How Software Is Eating the World
A virtuous cycle of software development and business model innovation
I’ve been thinking a lot about business models and go-to-market strategies lately. We’re building a B2B marketplace at Inventa powered by credit and selling tools (and someday, ads and logistics). Pricing these products and combining them into a cohesive platform are critical things to get right. Instacart’s recent S-1 filing provides one relevant example with their marketplace, enterprise platform, and ads product (as usual,
explains it best.) All of these products are software but they create economic value in very different ways. Ultimately, we’re trying to get more people to use more software. This creates economic value and we aim to capture some of it.There’s a well-known quote by Alfred North Whitehead from his An Introduction to Mathematics that “Civilization advances by extending the number of important operations which we can perform without thinking about them.” Software didn’t exist in 1911 but Whitehead was anticipating Marc Andreessen’s Why Software Is Eating the World by nearly a century. Whether we frame it as advancing civilization or creating economic value, the goal is to automate existing or novel tasks with software. But you can’t just write the software, you have to get people to use it.
SaaS is one way to achieve this. You write software, host in the cloud, and charge fees based on some proxy for value creation like seats, compute, or usage. The people who work for your customers can do more while thinking less. SaaS is having a pretty good run. Global SaaS spending will be about $200B this year.
But most of the world isn’t software-driven. Global GDP is over $100 trillion and we’re spending much less than 1% of that on software. To be fair, we spend around 4% on IT broadly. Even if tech companies are only capturing a small sliver of the value they create, we have a long way to go. Part of this gap is just a waiting game as SaaS spending is forecast to grow in the low double digits per year. But another part is that SaaS is not the only (or often the best) model to replace what people are currently doing with software.
Getting people to use software creates economic value and you aim to capture some of it.
A spin on the single-product SaaS approach is
’s compound startup approach. His point is that for some sets of products, integration is the product. Even though you’re violating a startup golden rule by building multiple products in parallel, you are actually focused on the dimension that matters most to your market. You can also price the bundle better than your collective competitors. Rippling is of course an example, as are HubSpot and Microsoft. This is a turbo-charged version of “SaaS is eating the world”. Parker is automating more stuff by selling the bundle but he is also automating the work needed to operate the pieces together.Another approach I’m familiar with to get software into companies is to build a marketplace. In an inversion to
’s now-famous come for the tool, stay for the network, suppliers participate because they can acquire demand that is as profitable as what they can find elsewhere. You build software tools for them to make your platform usable and also to drive your flywheel. The suppliers benefit from this software but you typically aren’t monetizing it directly. We’re a couple of decades into building digital marketplaces now and I’m sure that will continue, probably in the domain of services powered by LLMs.If you take the managed marketplace model to the extreme you end up vertically integrating the supply side. Opendoor is a good example that’s explicitly driven by software and machine learning. This type of model works best when there are significant gains to scale and low benefits to fragmented supply. This makes it easier to adopt software on the supply side because you own it. You can also tailor the software to maximally integrate it with your operations.
If vertically integrating is the “build it” approach, Private Equity is “buy it”. PE firms are an important driver of software adoption, especially in older, fragmented industries. Rather than convincing people to pay for software, you just buy the company and make them use it. This can be the right approach in markets with a lower inclination to adopt new technology, like SMBs and blue-collar firms. But what does this look like if a startup is doing both the software building and the company acquiring? Teamshares has acquired more than 80 SMBs, using a combination of Fintech and employee ownership to create value and monetize. And Keith Rabois is having success buying Shopify stores at Openstore.
Another model that could be a good fit for certain markets is franchising.
at Slow Ventures is thinking about this. Startups that build software could use the franchise model to penetrate markets in a way that borrows from the top-down PE approach without the PE capital requirements. The more predictable nature of franchiser cashflows could help make smaller markets more appealing to software builders. These startups could also manage the franchise portfolio using their own vertically integrated software stack.Finally, generative AI creates the opportunity to simply sell work rather than software.
points out that EvenUp allows personal injury lawyers to buy demand packages externally rather than purchasing software to generate them internally more efficiently. I don’t know the team but Monk is doing something similar for DevOps using LLMs. Instead of tools to make your DevOps team more productive, their AI will do the work for you. This is the virtuous cycle of software enabling a business model that gets society to automate more activities with software.You can’t just write the software, you have to get people to use it.
There’s no doubt we’re still in the early innings of Andreessen’s prediction. As we start to exhaust the opportunities available in the space of vanilla SaaS businesses, compound startups and marketplaces will continue to offer fertile ground for entrepreneurs to uncover new combinations of products and markets. The PE-like acquisition route is newer for startups but is working in some cases. And it has a younger cousin in the tech-driven franchise model (an idea I need to explore more).
My guess is that the LLM-driven work model is the one that will change things in a big way very quickly. Given the nature of the technological shift, it’s certainly the highest variance bet. Selling work makes the value legible to the buyer and API access is a low enough bar in enough markets that growth will be relatively unconstrained. That said, it’s difficult to make predictions, especially about the future, so I’m very likely wrong. Either way, the business model innovation that we’ll see in the coming decades should be as interesting as the technology that enables it.
If you’re thinking about these things leave a comment or DM me to chat.
Thanks for the ideas and for reviewing a draft of this post.