@cornbreadfed614 Our experience (DevOps B2C/B, PLG) has been the same as Ellipsis', most people asking for usage-based pricing are hobbyists. It never comes up with enterprise or prosumer. If anything, they explicitly ask for fixed pricing.
I interviewed 3 YC founders on pricing AI products (they all hate usage-based)
- Thread starter cornbreadfed614
- Start date
-
- Tags
- startup y combinator
nickolasss
New member
@cornbreadfed614 Yep I find the usage based model to be unrelatable and on my new startup (building prototype atm) I'm pricing based on outcome rather than usage.
The only time you price per usage is if you sell API access.
The only time you price per usage is if you sell API access.
cornbreadfed614
New member
@nickolasss Yes I agree. I think the closer you are to selling "raw materials" that others use to build, the closer you should get to usage-based pricing
summersunshine1217
New member
@cornbreadfed614 What do you think about a "bring your own keys" model, where you have a super cheap tier that's commensurate with your product's added value? The customer pays by usage with AI services directly.
@cornbreadfed614 Ugh, I had a cofounder who was very adamant that the only way to align our interest with the customers' is to price based on usage. I mentioned that our buyers want certainty for approval and he didn't believe it.
@cornbreadfed614 100% we found the same thing
sunshine1973
New member
@cornbreadfed614 Pricing's tough, right? What's your take on no-code AI integration?
silent_wife
New member
@cornbreadfed614 The point about businesses preferring to know what their cost is going to be upfront is pretty big. Every dept has a defined budget and It’s difficult to police how much your team is using a product and. Companies, both sellers and buyers, need some degree of predictability in their finances.
I do think usage based is potentially more feasible with b2c, although consumers are fairly unfamiliar with that model so they may be skeptical at first. I bet price sensitive or frugal customers would fuck with it though.
I do think usage based is potentially more feasible with b2c, although consumers are fairly unfamiliar with that model so they may be skeptical at first. I bet price sensitive or frugal customers would fuck with it though.
@cornbreadfed614 Can you share how you arranged call with them, whenver I try to research the market people are not so cooperative.
Thanks for the content, its gold for us
Thanks for the content, its gold for us
@cornbreadfed614 What about different subscription levels based on usage?
@cornbreadfed614 This is definitely not wrong, the best thing you want as a startup is like a set revenue from each customer because you need reliable cash flow; and frankly I don't subscribe to AI services because per use would essentially like mount up and cost me a lot more if I'm not careful; I lose the chance to mess around and find out more.
pamelalove
New member
@cornbreadfed614 so they’re fine usage-based cloud services but not with usage-based ai
cornbreadfed614
New member
@pamelalove The point is more that if you sell something others build with (APIs, cloud services), usage-based makes more sense.
It's more about the fact that if you (like most people) sell a solution that users use directly, burdening them with usage-based pricing is annoying and most customers prefer to buy with more conventional pricing.
It's more about the fact that if you (like most people) sell a solution that users use directly, burdening them with usage-based pricing is annoying and most customers prefer to buy with more conventional pricing.
pamelalove
New member
@cornbreadfed614 i agree but unless you’re vc-backed and have tons of money you can’t provide fixed-price subscriptions without setting usage limits. i hear you i’m not a fan of usage-based neither but it still makes the most sense when the service is all about running on a server.
@pamelalove A big part of it is the unpredictability.
If it's $x per gigabyte and you store y gigabytes you know your costs are $x*y. Same with API calls, $x per API call and you make y calls.
But with LLMs you have no idea exactly how many tokens its going to spit out or if its going to loop for how many messages.
If it's $x per gigabyte and you store y gigabytes you know your costs are $x*y. Same with API calls, $x per API call and you make y calls.
But with LLMs you have no idea exactly how many tokens its going to spit out or if its going to loop for how many messages.
pamelalove
New member
@pamelalove Honestly not 100% sure.
Taking some clues from OpenAI themselves: you can set a monthly spending limit. This gives some comfort that you aren't going to be billed a gazillion dollars at the end of the month.
Obviously cutting off service completey isn't going to work for a lot of customers, so maybe you could start reducing price per token as the user uses more per month. So it starts curving downwards towards cost per token. The user doesn't get a massive invoice, they can still use your service and you keep making at least some money. Far from perfect though. Dunno how much I like it.
TBH... IDK. As OP noted, entirely new pricing models are going to emerge. I'd imagine once the tech matures and its no longer so computationally expensive & electricity-hungry, things will lighten up.
Taking some clues from OpenAI themselves: you can set a monthly spending limit. This gives some comfort that you aren't going to be billed a gazillion dollars at the end of the month.
Obviously cutting off service completey isn't going to work for a lot of customers, so maybe you could start reducing price per token as the user uses more per month. So it starts curving downwards towards cost per token. The user doesn't get a massive invoice, they can still use your service and you keep making at least some money. Far from perfect though. Dunno how much I like it.
TBH... IDK. As OP noted, entirely new pricing models are going to emerge. I'd imagine once the tech matures and its no longer so computationally expensive & electricity-hungry, things will lighten up.
@pamelalove You translate tokens to a metric that users can understand. For example, if you're a document processing company, figure out how many tokens per page and charge $x per page.
@pamelalove A lot of people are not fine with usage based cloud services. Plenty of a large organizations are going towards on-prem repatriation, where they know what a specific workload is going to be like, they'd rather just have it on-prem so they have better control of the costs. Really depends on the bells and whistles and additional functionality, security, and difficulty of implementation and ongoing support for on-prem versus SaaS.
pamelalove
New member
@ironicall but i don’t get how one would do it otherwise without setting usage limits. i mean one can always set their own server and even run their fav llm model locally, but i think this will cost way more.