the steep 57% savings of 3-year no-upfront RIs with none of the commitment (you can sell them back to us anytime after 30 days) … We make money off a 20% Savings Fee

Isn’t this business model fundamentally flawed? And not profitable (unless a customer uses you for > 29 months)

Because ONLY companies who plan to use your service for less than 29 months (the break even of a 3 year contract where you keep your 20% cut), should be using your service otherwise it’s just cheaper for them to call up Amazon and get their own 3year RI contract.

Meaning, you’re always going to be in a situation where a customer is going to sell back their RIs and you’ll become unprofitable.

Said differently, you’ll only be profitable if a customer uses you for more than 29 months.

I'd assume they have volume discounts or negotiated rates with Amazon that bring that point forward.

There's an entire sub industry of billing partners who are profitable enough just offering eg 8% off retail prices.

In my experience Amazon can be difficult to negotiate with directly if you're not willing to commit to YoY spend growth. It's easier to deal with billing partners.

Why orgs wouldn't go to Amazon directly via retail is the 3 year commitment that comes with RIs.

20% over the discounts is a bit steep imo but I'm sure there's a market for RIs as a service (particularly if the RI pricing they charge is the 100% upfront covered by their own internal funds)

Oops, I need to learn to read more closely -- the RIs are no upfront so it's a slightly less good deal.

Not totally sure what you mean by we are only used by companies who plan on using our service for less than 29 months. Can you elaborate?

We sell the RIs to our other users (now in the hundreds) so we don't take a loss. We only take a loss when we aren't able to sell your RIs (which is what we use some of our Venture Capital money for).

Looks really cool! The verification email went to spam – I used my Gmail email. Also, I refreshed the page when I finished setting up and the "Generating Recommendations" message disappeared. I can't tell if it finished and didn't come up with anything, or if it's still crunching the numbers.

How are the payments handled? What happens if I change the sizing of my instances?

If you don't see anything it likely means there are no recommendations at the moment. We will make it more clear on the app when that happens.

Very interesting idea - I'd be wary of this rubbing the wrong way against AWS's TOS though.

I only recently heard of this idea at the AWS Summit when talking to a rep of one of your competitors (Zesty), they have a minimum spend of $7k. Your pricing page says there's no conditions, so I assume you don't have minimum spend requirement, is that correct?

No conditions. We are fully automated, we have customers saving as little as $10k and some as many as $1M+ per year.

Any plans to expand this to other AWS services in the near future? At least the popular ones like RDS and Elasticache?

Also, what happens if an organization is currently using Savings Plans and is looking to pivot towards your offering?

We have something launching soon for RDS!

For a customer with an SP, we typically get them boosted to 100% coverage with our Flex RIs.

Once the SP expires, we automatically switch the customer over to 100% Flex RI to maximize savings.

Flex RI = 3-year no-upfront Standard Reserved Instances with Guaranteed Buyback

Why do you use RIs and not sign orgs up for savings plans?

I know RIs are still supported though I'd heard they're intended to be deprecated by SPs (my source could be wrong)

Probably because you can't resale the savings plan.

It is nice. I do this manually for companies (short contracts); great if I can do more automation and less manual work as often it is repetitive.

> We make money off a 20% Savings Fee

does that mean if we save 50%, you keep 20%?

say our bill is 10k and you save 5k, how much do we pay you?

You would pay us $1k (20% of $5k)

cool, is this against any terms of AWS? Seems at the very least it'd be a gray area.

We have a very strong positive relationship with AWS and meet with them often, both with senior leadership and account managers.

You're not really answering the GP's question.

No. We aren't in violation of any of AWS's terms.

The transfer of RIs are done through AWS APIs.

AWS doesn't care, they make the same money off of reserved instances as they do on on-demand instances as they do on spot instances. They know exactly what instances cost to provide under their various pricing plans and that's what they set the price to.

Im uneducated in this space, whats grey about it? isn't this effectively what heroku and other services do? They put a layer on top of AWS, up charge a bit and make profit that way?

terms for something like AWS have all sorts of restrictions. Just thought there could be something in there against transferring instances.

Can your software also implement hard budget limits for users/groups?

Cloud formation dissipation

Why stop at 50%? ;)

Because this is largely a trivial usage calculation and long-term contract automation.

57% is the common savings number by moving to a 3 year commitment on AWS. Many people are hesitant to commit to 3 year contracts, it's hard to predict usage that far. Consider we may be going into a downturn and you are now legally required to pay for today's usage. Seems like OP is taking on some (all?) of that risk? I for one would not want to be the one holding those bags today, given the near-term economic outlook.

To get towards the 80% figures requires a deeper understanding of a company's usage, processes, and code. These need code changes

Usage does take on all the risk. The good thing about cloud compute is that it's recession-proof -- compute volume isn't something companies can cut.

You may want to reconsider this assumption. The cloud companies may be relatively "recession proof" because more companies are migrating to the cloud.

However, their customers are not. I've helped several companies reach 80%+ reductions in compute/storage needs, which then translates into similar cost savings. No reservations needed, you have to provide deeper value and changes to achieve this.

> Usage does take on all the risk.

Seems like a great way to get 3-year commitment prices without the risks. Are you worried about assuming all that risk? What about abuse of this relationship by your customers?

This assumes all compute was crucial for business operation. In practice however, a lot of compute volume just serves expendable reasons such as for example marketing people trying to data-mine interesting leads out of terabytes of sales data. Stuff you do, as a company, while you're flush with cash and thus ready to spend some on potential opportunities. These things however are the first projects to be cut when times are getting rough and budgets must be tailored down to survive.

Sure they can: if you lose half your users because of economic downturn, you don’t need half your compute.

There is also the RI marketplace, they can probably sell overcapacity there even if their direct customers don't have any need for additional capacity.

Or you could change your tech reduce your compute need by 10X+

There are plenty of blog posts around about that for X->Y

And if a company folds, it doesn't need any compute at all.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact