The Tetris Opportunity For Services As Software Companies
A while back I wrote a piece on the Services-As-Software business model for AI, where you take a workflow that is largely human today, keep the same user interface, but replace the rest of the workflow with automation. Today I want to explain why the first companies to adopt services-as-software models will end up dominating multiple markets — sometimes unrelated ones.
There are two pieces to understanding this opportunity. First, it exists because every AI company I am invested in that is working with a services-as-software business model is struggling with the same question — should they provide their services to other businesses? Or should they be full stack and compete with the other market players, or both? To keep strategic discussions confidential, let’s think about grocery store AI, an area in which I am not invested.
If you are a startup building the technology to power an Amazon Go style store — all tech, no employees, the question is, should you sell the tech to retailers, or should you build stores and compete with them? You capture more of the value created if you become a retailer, and you can customize and more tightly integrate your offering rather than make it standardized enough to be a product that appeals to a large part of the market. But, that takes more capital and time, and expertise in retailing that you have to hire.
What is starting to emerge from watching a lot of these companies struggle with this decision is that many of them will be able to do both.
There are plenty of precedents for this. The most notable is Amazon’s realization in the mid 2000s that running scalable on-demand I.T. infrastructure was a core competency, thus a business it should be in for others. Uber has also done a similar thing recently with mobility software.
In the AI space, this will be even more possible than before. Digital business models and cloud computing made this possible because the marginal cost of serving a new customer was low and these models are very scalable. Now AI will make that possible for things like learning, prediction, and automation.
So the two versions we will see are like this:
Version 1 — A company provides a full stack service that competes in established markets by building an AI-powered value chain at certain key steps, and then offers those pieces of the value chain that are AI-powered to others in the industry as a new business.
Version 2 — A company that provides one key AI-powered piece of a workflow to customers will get so good and so smart by learning from all the companies it serves, that it will compete in some part of the market with a full stack offering.
Version 1 will be more common initially, primarily because AI adoption was slow enough the past few years that even though companies wanted to do it, implementation was tough because of the culture change required. So competing head on full stack was a smart move.
Version 2 companies will show up in the second wave, when, as the strong provider of a key part of the service with decent scale, they can now gain attractive financing to go full-stack (or, acquire/merge with a full stack player)
I was going to call this a T-shaped business model because you have a company providing a horizontal service to multiple industry players, plus a full stack vertical slice — making a “T.” But as I was writing this over 2 glasses of wine, I realized (or maybe was misled by the wine) that we you could end up with multiple competitors with T-shaped models but competing with different parts of the stack as the offshoot.
So, think of a version 2 grocery warehouse robot company, and a version 2 grocery delivery robot company, both different parts of the grocery automation value chain, deciding to compete as full stack grocers. If you lay their value chains beside each other, one looks like a “T” and one looks like a “+”. So, it’s more like tetris pieces in that different companies will even compete by pulling off different AI pieces of their value chain as they compete full stack, thus having all types of different shapes.
And really, some may build in multiple directions — starting as a Version 1 full stack company, pulling off a service to offer to others, then building a new full stack on that service as the extension pulls them into a new market. (Note that new markets here may be defined more by data types than by customers type). As a result you can get all kinds of funky shapes of value chains going in many different direction, bolting on pieces everywhere that they are the first company to learn/automate something that can make them competitive somewhere else.
If you are building a Services-As-Software company and thinking through this, we’ve already spent a few years investing in some businesses like this at PJC and it’s one of our favorite strategic topics, so please consider us as an investor.
Thanks for reading.