Pricing for AI Solutions

Alex David
Alex David
Originally published: 03/Apr/23
If you are thinking about introducing Generative AI (GenAI) into your product, you need to also be thinking about how you are going to introduce GenAI into your pricing and packaging.

In this article, we look at different product objectives, talk about cost structures and pricing options, and project how AI tools would be monetized both in the short term and the long term.

Why ‘how you price’ Generative AI might be more important than if you have it or not

You want to integrate Generative AI into your product offering - maybe because you have a truly cool use case for it, maybe because you think you need to since ‘everyone is doing it’. Should you do it? 🤷‍♂️ That's between you and your deity of choice. What I want to discuss is this: if you are thinking about introducing GenAI into your product, you need to also be thinking about how you are going to introduce GenAI into your pricing and packaging.

‘Why does this matter? People have been doing AI/ML things for a while now. It’s not that big of a deal. I just need to get this into my product and #ship-it!’ - I can hear it. I can understand it. But it worries me. I am not going to go into the details of how Large-Language-Models work, or the nuances of these types of generative AI models - these days it feels like that’s what half the content of the internet is about. What I do think is important to call out is the following:

  • Previous ‘AI’ had complex IF-THEN logic and some decision-tree style algorithms that did a decent job of providing insights or answers around simple questions. Charging for this at scale was often hard because either it wound up bespoke, or wouldn’t always work consistently making it hard to trust if a dollar number was attached to it.
  • This generation of AI is different - it actually creates a pretty massive impact. And more importantly, it's readily available. This isn't necessarily new - MoveWorks has had pretty impressive AI for a while, but it was integrated into a pretty beautiful and closed-off system. GenAI is in your face and begging you to directly interact with it.
  • At Segment, we used to communicate the value (partially) in terms of how much engineering effort we could save you. Use Segment to integrate your sources and destinations with a click instead of engineers having to build and maintain. Very compelling. Worked pretty well for anyone keeping track 😉 A lot of the use cases of GenAI are similar - generating marketing copy, for instance, can massively empower a company to do more with fewer people. is throwing up some crazy stats with that very idea. People are willing to pay for that and so thinking about the pricing is worth your time. would definitely say so.
  • To be clear, I do not think having GenAI means you should cut your team sizes. I am a growth/topline guy - I would argue all day long that you should use GenAI to make your 100-person sales team as effective as a 1000-person team, rather than lose 90 salespeople and maintain the effectiveness of 100.

There are a few key questions I would be asking myself at this junction:

  1. What are my product objectives with GenAI?
  2. What is it going to cost me?
  3. What are my options for pricing GenAI?
  4. What are other people doing?

So let's discuss each in turn, and maybe even look a little into the future of where this might all be headed.

What are my product objectives with Generative AI?

Well, are you enhancing your current, fundamental product offering or are you expanding to a whole new use case? Depending on the answer to this question, you should be thinking about what packaging structure makes the most sense to support it. E.g. if it’s a new use case, a new package could be created alongside the current offering (aka an ‘add-on’) vs simply integrating it into the current offerings.

If the goal is to leverage AI to capture new customers, then the standard acquisition rules hold true. Cheaper, lower commitment, more value upfront, etc. all help get your product into the hands of the masses. Potentially as a completely stand-alone option.

If AI is meant to further expand on the current product value, then it should either be added to the line-up (for limited/no cost if trying to capture market share) or added as an add-on with its own (though complimentary) pricing structure.

‘Well, why can’t I just throw it in for free and retire on my amazing success?’ 🙃 You could, potentially! But I think it's worth noting another major difference with this new form of AI. Namely…

What is it going to cost me?

This version of AI is not cheap. You are really facing two options: using an existing option (like OpenAI’s GPT3) or building your own LLM (or similar). Existing tools allow for rapid integration and setup, this means a quick GTM and the ability to leverage market momentum. However, you are limited by the restrictions OpenAI puts on the solution and have to pay them for its usage. You could try building your own, which allows for complete customization and vertically integrated costs, but unless you have been working on this for a while already, it is both time and resource intensive.

Thus, assuming you are going with an in-market solution you need to be mindful that the variable cost is high (especially considering the usual software cost model). Every prompt from a user requires a massive amount of computation and often many prompts are required to get at a result that the user finds acceptable. OpenAI has very transparent pricing and it's clear you would be benefiting from the fact that it wants people using it (read: it's pretty cheap). Another thing to consider is that integrating the technology can be arduous depending on the application / intended purpose, meaning one should potentially expect a high engineering cost to productize as well.

‘But I am a software company, I care about growth, not costs!’ - SaaS founder from 2021 🤦‍♂️. Unless you are one of the GenAI/NLP companies you probably do care about costs right now. With the shift in focus toward CAC reduction and runway extension, embedding a high-cost capability into your product should be done with care, and making sure you are monetizing the additional value can be a great way to ensure you are growing both in the market and on the balance sheet.

‘Okay fine, I need to charge for it. So how do I do that?!’

What are my options for pricing GenAI?

Generally, there are two models emerging right now: charge for access and charge by use. I see these models as a direct reflection of who the audience of the tool is. If you are a user of GenAI, in so far as that you are using it to, for instance, write copy, or generate content, a charge for access likely makes sense - e.g., a fixed fee per user, like what Notion is doing.

If you are a developer using GenAI in your product, i.e you are integrating it for your users to use, then I believe a per-use/prompt/token model makes more sense - e.g., a token or prompt-based fee, like what is doing.

If you have both user types, aligning your metrics to your user types is not uncommon and can become an easy way to help differentiate between them since it's hard for them to price compare those use cases - which is great given they often have different WTP.

Segment did something similar, in that its original price model was more developer-focused in the early days around API calls and sources/destinations vs. aiming more towards marketers in later stages and then adopting MTUs.

In either case, it's critical to adhere to the basic principles of price metrics…

  • charge based on value/outcomes* delivered
  • make it easy to understand for your intended user
  • charge for it distinctly/make it easy to validate
  • make sure it can scale as customers do

*potentially charge for outcomes if this can be measured and doesn’t impede usage (i.e. per prompt by per published post)

For many, given the high cost (and recent market focus on gross margins) it's likely worth pursuing a fixed model that covers most users’ use cases with some form of an overage rate for extreme users.

I would discourage charging per prompt for non-dev users, at least initially, as this will limit adoption and usage which is the opposite of what you want with novel technology. A purely fixed fee would be great for adoption purposes but depending on the user base could create a high risk of margin erosion. Depending on the stage of the company this may be a more or less critical factor.

One could construct usage limits to impede the costs (processing speed in which results are returned, delaying queries to off-peak computation times, etc.) although all of these would create poor user experiences that may potentially erode user adoption/retention.

‘Okay…that was a lot. Can I just copy other people? What are they doing?’ 🥲 Sure, that is an option…

What are other people doing?

We spoke about Notion and Jasper, and I think both have a good sense of how they are approaching their customers. OpenAI is clearly targeting the developer market with its token-based pricing model on the backend, and the creator-user with the access-based model for ChatGPT Plus.

Clearly, there are a number of ways GenAI is being leveraged, either as standalone products (e.g. ChatGPT) or by being integrated into existing products (e.g. NotionAI). The variety of the application is exciting for its impact on adoption and acceptance to market but also creates data points around how this may determine monetization models.

The ChatGPT Plus and Notion models are fairly standard approaches - given the widespread adoption and usage of the tool this could be an interesting model, though presumably, those willing to pay are also the most aggressive users and thus likely to drive the most cost. They may even be encouraged to use it more given the ‘skin in the game’. Time will tell what this means for their bottom lines.

Microsoft is doing something completely different by integrating GPT into Bing, its search engine. This seems to be a competitive play to steal market share from Google. The assumption being Bing will remain free but drive ad dollars to Bing as it claws the market from Google. This could be a very interesting play to make Bing relevant and one I will be watching!

Notion integrating AI into its beloved product is a cool application. I have been using the Notion AI as a writing assistant and it's both intuitive and very useful. My guess was that they would enable it as a paid-for add-on (what they did) and I think that makes a lot of sense - not restricting it to a particular tier means they can drive adoption across users and drive up ARPU.

‘Man, this has been helpful. It’s a really exciting time and now I feel confident about how to think about pricing my GenAI adoption. So what's next?!’ 🥰 Thank you! And yes, that is the question, isn’t it? I have a few thoughts there as well, though it’s really just an educated guess.

So what's next?

In the near term, I believe the ‘co-pilot’ concept will continue to take off. I don’t think companies will trust it enough to exchange human resources for AI models. This means we will see the rise of tools that enable humans to be more effective. We are already seeing this in the space of copywriting, code creation, testing, etc.

From a pricing lens, I could see this impacting how the AI co-pilot is ‘comped’. i.e. if an AE is given a GenAI co-pilot to support with lead sourcing, outbound activities, conversation analysis, and negotiation support, I would not be surprised to see GenAI collecting a commission on those deals.

In the long term, and hear me out here, I could see these AI tools actually becoming the default interface for people. I can easily imagine that same AE interacting with a simple AI interface, asking it what meetings it has coming up, pulling up reports (summarized of course) on the prospect calls, what those prospects likely will be most interested in, pulling relevant battle cards and value positioning playbook, all the while none of those tools we use today are ever touched by the AE. The AI is running the show.

In this world, monetizing this AI becomes interesting. Is it revenue sharing? Is it a massive ERP/CRM-style platform? Or will we see a Netflix-style streaming revenue allocation? I.e. the sales organization pays the AI tool $10M a year, and the AI tool divides $9M of that across the dozens of tools it’s pulling from on the backend based on how much they are contributing to its impact.

I am excited! 💪 It’ll definitely be interesting to see where this goes. At the very least, those of us who spend our days thinking about pricing will continue to have some interesting challenges to figure out.