Thanks for reaching out. Be sure to check us out on LinkedIn for all of our current news and announcements.
Oops! Something went wrong while submitting the form.
Part 3 of 5 in a series on Lean Analytics in an AI era. Haven’t read the last two editions? You can find read Part 1 here and Part 2 here.
For years, stickiness meant building walls. Switching costs, lock-in, data gravity. Keep the user captive and they can't leave.
AI is killing that model, and replacing it with something better. Let’s look at the next three product shifts: stickiness, quality, and trust
Traditional stickiness was a frequency game. DAU/MAU, return visits, habit loops. Andrew Chen wrote years ago about where DAU/MAU breaks down: episodic but high-value products, weekly-rhythm tools, anything that isn't a daily habit. AI doesn't kill DAU/MAU. It amplifies the limits he was already calling out.
Two things are happening at once.
First, users expect to do more with an AI product than they did with the single-function tool it replaced. "Surely I can do more with this than the old tool" is becoming the default mindset. That's an opportunity. The signal for where to expand your product is sitting in your prompt logs right now. Task diversity per user is a growth vector that didn't exist before.
Second, sticky AI products are less about walls and more about being in the flow of the user's work. Trace Cohen put it well: "moats are dead, long live canals." Moats scale through exclusion. Canals scale through throughput. You're indispensable because of how much flows through you, not because the user is trapped.
What to measure alongside DAU/MAU: task diversity (are users stretching you into use cases you didn't scope?), integration depth (each connected tool is a canal), trigger diversity (one reason to come back, or many?), and workflow chaining (do you hand off to other tools and receive handoffs back?).
One warning. Being "in the flow" can mean being a tiny canal between two big rivers, useful today but easy to reroute tomorrow when the tool upstream ships its own AI feature and absorbs your slice. The defense is to eat more of the chain. Claude Code didn't sit alongside the IDE. It replaced enough of the editor, enough of Stack Overflow, and enough of the manual work to become a different shape of product. So add one more metric: replacement breadth. How many tools and manual processes did the customer drop when they adopted you? If the answer is zero, you're still a canal that can be rerouted around.
Traditional software works or it doesn't. You ship it, instrument it, move on.
AI output is a distribution, not a property. An 80%-good product and a 95%-good product feel like completely different products to a user. That gap matters more than anything else in your funnel. Quality isn't something you ship once. It's something you watch like retention.
Klarna is the cautionary tale. They went big on AI-only support in 2024, claiming it did the work of 700 agents. By mid-2025 the CEO walked it back and started hiring humans again.
There's a second piece most teams miss: brittleness. Your quality depends on models you don't own and prompts that can silently regress when an upstream provider changes something. Quality can drop without anyone on your team touching the code. The defense is to run the same evals across every model you use and watch the gaps. Your model is now a vendor you have to actively manage, not a dependency you set and forget.
What to measure: thumbs-up rate and regenerate rate as your core signals. Eval-harness scores tracked over time, the way you'd track retention, across every model you touch. And quality distribution by cohort, because new users experience a worse product than power users by default, and most teams aren't measuring that gap.
This wasn't in the Lean Analytics book, because it didn't need to be. With traditional SaaS, if a user could click buttons and read labels, they could use the product. With AI, comfort with the technology itself is a variable, and it shapes every downstream metric you care about.
Gallup's February 2026 study of 23,717 US employees found that what separates AI adopters from holdouts isn't access to the tools. It's whether they see AI as useful, ethical, and a fit for their workflow. Stanford's 2026 AI Index puts global employee adoption at 58%, with the US trailing at 28%.
In B2B, you'll see meaningfully different activation, stickiness, and task-diversity curves between AI-native users and AI-hesitant ones. Same product, same plan, same role, different behavior. Measure them as one cohort and your averages hide the real story.
Trust isn't one thing, either. It's at least four, and they move independently:
What to measure: adoption curves segmented by AI-comfort cohort, accept rate sliced the same way, and override rate (how often the user rewrites the AI's output). A falling override rate is a rising trust signal.
Trust and comfort aren't soft, unmeasurable things. They leave real signals in your data. In an AI era, they effectively are retention.
So here's our question for the corporate innovators, product managers and founders reading this: are you still optimizing for walls, or are you measuring throughput?
