Are data effects a moat against AI?
Yes: when usage produces data that makes the product better, and a better product drives more usage, the loop compounds in a way a copier can't buy.
Holds against AI
Updated November 19, 2026.
What is it?
A self-reinforcing loop. Every use of the product generates data, that data improves the product, and the better product attracts more use. Cursor's accept-and-reject signal on code suggestions is a current example.
Does it hold against AI?
It holds, and AI makes it more achievable. Hockenmaier names data effects as one of four compounding advantages that count as real moats. Modern models can learn from feedback signals that used to go unused, so more products can build this loop than before.
It fails when the signal isn't unique (anyone with similar users collects the same data) or when the improvement is small enough that a better base model erases it.
How do you know if you have it?
Can you point to a metric that improves because of usage, not because of engineering work? Would a competitor with the same model but none of your usage data be meaningfully worse? If not, you have data, not a data effect.
How do you deepen it?
Instrument the moments where users accept, reject or correct output; those are the most valuable signals. Close the loop quickly, so improvements show up while users still notice. Measure it, so you know the loop is real.
Questions people ask
- What's the difference between proprietary data and data effects?
- Proprietary data is a stock you hold. A data effect is a flow: usage that keeps improving the product.
Sources
Does your business have a moat?
Five AI attackers, live web research and a published rubric. Free, about five minutes, no account. Or talk it through with Matt on a strategy call.