Consumer AI product retention

Olivia Moore recently posted a thread on Twitter about retention in AI consumer products:

It got my wheels turning, forcing me to flesh out my intuitions and assumptions about retention with consumer AI products.

Retention differs by era 📈

Consumer retention differs by era but always ties to what that era uniquely enabled.

For example, Internet 1.0 enabled websites and search. So, the strongest indicator of retention for the largest companies of that era was getting set as a consumer’s start page.

Internet 2.0 enabled social media. Connecting with friends and engaging with the content they share (time spent) is the strongest indicator of retention.
But from Internet 1.0, getting set as the start page has evolved into having your app on the home screen with mobile.

But what is it for “AI 1.0” or whatever this new era of AI is called?

  • Internet 1.0 (web): Making it your browser start page.
  • Internet 2.0 (social): Friends added/followed, engagement.
  • AI 1.0: ??? 🤔

What does AI uniquely enable? ✨

Today, the exciting implementation is answer engines (discussed more here). Ask a question or give a prompt and get an ‘answer’ as text, image, video, etc.

But there’s no stickiness. What prevents me from just trying the next shiny version of this? What makes anything better aside from being newer (using a larger/better model)?

This gets to the crux of retention in consumer AI apps.

Here’s the gist:

  1. AI uniquely enables near instant, custom responses based on unique inputs.
  2. Part of these inputs should be what’s unique about a consumer: needs, preferences, aspirations, etc.
  3. These unique inputs enable a differentiated, personalized experience that ‘knows’ consumers and what they want better than they know themselves.

The longer it takes to build a profile that deeply knows a consumer, the higher the switcing costs, the larger the moat.

The hard part 😅

This kind of ‘progressive enrichment’ requires investment and tuning over a long time. But how do you get the investment without making a consumer feel like they’re doing the work?

This is one of the secrets startups should be approaching in their own way to get a headstart on building their moat.

Tech giants are at a disadvantage here because all of the existing data they have feels like it should be an advantage, but it isn’t because it keeps them tied to the way things used to be. It surreptitiously becomes a liability.

Important questions for startups ⁉️

For consumer AI startups there are only two questions to determine the viability business: 1) can I get sufficient data to personalize the experience and 2) if I do, does that provide sufficient differentiation vs. competitors?

If the answer to #1 is no, you can’t do #2. If the answer to #2 is no, then you have no defensible moat (no switching costs), and any flashy new AI app could take your share — especially if an existing tech giant with superior distribution owns it.

No switching costs? No business 📉

And context windows are getting so big that they’re minimizing switching costs. Customers can easily drag and drop their data into any new app and get similar personalization.

The winners will ‘earn’ the proprietary data that differentiates their product experiences. Any consumer AI product that cannot provide a differentiated experience via personalization (data) will be relegated to low-value markets and easily replaceable by something newer and shinier.


Gabe Avatar