The History of Software & Pricing:
In the 90s, buying software was an adventure: you'd drive to a store, grab a colorful box filled with disks, pay upfront, and hope you'd actually use it. Forget about updates, if a new version was released, you'd repeat the entire ritual.
Then came the internet, where software was forever changed with the SaaS model. Suddenly, software lived online, always fresh and up-to-date. Companies like Salesforce and Google soon jumped aboard, making SaaS the standard. But one pesky issue lingered: subscription fees. Companies often paid annually per user, even if those seats sat untouched, gathering digital dust also known as “shelfware". But, buyers just put up with it or waved their hands at “boosted productivity” numbers since paying for a subscription was easy and predictable.
This was until cloud providers tackled this problem head-on with usage-based pricing. The message was simple: pay exactly for what you use, nothing more, nothing less. The model made sense at the time: more usage meant higher costs, less usage meant lower bills.
AI-native companies have been rapidly adopting this model, promoting usage of LLMs and LLM-enabled services. But, with the arrival of AI agents, the natural progression of pricing for companies building on the application layer shifts into what we call “outcome-based pricing”.
AI Agents:
In Layman's terms, agents are just pieces of software that are able to work autonomously. They are powered by large language models (LLMs) often built by frontier labs like OpenAI and Anthropic. Every time an agent completes a task, it costs their user “tokens” which are similar to credits, but under the hood are really just how much it costs their provider to display parts of a word (usually a few cents).
Independent software builders can use the underlying LLMs to create their own domain-specific agents that excel at completing niche tasks. In our Y Combinator batch, founders are building agents to carry out tasks from domains ranging from debt collection to medical billing to software engineering.
These agents are programmed and trained with domain-specific information, guardrails and evaluations, to make sure they complete the task properly. Oftentimes, the output these agents generate already trump the speed and ability of a human to do the same task.
We are at a point where change is no longer a possibility in the workforce. It’s already here. Every single company in the world has a plan for how they want to embed AI agents into their business, at the center of their whiteboard. But, those same companies in large part have no plan for buying or selling AI the right way, with a few exceptions.
The Impending Change:
AI agents are just another word for software that can autonomously complete workflows from start to finish and deliver results. So, wouldn’t it make sense to pay your agents accordingly?
We think yes. It doesn’t make sense for those selling agents to adopt a traditional subscription model, when they are no longer augmenting the productivity of a human, but replacing it. It also makes sense for buyers to pay agents for a job well done on a completed task, rather than on a per token basis – the equivalent to paying engineers a fee for every line of code they wrote no matter the quality.
In our eyes, this is the best thing that has happened to software. Incentives are finally aligned between buyers and sellers. No more shelfware and no more undercharging.
Build your agents to deliver results and get paid when they do.
Outcome-based pricing is still a relatively new term. Only a few companies at the cutting edge have truly adopted it, for a few reasons.
The criticisms of outcome-based pricing are often that:
Customers won’t like the unpredictable costs associated with the model
Companies selling agents like to see predictable revenue (as do their investors)
There is no standard way to enforce what a successful outcome is
There is no billing platform that natively supports it
We think that in large part these are short sighted claims that dismiss the core principle of the alignment of incentives between buyers and sellers. That’s a big deal.
Also:
If a seller can prove their product can fundamentally drive more business value while allowing the buyer to assume no risk, paying only if it works, they will adapt.
If a seller is able to drive that value and capture a portion of it, they will generate much more revenue than they ever could’ve with standard pricing models.
If a buyer and seller cannot agree what a successful outcome looks like contextually, then this model probably isn’t a great fit for them.
The infrastructure to support this will catch up, it already is.
Early Adopters:
Salesforce, Sierra, Zendesk, and Intercom are a few of the early movers in adopting an outcome based model. Their definitions of a "successful outcome” vary from simply facilitating a conversation (Salesforce) to completing a customer support query with no human elevation needed (Sierra).
Chargeflow is another company that automates the process of collecting revenue and preventing chargebacks for ecommerce, which has adopted this model. They take 25% of each recovery and charge $39 for each chargeback prevention. Their pricing page explains the idea perfectly: success first, pay second.
The early feedback from Sierra’s customers, including ADT, Sonos, and Clear, are evident: they love it. Their users are able to prove the return on investment, with just some simple spreadsheet math – a new world! The costs are even predictable at scale.
For Chargeflow, they’ve recovered over $100M in revenue for customers and are able to guarantee a 4x ROI for every single user.
What Works for My Business?
It’s important to note that this will not be the best option for every company. We understand that subscriptions, usage and hybrid models will still have their place in certain scenarios.
But, there is no debate that properly pricing and buying agents HAS to be a priority.
Future Predictions:
We don’t have a crystal ball, but this conviction stems from being surrounded by AI native builders. Every single AI native company has at least thought about these models and what the best course of action is for their company.
The same cannot be said for incumbents across industries.
As underlying models get better and moats are harder to build, pricing and business models WILL become a moat. The ability to adapt, iterate and shift perspectives will be a differentiator and help win deals.
we’re pushing this topic forward at useskope.com :)