Horizontal vs Vertical Software

Transcript

It's interesting. So Anil Bouchri is the founder and CEO of Workday and also is a very successful partner at Greylock. He came to a partner meeting in 2017, a year after he joined the firm. And he said, guys, why are we looking at horizontal software? There's no paradigm shift that will allow us to go take on the incumbents. We'd be better off just buying the public stocks of the horizontal SaaS companies, and we should actually go do vertical software, right? Because that's where, as a VC firm, there's low-hanging fruit. And I remember there was debate and skepticism, and a lot of horizontal software companies got founded. But by the way, fast forward the clock, you would have been better off buying Salesforce, ServiceNow, Workday, Palo Alto, CrowdStrike, and the public markets. That basket would have done better than any horizontal startup SaaS basket at the time. Because it turned out that either people were building things for SMB, they were building niche add-ons, but no one really could go build a disruptive core company. Now, fast forward, I was with Anil at a Warriors game spring of 2023, and he was like, his advice to me in Greylock was, now's the time to go do new horizontal things. And why is it the time? Because it's the first time since 2005 when Palo Alto and Workday get started at the start of the cloud boom, where you have the confluence of three things that need to be true in order to build new disruptive horizontal companies. One is there's a new pricing model, right? You went from license to SaaS to now outcome-based pricing. The second is there's a new abstraction or unit of value for work, because now you can actually complete end-to-end tasks. So you're selling something very different than workflow software that's used by people to complete tasks. And then the third is the data model's different, right? When you went from on-prem to SaaS, you had this multi-tenant elastic architecture that could support new use cases. In an AI world, why does the entrenchment of the system of record even matter? Salesforce has this whole schema that they think about customer objects with. The whole world is standardized on that schema. I'd argue that gives them a lot of defensibility. Why do you need that anymore? Why should you as a CEO go hound your reps to update their account information in Salesforce? I should have an AI system that integrates into email, into my calendar, there's a recorder in the sales conversation. And it's just implicitly on the fly, creating the data schema that I exactly need to power these outcomes because I want to. If I want to do a pipeline forecast, wouldn't it make more sense for that to be substantiated based on the texture of the actual customer conversation than what my Salesforce instance says? [SPEAKER_00] Yeah. I mean, the data needs to live somewhere, but it doesn't have to be in a... And a structured schema. [SPEAKER_00] It's... I would argue... [SPEAKER_00] It doesn't have to be structured. Exactly. But it needs to exist. It needs to exist. [SPEAKER_00] But I would say that a lot of what makes these systems of record that we look at today defensible is they have defined the ontology and they've done the structure. Yeah, exactly. And then if you want to plug in and you want to integrate with this thing, you have to match it. Correct. And that's why the whole ecosystem has been able to run that. And by the way, you as a rep have learned how to use Salesforce. [SPEAKER_00] And what I would argue now in AI is it's all dynamic, it's all generative. [SPEAKER_00] And so all of that work that you've done is no longer needed. Like, Salesforce has this whole schema that they think about customer objects with. The whole world is standardized on that schema. I'd argue that gives them a lot of defensibility. Why do you need that anymore? Like, why should you as a CEO go hound your reps, like update their, you know, account information in Salesforce? I should have an AI system that integrates into email, into my granola, into like, there's a recorder in the sales conversation. And it's just implicitly on the fly, creating the data schema that I exactly need to power these because I want. If I want to do a pipeline forecast, wouldn't it make more sense for that to be substantiated based on the texture of the actual customer conversation than what my Salesforce instance says? Yeah. I mean, the data needs to live somewhere, but it doesn't have to be in a... And a structured schema. It's... I would argue... It doesn't have to be structured. Exactly. But it needs to exist. It needs to exist. But I would say that a lot of what makes these systems of record that we look at today defensible is they have defined the ontology and they've done the structure. Yeah, exactly. And then if you want to plug in and you want to integrate with this thing, you have to match it. Correct. And that's why the whole ecosystem has been able to run that. And by the way, you as a rep have learned how to use Salesforce. And what I would argue now in AI is it's all dynamic, it's all generative. And so all of that work that you've done is no longer needed.