New to WIP?

We're a group of makers shipping together. We help each other stay accountable and reach our goals.

Apply for access
Topics
🤖 Artificial Intelligence
Other questions
What steps are included in your CI pipeline you should have added sooner? 11 answers
AI to work with a lot of data (two usecases) 8 answers
Customer Support SaaS with unlimited brands support 2 answers
Does investing time and effort in social media marketing still make sense? 9 answers
Back
 Question
Asked

Thoughts on this pricing for this AI-powered SEO tool?

Hey everyone,

The tool I'm building helps companies/agencies improve their SERP rankings. I'm wondering what the best way to price an AI SEO software like this would be:
  • User brings their own OpenAI API keys (Bring Your Own Key)
  • My software would do bulk API calls upfront (anywhere from 100-1,000 API calls). Each call costs ~$1.30
  • There is a monthly hosting aspect (my tool improves their existing blog content via SEO and also makes loading them 5x faster)

During beta, plans will be free so I can learn more, but initially I'm thinking on pricing based on:

  • Number of blogs (websites)
  • Number of blog articles
  • Number of blog pageviews
  • Number of team members/clients

The biggest issue I'm having here is determining whether a BYOK model is the best here. I've chatted with AI guys like the CTO of JenniAI and removing the API key aspect is super helpful b/c it simplifies the product. Also, the bulk aspect here is the entire selling point of the product, b/c it minimizes the amount of work for users.

Kevin Yun
@kevinyun

👋 Join WIP to participate

Luca
@lvca

Hmm ok makes sense. The only question is how the heck do you have a $1.30 cost per request? Are you generating 50k gpt4 characters or something?
With costs like that I'm pretty unsure about what the best pricing would be ngl...

 Reply
Kevin Yun
@kevinyun

Estimated max input/output as worst case scenario (still gotta test more), but will prob optimize eventually after launching. It is all new GPT-4 calls.

Yeah the bulk upfront model is tough. Not doing bulk may be worse for users but more sensible as SaaS

 Reply
Luca
@lvca

To me, having a cost per user of $130 MINIMUM means you gotta do a ton of optimizations to be profitable... That is while still being able to charge a monthly sub that isn't outrageous compared to other AI blog tools...

The idea sounds cool, but I still haven't rly understood how it works...

Mind DMing me on telegram? Think I need to see/test how it works a bit before giving you a better answer :)

Will give you my API key for eventual testing.

 Reply

BYOK seems nice when targeting developers and/or dealing with potential abusive content and you want to put the risk back to the user.

In most other cases, including yours, I wouldn't do BYOK. If you do, not only does it add complexity to your codebase and lock you into using OpenAI going forward, it also makes it harder to price your product because customers will "feel" like most of the value is coming from OpenAI anyway.

In terms of pricing, it really depends on the value you're providing and the alternatives customers are replacing with your product.

At first glance, number of blog articles seems the closest linked to the value you provide. Sites with more articles, will likely get more value, etc.

Pageviews are hard to measure.

Team members/clients might make sense as an additional pricing knob. For example, you can have fixed monthly plans with different team limits, in addition to a variable pricing for the number of articles.

I recommend looking at analogous services like newsletters, etc that also deal with variable pricing (subscribers) and fixed pricing (features)

 Reply
Ben Katz
@ben

I considered BYO OpenAI key for rmflags.com as well and decided against it for the reasons you mentioned. Ultimately if the customer has to sign up for and manage their own OpenAI credits it just adds friction to the entire signup process.

However for simpleotp.com I did force users to supply their own SMTP credentials. I can’t risk ruining my email sending reputation because one user decides to send emails from a bad domain for example

For pricing, I thought about this for a long time and I think I’d rather charge according to the value the customer gets out of the solution instead of considering AI cost in too much detail upfront. I’m not saying I wouldn’t look at the AI cost, but I like to start with “okay, customer would otherwise have to pay an engineer $200/hour to do this task, so I’ll charge close to that” - and if AI spend is > $200/h, work backwards and figure out how to reduce the cost from there, or if not possible maybe it’s a bad business idea.

 Reply