How the AI gold rush is going
Last year, I wrote a popular piece about why it is hard to sell AI software. I discussed the AI gold rush dynamic, where AI companies initially find it easy to raise money but then much more difficult past Series A. In the situation I described, businesses didn't know how to buy AI tools yet, and for startups, competing with big incumbents was even harder than before.
It's now one year later, and most organizations have figured out how to buy AI tools by now. As a result, AI startups are getting off to faster and faster starts with revenue. But even though capital is still comparatively free-flowing for great early-stage AI companies, it's harder than ever to raise a growth round.
Why is this happening, and what should AI startups do about it?
Buy to test
Over the last year, I have spoken about buying, building, and using AI with too many senior executives to count. From engineering to sales to HR to legal to finance, it's a lot easier to buy AI solutions than it was a year ago. Companies have their vetting and purchase process much more dialed in, operationally speaking.
Maybe that's why so many leaders are intentionally buying 12-month subscriptions for 4-5 times as many AI tools as they plan to renew the following year.
Yeah, you read that right.
There are too many AI tools on the market for buyers to tell them apart during the purchase process. Well-resourced teams are responding by extending their evaluation process into paid pilots; the first year's subscription to an AI tool is commonly viewed as a pilot whose value is TBD. It is easier to buy a new tool and perform your own structured evaluation than it is to ask vendors a bunch of questions ahead of time.
Teams that subscribe to several competing tools at the same time are often running controlled bake-offs. At the end of the year, they will pick one tool (at most) and scrap the rest.
In other words, most leaders purchasing AI tools plan to churn out of 80% of them a year later. By design.
It's easy to see why so many AI companies have impressive-looking revenue growth in their first year or two. It's a whole different story when the renewals come due, which is why many companies have a harder time raising later funding rounds.
Buy to copy
Vendor bake-offs as such are not new, though we are seeing more bake-offs across more tools simultaneously than I've ever seen in my career. The more interesting dynamic is not bake-offs across different vendors, but bake-offs between a vendor and their potential customer's internal dev team.
Large tech companies are racing to snatch up AI-native leaders as fast as they can to build AI tools internally. Here's what's happening:
- BigCo signs up to try a bunch of AI tools just to see what kinds of solutions are available
- BigCo hires top-tier technical talent not just to build customer-facing products, but to build internal tools
- BigCo uses their AI pilot subscriptions as inspiration to figure out what to build internally
- BigCo cancels their AI tool subscriptions because they've built their own versions
This has been happening at large tech companies for a while, but I am increasingly seeing this in other segments too; companies kick the tires on various AI tools with the explicit goal of copying them for internal use. Lately, even non-tech companies are beginning to staff with this intention.
In this context, it's easy to understand why so many AI startups are struggling with churn problems.
What does this mean for AI startups?
No, this is not the end of B2B startups, but it is going to take some time before internal dev teams at large companies figure out their limitations.
Enterprise renewal cycles in particular may stay challenging for AI startups over the next couple of years as large companies go through the paces of trying to automate their operations with AI on their own. Imo, midmarket is likely to be a little easier, simply because smaller companies aren't usually resourced to build internal tools the same way.
But just because bigger companies are trying to build their own AI tools internally does not mean they will be successful, even if they manage to snag best-in-class tech talent to lead the effort. Delivering on usable internal tools can be time-consuming and distracting, especially if those internal tools are not also products for your customers.
The more an AI tool feels like a point solution rather than an end-to-end application, the more vulnerable it is to being cannibalized by internal builders. Few companies will attempt to build their own internal video meeting software or email client. By contrast, a whole bunch will use homegrown tools (or just ChatGPT) for writing social posts or contract analysis.
It's tempting for internal builders to cannibalize point solutions because 1) they seem easy to build ("seem" being the operative word sometimes) and 2) internal versions can be more closely integrated into the company's custom workflows. Many fewer have the appetite to build production-grade end-to-end apps internally.
Of course, it takes longer for a startup to build an end-to-end app than a point solution. It's also typically harder to make your first sales because you have to convince an organization to take a bigger bet when your solution is comprehensive. Building end-to-end is a more daunting proposition and your initial revenue may take longer to ramp up.
But here's the hot take: For AI startups that want to stick around, it's better to start a little slower and build end-to-end than it is to 10x your growth in Year 1 and churn it all out in Year 2.
What do you think?
Kieran
From now until August 31, my Viral Data Stories 101 course is half-price for nerd processor subscribers. Learn to tell your own data stories clearly and effectively!
My latest data stories | Build like a founder | nerdprocessor.com