Outcomes-based pricing

Bayer, founded in 1863, has one of the most intriguing AI use cases. Bayer sells seeds and pesticides to farmers, among other things. These products come with an AI offering that predicts crop yields and offers tailored growing advice.

Bayer's pricing strategy is just as unusual as its product bundle. Bayer splits revenue with farmers if crop yields exceed predictions and issues refunds if crop yields fall below forecasts. Bayer puts a lot of faith in its AI models and is willing to bet on it. Pricing is outcomes-based.

Similarly, an emerging set of products reflects confidence in their AI offerings via outcomes-based pricing.

Some examples that have caught my eye recently:

Outcome-based pricing is becoming popular because it reduces buyers' skepticism about nascent AI products. There's a common thread among farmers, lawyers, and the archetype of software engineers convinced they could build your product in a weekend. They've all heard about AI and are excited about its potential. Yet, they're skeptical. Tying pricing to outcomes gets rid of this initial skepticism.

All the outcomes-based pricing examples above take a similar shape:

1. Customers would rather purchase the outcome of some process instead of just software — resolved customer questions, translated documents, and contacted sales leads.
2. The purchased outcome is easily measured—you can track the number of questions an automated bot answered, the number of documents an AI translator produced, and the number of sales leads an AI bot contacted.

These two ingredients are necessary prerequisites for implementing outcomes-based pricing. However, they don't guarantee success and introduce new problems for vendors and buyers. 

If you're a vendor, you incur all the variable costs of delivering the product without a guarantee of revenue unless customers hit their success metric. This uncertainty does not exist in other popular pricing models. In a subscription model, users pay ahead of time and churn if they don't like the product. In a usage-based model, you're guaranteed payment based on usage, not success. Upfront variable COGS is particularly tough for startups with a high cost of capital.

From a buyer’s perspective, outcomes-based pricing limits budgeting flexibility, as the true cost of buying a new product becomes less predictable. This issue is particularly relevant for recurring users who need reliable cost estimates. For such users, the unpredictability of outcomes-based pricing means the are punished for successful usage.

To address these concerns, most startups end up with some combination of usage-based, seat-based, and outcomes-based pricing. Going forward, I expect this to be the norm. One way this is implemented is by selling credits in advance and offering a refund if the expected results aren't achieved.

The more important trend here is what Sarah Tavel describes as startups "selling work, not software." Outcomes-based pricing is a consequence of a shift from selling workflow software to selling the workflow's outcome.

This trend's most significant side effect is that bundling services and software creates more value than software alone. Global software TAM is currently ~$700B. That is a tiny fraction of the tens of trillions of the worldwide economy that is services-heavy. As language models improve in reasoning capabilities, startups have an opportunity to build businesses that convert some of that services revenue into technology revenue — possibly priced based on outcomes.