AI Monetization Approaches

Everyone’s talking about AI pricing models. Few have actually priced an AI product that is being sold/paid for. When we launched our Copilot product, we talked to a ton of customers about pricing. Here are 4 ways to price AI features/products that came up and we considered: 1. Success-based: Great in theory, sucks in practice. Sarah Tavel has this awesome framework: “sell the work, not the software”. The problem in many software markets bulging with AI right now: turns out the “work” is really hard to define. If your support chatbot charges by resolution, what does a resolution actually mean? If a user is so frustrated with AI answers that they close the chat, they didn’t write a ticket, so is that a resolution? But they also didn’t solve their problem. We hear so many complaints about products that are priced this way. 2. Usage-based: Charge a margin on top of how much you pay for software. This is a very good way to avoid losing money on a contract and also a very good way to not make money. It’s really hard to convince customers to pay a large margin for cost-plus products, mainly because people don’t like paying you a lot of money when it’s clear exactly how much profit you’re making on top of COGS. The other, arguably bigger problem with this approach, is that AI API volume doesn’t always correlate to value. Usage of your AI chatbot could 10x, but that could all come from low-value interactions because your chatbot started behaving poorly and users had to rephrase their questions a lot. Don’t make your customer think about this trade-off. 3. Bring your own keys: Customers paste their own API keys and foot the AI bill based on how much they use it. Our customers were almost unanimous: We don’t want to manage our own keys. They want outcomes: not outcomes that can be achieved if they do a little bit of the work themselves because AI. 4. Don’t break it out: Customers don’t pay for software, they pay for value (shh… they always have). That’s why we price Copilot like we’d price any product: standard SaaS contract (in our case, that means metered by MAUs) that doesn’t directly touch on how much that customer costs us in AI fees. Yes, this means we could in theory lose money on a contract if some of our customers users spam our AI features. But with some rate limiting and paying attention, it’s easy to flag suspicious patterns and investigate. Unless you are grossly negligent I can’t imagine losing money on a contract this way. The massive benefit of this approach is that your customers “just use” your software. They don’t have to think about whether each use case is “worth it” or quibble with your definition of “success”. They just agree to a fee upfront that we justify to them based on what the software is going to do for them and their users, and then next year they can evaluate if that fee was worth it, like normal software. (For some great research on this that inspired us, check out Kyle Poyar post on AI pricing!)

The AI craze will implode faster than #Threads (🫢) if folks can't figure out how to make money from AI. With all of the AI buzz, you might not have heard that ChatGPT's traffic actually dropped by 9.7% in June according to Similarweb. Time spent per user was down by 8.5%, too. So, how do you monetize AI? Here's what I'm seeing from the early adopters. 👉 The majority are in 'wait and see' mode. They want to put AI in the hands of customers, see where it creates value, and then monetize later. In many cases the play is to embed AI capabilities into customers' existing workflows & with their own data -- which makes AI both stickier and more valuable. I wonder: how will folks actually measure value creation and willingness-to-pay? In the words of Madhavan Ramanujam, knowing how you'll monetize is >>> than hoping you can monetize. 👉 Free plans are universal (#PLG). AI has the potential to deliver not only fast time-to-value, but near instant time to value. It can feel like magic for the end-user. Why not put that magic in the hands of more people? 👉 I'm fascinated by early adopters like Intercom who are testing disruptive pricing models. Intercom is charging $0.99 for every successful AI resolution of a customer support case. That's usage-based pricing tied to real customer outcomes and business value. If customers accept that model (which is a big question), it has the makings of a compelling business model with built-in expansion. 👉 There's a trend toward complex & hybrid pricing models. Many folks have some concept of a usage paywall (tied to # of prompts, queries, words, responses, etc.) while still monetizing based on user seats or feature-based packages. I suspect this approach aligned with how people were already pricing their products & made pricing predictable for customers. --- I unpacked this topic with the one-and-only CJ Gustafson in his latest Mostly Metrics newsletter. Check it out, then chime in: will AI products make money? If so, how?! #ai #pricing #monetization #product

Let’s talk about the unit economics of AI usage, and who foots the bill. AI cost structures can be complicated and costly. From API calls to models from OpenAI or Anthropic, or the costs of hosting and deploying open-source models, or charges associated with fine-tuning and training models: COGS add up, and gross margins can take a hit, especially if you don’t charge based on usage. Here’s the twist: we’re seeing highly-leveraged tools like Notion AI stick with per-seat pricing instead, meaning every customer, regardless of usage, pays the same monthly price. At Orb we think experimenting with pricing is core to your company’s success - not a one-and-done deal. But, if you’re just starting to build in AI, you’re not wrong to keep pricing straightforward - for now. Here’s how you should think about monetization while you’re in early learning mode: 1. Learn how much AI costs you: know your cost structures like the back of your hand, including an understanding of your OpenAI bill - know this even if you aren’t yet charging for AI capabilities. 2. Experiment and learn: companies often wait until it’s way too painful and costly to experiment with their monetization - build a cadence and culture of pricing experimentation into your product roadmap. 3. Test out various AI pricing strategies: explore willingness to pay, think through the fundamental unit of value for your customers, and think through the needs of each customer segment. AI pricing simply comes down to pricing fundamentals and thinking from first principles to arrive at what's best for your customers. And while cost structures and gross margin profiles will always be in flux, the key to achieving strong unit economics will be to stay focused on the customer and the value you provide, and adopt your pricing framework accordingly.