There is a question I ask myself when I'm reviewing AI deals, and it cuts through faster than almost anything else in the diligence process.
If a customer opened their laptop tomorrow and this product wasn't there, would they notice?
Because if the answer is probably not, you're looking at a tab. Something that gets opened occasionally, feels useful in the moment, and gets replaced the second something shinier comes along. We’ve seen a lot of those across the 3,000 deals we review each year, and they can be genuinely hard to spot because they look like real businesses at the early stage.
Before reading further, pause and answer this:
When you last evaluated an AI deal, what did you focus on most?
What I have learned is that the companies worth backing have built themselves into something the customer depends on.
And in 2026, that comes from one of three places.
Data that compounds
The first thing I look for is whether the startup has data that gets more valuable the more it's used.
A health company we've backed brings together electronic health records, wearable data, clinical history, all the things that have always sat in separate systems with no connection between them. Every data point added makes the product more accurate and more specific to that patient. A competitor couldn't replicate it tomorrow because the data itself is what the customer depends on, and it gets more valuable the longer they stay.
That's what I mean by a data moat; not just that they have data, but that the data compounds with usage in a way competitors can't easily catch up with.
A feedback loop that moves with speed
Between a commercial team and a product team, there's a process: customers give feedback, commercial passes it to product, product ships new features, customers react.
The pace at which AI is moving means a slow feedback loop doesn't just cost time; it costs the right product entirely.
One company in our portfolio has 0% churn at a stage where most startups are still working out who their customer is. That number tells me that the feedback loop is working. They're hearing the right signals and getting them into the product quickly enough to matter.
Distribution with real switching costs
This is the one I care about most in 2026.
Technology can be copied, and traction can be manufactured, but if a startup has built a community around its product or created a distribution strategy with real switching costs baked in, that's much harder to displace.
One company we've backed sells through financial institutions and has spent years building regulatory trust inside those relationships. A new entrant with the same product couldn't walk in tomorrow and take those customers. The trust, the compliance process, and the relationships take years to build.
Running the tab test yourself
When you're sitting across from a founder, these are the questions worth asking:
What happens to your customer's workflow if your product disappears tomorrow?
What does churn look like at 12 months, not 3?
And who owns the relationship with the end user, you or the platform you distribute through?
The answers tell you quickly whether you're looking at something a customer depends on, or something they could replace by Friday. Most AI startups at the early stage are closer to the second than founders will admit, and the diligence process rarely surfaces it unless you ask directly.
That last question is one I go into properly in this week's video if you want to go deeper.
Warmly,
Rupa
P.s. When you're ready, here are 3 ways I can help:
Follow me on LinkedIn: I share quick takes on deals, founder patterns, and what I am seeing across the ecosystem between newsletters.
Subscribe to my new YouTube channel: I'm releasing in-depth videos every week on how to succeed with angel investing.
House of Arāya Membership: Access pre-vetted deals, co-invest alongside Arāya Ventures, and join a community that pools diligence and shares real perspectives.
