🩺 Value-based Software
It's like usage-based software but with better incentives.
Happy 2024, everyone! We’re glad to be back publishing our memos, now with a monthly cadence. Keep an eye out for a new memo from Paul or George at the end of every month!
Why does buying software almost always feel like a bad deal? I've rarely been thrilled to pull out my credit card to purchase some SaaS, even though software makes me much more effective at what I do.
Maybe it’s because most software doesn’t really automate the “job to be done.” It only makes the process of working marginally more efficient. Often, I still have to do the heavy lifting myself.
It’s easy to measure how much I would pay to outsource a task. It’s hard to estimate how much I would pay to be marginally more efficient at completing tasks. Pricing software (as it currently exists) requires hoop-jumping and hand-wringing to justify capturing the maximum amount of value from customers.
That’s why I’m excited about Intercom’s new AI support chatbot, Fin. With Fin, you only pay when it successfully resolves a conversation on your behalf. A game changer.
I admire the boldness of this pricing model. Intercom doesn’t get paid unless they complete the job for you. Their incentives are aligned with yours, and you take almost no (financial) risk by using their product. As Sarah Tavel puts it, they’re “selling work, not software.”
Many existing software companies are well positioned to start selling work in addition to their software. In order to reliably automate tasks, you need a deep understanding of the many different paths a task could take, as well as the required data to complete it. Vertical software companies often have robust workflows and integrations built into every necessary data provider for their domain, making “flipping the switch” to fully automating the work much more feasible.
Obviously, innovations in AI enable the last mile required to flip that switch. And fine-tuned models optimized for specific problems (like responding to customer questions) are getting more reliable by the day. Sam Altman was recently quoted saying, “Build with the mindset that GPT-5 and AGI will be achieved relatively soon, and most limitations of GPT-4 will be addressed in GPT-5.”
From my perspective, AI unlocks a tremendous opportunity for existing software companies to “break out” of their software, and interact with the rest of the internet to accomplish tasks on behalf of their users. I believe vertical software companies will leverage web agents (web browsers powered by AI) to extend their applications beyond systems that they control. Now, software companies will be able to automate much more complex workflows that currently require several different off-platform vendors. End-to-end solutions that were practically impossible to implement are now viable.
A second order effect of this shift will be the importance of AI infrastructure companies powering these features. If these new features are only generating revenue when they successfully accomplish their task, it’s paramount that the underlying systems work reliably. Having to re-run a task means doubling your AI inference costs. Infrastructure companies that are selling the “picks and shovels” to software companies that need reliable primitives to power their automations will act as an index of the entire space.
I want to use software with incentives aligned with mine. They should make money when I make money (or save time or save money… etc). This is a risky bet for a software company, because that means they’ll only make money if they’re solving actual problems. That’s rarer than you’d think. But most importantly, it requires a bold belief that they can solve my problem equally or even better than I can. It's remarkable to live in an era where this level of problem-solving is not just a possibility, but a reality.