Generative AI is capturing the attention, imagination, and funding of the global ecosystem these days, and we’re not exempt – see our coverage on the Israeli Gen AI landscape. But in Gen AI, just like any other technology or vertical, business model is the backbone of business strategy. 34% of startups fail because of flawed business models and pricing. In this post, we aim to review the prevalent business and pricing models of this emerging domain, raise some questions founders should consider when deciding on their business models, and take a look ahead at the business models we expect to see as Gen AI technology matures.

When it comes to SaaS startups in general, we are used to two main pricing models: subscription-based and usage-based. Subscription is a set fee per month, year, etc – which provides access to the product, no matter the usage. When it comes to usage, the user is charged per the resources they consume – which can be defined as an API call, a certain amount of compute resources, etc. There is often an overlap, and we see hybrid models on this spectrum.

However, when it comes to pricing Gen AI tools, some variables need to be considered, which add complexity:

  • Cost – The resources required to run Generative AI applications can be prohibitive.
  • Automation – Gen AI can potentially replace entire functions, so seat-based pricing may backfire.
  • Expansion – The value can also be exponential, because it’s not just another efficiency tool but can potentially handle an entire job, paving the way for a greater emphasis on outcome-based pricing models.
  • Verticals – In our previous blog post, we discussed vertical applications of Gen AI, and verticals indeed represent some of the most intriguing opportunities for Gen AI. However, if your Gen AI product is targeting a specialized industry, a traditional subscription may not cut it – you might need to get creative with the biz model to get to scale.

We analyzed the pricing of over 150 Gen AI products. While we should take into account the stage of Generative AI and the fact that the domain is still expected to mature and shuffle quite a bit, affecting business models, here’s what we found:

  • 29% of Generative AI products are using subscription pricing (for example: Grammarly)
  • 19% are entirely priced per usage (for example: Cohere)
  • 52% have hybrid models, in which usage can be one of a few factors to determine pricing (example: Otter.ai) or serve as an additional pricing layer for extra usage (example: Runway)

 

As mentioned before, pricing Generative AI tools requires more considerations, making choosing a business model a fine balancing act. While subscription-based pricing allows for more simplicity and predictability, the usage-based model helps align both the value that customers get and the cost to the startup with the pricing. It also allows for more scalability and flexibility. However, effectively aligning price with value and cost in usage-based pricing models demands careful consideration and effort. Let’s see how this plays out in reality.

Midjourney –

Midjourney places a big emphasis on GPU costs. While they have a tiered monthly subscription model, the tiers are primarily determined by the amount of fast GPU time the user gets. The subscriptions also limit the number of concurrent jobs you can do and allow users to pay for additional hourly usage.

Pricing considerations: Midjourney’s pricing model is very good at aligning cost (GPU) with pricing. However, Midjourney does not capture all aspects of value here. For example, the quality of the images generated isn’t factored into the cost at all. In addition, it lacks simplicity, not intuitive to users. In Gen AI, key cost drivers include broad product scope, time sensitivity of tasks, and model complexity. Applying these cost drivers to Midjourney’s case helps clarify why their pricing model is heavily cost-focused. Their expansive ambition to generate any conceivable image, the availability of rapid processing via fast GPUs, and the intricate complexity of handling detailed, pixel-rich images – all contribute significantly to their computing cost.

Jasper –

Jasper has recently transitioned from a usage-based to a subscription pricing model, reflecting a strategic market segmentation. The current initial two tiers offer one to five seats with unlimited words, while the third tier allows for tailored plans, including API access.

Pricing considerations: This transition in Jasper’s pricing strategy reflects a thoughtful market segmentation. The first two tiers cater to smaller players with a straightforward, hassle-free package, resonating with their needs for simplicity. In contrast, the custom tier for larger organizations acknowledges the diverse requirements and capabilities of larger businesses, offering the flexibility and scalability they require.

ChatGPT

ChatPGT started as free and now has a straightforward pricing model. ChatGPT Plus has a pure subscription business model, while ChatGPT API is completely usage-based.

Pricing considerations: Separating the pricing plans based on API access makes sense. API usage is much less predictable, making fixed subscription pricing very hard and risky to implement. In addition, the API buyer is more sophisticated and they can deal with a more complicated, token-based model. It allows ChatGPT to provide simplicity for the direct access users, align cost with pricing with the API use case, and have scalability when it comes to enterprise customers.

HourOne –

This AI video generator poses an interesting case because it is rethinking the metrics used in usage-based pricing. They charge for the minutes of videos published rather than those that are generated.

Pricing considerations: The download-centered approach aligns pricing with user value, but can be dangerous since it can veer very far from being cost-aligned for a multitude of reasons, whether users just enjoy experimenting with generating videos, or the quality of videos isn’t high enough for them to want to download them.

GitHub Copilot –

GitHub Copilot, as an add-on to GitHub’s base product, mirrors this pricing model despite its unique operational costs.

Pricing considerations: This approach reflects a trend in Gen AI addons – the consistency with the product’s base pricing offers familiarity to users. However, when an addon presents exceptional value or significant added costs, diverging to a different pricing strategy may be warranted, to align more closely with the distinct value the addon provides or to make up for the additional cost structure.

More UBP – As technology matures, we expect the metrics that are crucial for user-based pricing to also develop and make this a prominent model.

API vs direct access – These two use cases will continue to diverge (as we’ve seen in the case of ChatGPT) and business models will reflect this.

Success-based pricing – The better Generative AI becomes at replacing roles and taking care of entire processes, the more we’ll see this business model develop.

Royalties-style model – As Gen AI becomes the orchestrator of processes it will charge customers for bundles and distribute the revenues. Think about Spotify, image stocks, etc.

Lastly, Generative AI is a domain in motion, and we are all witnessing it take shape on both tech and business fronts. While founders should be flexible and open to change, here are a few key questions to keep in mind while choosing a business model for your Gen AI product. It’s a good idea to revisit these questions as the domain evolves:

● Are you replacing human labor? Seat-based model creates a problem where your product is meant to reduce human force, because by the very nature of what you’re doing – you’ll require fewer seats, therefore cannibalizing your business model.

● Is model reinforcement significant for you? Improving your models entails reinforcement and training, which comes from usage. Charging per usage may discourage your customers from increasing usage, resulting in stalled improvement.

● Is your Gen AI product mature enough? Aligning value with usage can be challenging, especially if the product is still evolving and its quality isn’t fully consistent. A product that requires several ‘shots’ or prompts to deliver expected results may necessitate a more flexible pricing approach that accommodates these developmental nuances.

Wrapping up, navigating Generative AI pricing is far from a binary decision and involves a spectrum of subscription and usage-based models. Consider your product, costs, and customer preferences to find your ideal spot on this nuanced spectrum.

If you are a founder working on a cool startup, AI or not, we would like to hear from you: uril@viola.vc

If you’re interested to learn about the Israeli GAI ecosystem check out our previous blog – here