Hey there! It’s been a bit since I’ve posted here so allow me to reintroduce myself. My name is Niraj Naik and I’m a VP on our Product team overseeing our core subscription and billing solutions.  I’ve been shipping product at Recurly for almost 5 years now and have built a bit of a reputation at the company as a “dad joke” teller (hence the title of this post).  

Intros aside, let’s discuss how usage-based billing has emerged as a particularly effective approach for AI companies. Specifically, AI companies can use this billing method to align costs with actual resource consumption while providing customers with flexibility and transparency. Ready to break it down?  Keep on reading and don't worry, I'll keep it light. After all, good billing practices and dad jokes have something in common: timing is everything!

What's the deal with usage-based billing?

Usage-based billing is pretty much what it sounds like — you pay for what you use. It's like a buffet where you're charged by the ounce instead of the flat "all-you-can-eat" price. 

For AI companies, this might mean charging based on:

  • How many times someone calls your API

  • How much processing time they use

  • How many tokens or characters get crunched

  • Which AI models they access

  • Storage space they hog

  • Computing power they burn through

This approach offers several advantages over flat-rate subscription models, especially for services with variable usage patterns or high operational costs.

The main flavors of usage-based billing

Tiered pricing

Tiered pricing structures divide usage into predefined ranges, with each tier having its own rate. As customers reach higher tiers, they pay the rate associated with each tier for the usage that falls within that tier.

  • First 10,000 API calls: $0.05 each

  • Next 40,000 calls: $0.04 each

  • Beyond that: $0.03 each

This model rewards higher-volume customers while maintaining profitability across all usage levels.

Stairstep pricing

With stairstep pricing, the price per unit changes for all usage once a threshold is crossed. This creates distinct "steps" in pricing.

For example:

  • Use up to 10,000 tokens: $0.05 per token

  • Use 10,001-50,000 tokens: $0.04 per token for ALL tokens

  • Use over 50,000 tokens: $0.03 per token for ALL tokens

This model can incentivize customers to increase usage to reach more favorable pricing tiers.

Volume pricing

Volume pricing applies a single rate to all usage based on the total volume consumed. As usage increases, the rate applied to the entire usage amount decreases.

For instance:

  • $0.05 per compute minute if total usage is 0-1,000 minutes

  • $0.04 per compute minute if total usage is 1,001-5,000 minutes

  • $0.03 per compute minute if total usage exceeds 5,000 minutes

This model strongly encourages increased usage by offering the best rates to high-volume customers.

Making it work without giving yourself a headache

Track everything

You need solid systems to count what your customers are using — whether that's API calls, tokens, or compute time. This needs to be reliable, because nobody wants billing surprises. Those are even less popular than dad jokes at the dinner table!

Keep it simple, smarty

Sure, you could create a pricing model so complex it would make quantum physics look like kindergarten math. But maybe don't? Your customers will thank you.

Give them:

  • Simple calculators

  • Usage dashboards

  • Alerts before they rack up huge bills

  • Clear breakdowns of what they're paying for

Dealing with unexpected usage spikes

AI service usage can be unpredictable, with sudden spikes in demand. To protect yourself and your customers, consider:

  • Setting usage caps

  • Creating throttling options

  • Building in some forgiveness for legitimate spikes

Real talk: Companies doing this right

Tons of AI companies are killing it with usage-based billing:

  1. Those big language model folks charging by the token

  2. Image recognition APIs billing by the picture

  3. Development platforms with pricing that scales as you grow

Wrapping it up

Usage-based billing lets AI companies charge fairly while giving customers flexibility. It's a win-win, kind of like when dad agrees to grill but mom makes the marinade.

If you're looking to set this up, check out Recurly’s solutions for usage-based billing and tiered pricing that handle all the complex metering and billing stuff, so you can focus on building cool AI.

Remember: the only thing worse than poor billing practices is a dad who thinks he's a comedian. Unfortunately, I've given you both today!