Another VC recently shared their list of the most common startup ideas they often get pitched. It made me reflect on the most common AI-related pitches I’ve heard of late. Here’s the list: 

• Robotic Process Automation (RPA) 2.0: building a platform that enables enterprises to build their own GenAI workflows 
• Enterprise adoption platforms: very related to RPA 2.0 but more focused on solving technical issues around data security, deploying into production, managing models (ML Ops)
• Enterprise knowledge management: ChatGPT on your company knowledge base
• Legal tech: automating document generation of some kind
• Brand/marketing workflow automation: help people develop brand-complaint designs/copy, complete brand work, or automate specific processes within marketing (like drafting creative briefs)
• Agents for [X]: agents for customer support, sales, etc
• Personalised personal assistants: either in a work or personal capacity

I thought I’d share some thoughts on Robotic Process Automation (RPA) 2.0 where businesses are building platforms that enterprises can use to build their own GenAI apps and workflows on top of.

It’s a huge opportunity space: much larger than the current RPA market given how much more we can do with GenAI than rule-based automations. While I have a lot of excitement for what will come from this space, I also think it’s one of the hardest spaces to be building in during the current generative AI wave.

Why? A few key reasons:

1. Many enterprises don’t know what they want from AI yet and those that do aren’t easy to sell to

Enterprises that are sophisticated enough to know what they want from AI are more likely to experiment with building in-house directly on top of foundation models than buying a low-code automation tool. A16Z reflected on this in their recent article on enterprise AI adoption:

Enterprises are overwhelmingly focused on building applications in house, citing the lack of battle-tested, category-killing enterprise AI applications as one of the drivers. After all, there aren’t Magic Quadrants for apps like this (yet!). The foundation models have also made it easier than ever for enterprises to build their own AI apps by offering APIs. Enterprises are now building their own versions of familiar use cases—such as customer support and internal chatbots—while also experimenting with more novel use cases, like writing CPG recipes, narrowing the field for molecule discovery, and making sales recommendations.

Those who aren’t sophisticated enough to know what they want from AI require the startup to; A) help educate them on what’s possible, B) help them craft the business case and C) hold their hand through lengthy implementation projects.

A) Educating: My observations from my time at Uber are that when the market requires educating on a concept it can put first movers at a disadvantage as you are the one who has to spend capital on educating the customer base. Later entrants can avoid this “first mover tax” and spend their capital on acquiring your hard-won customers instead. I expect vertical software players will reap the benefit of the education work horizontal players are currently doing around AI use cases (see section 2).

B) Business case building: Have you heard of the Law of Shitty Cohorts? Another Uber learning. Your best customers are usually your earliest adopters because they often find you out of a deep desire to solve a painful problem. I worry that in the case of horizontal platforms, you need to not only sell them on your product but also on a problem to solve. In the case of single-problem startups, you can focus your time on a group that feels that problem acutely.

P.S. I do think there are ways around this, like focussing on a specific problem area to start and then scaling from there or offering templates for specific problems.

C) Long implementation times: The low technological readiness of enterprises favours hyperscalers like AWS and Google who have armies of Solutions Architects (SAs) ready to hold their hands in the hope of locking them into their cloud solutions. Hyperscalers know that they need to deploy serious OpEx $$$ to protect their cloud revenue from the AI shift. For startups, it makes scaling much more cumbersome as it requires more human capital. Either that or you partner with channel distributors and decrease your margins that way.

2. It’s hard to compete with best-in-class vertical software

We know that building in-house is one form of competition in this space. We also know that players like ChatGPT, Adept, Zapier and UI Path will form some of the competition, too.

That’s not all. If you’re building a horizontal automation platform, your competition is also all of the vertical software products that solve the problem your target customer most wants to solve.

From the same article as above on enterprise AI adoption:

However, the jury is still out on whether [the trend of building instead of buying] will shift when more enterprise-focused AI apps come to market. While one leader noted that though they were building many use cases in house, they’re optimistic “there will be new tools coming up” and would prefer to “use the best out there.”

So if your target customer wants to use your platform to build a customer support bot, you’re also competing with the thousands of software companies that specifically focus on customer support.

It’s difficult to offer a better product experience and design than someone who exclusively focuses on that problem. That is part of why we’ve seen so much unbundling in SaaS: people want best-in-class solutions.

3. Everyone has identified this problem and we can all see how large the opportunity is

As noted above: competition is severe and it’s coming from all angles. It’s easy to make a throwaway comment as a VC that “lots of competition! scary!”, but it’s hard to overstate how competitive this particular space is right now.

Higher competition = higher cost of customer acquisition, lower contract win rate, more difficulty in getting the best talent, and fragmented revenue capture.

4. There’s a deeper risk of disruption

While I don’t think it serves founders to ruminate too heavily on where the foundation model layer goes or when we’ll encounter AGI (whatever your definition), the rapid pace of development in Agents makes me believe that folks building horizontal automation platforms are at a greater risk of technological disruption than even some “GPT wrapper” companies. Hot take!

Let’s play out one possibility with ChatGPT. I suspect that they didn’t introduce the GPT Store purely to make more revenue. I expect that the workflows people build on top of the Store will, at some point, be used to train models. I don’t mean train models on the data that goes through those workflows, I mean train models on the structure of those workflows.

Why? I expect that there’s not enough openly available information online on how people solve specific business problems step-by-step.

If you have thousands of people building the steps of different workflows on top of your product, that’s a pretty powerful way to teach an AI model how to accurately solve more niche problems than it currently can. I think it’s part of their strategy to reach AGI.

If they can accomplish this, building automation and workflows might not require platforms and instead may become native to the next generation of models instead.

I know people will take issue with this as it’s so speculative and perhaps counter-intuitive, but it’s one example of how companies may look to train models that can solve a large variety of workflow problems with much less UI-level input from users.

The company that can take users from problem to solved workflow the fastest is likely going to be a company that has solved that at the model layer, rather than the interface layer, I think.

Final thoughts

To close things out I wanted to caveat that none of these concerns are reasons I would be against investing in this space. I have advocated for us to invest in companies in this theme of automation.

One of our values at Square Peg is Anchor to Optimism. As VCs we need to look for what can go right; in any given startup there’s plenty that could go wrong and if you fixated on what might go wrong you’d never make an investment.

These are simply some of the things I think about with this particular type of business, and I could create the same type of list for other businesses I’m commonly seeing in AI.

‍