PilotOpen protocol for stores.Explore ACP
Back to Glossary

AI / ML

Compound Learning

The architecture by which AI models improve continuously as they learn from every marketing decision and its outcome. Named after compound interest because the improvement rate accelerates: each cycle of prediction, observation, and model update makes the next prediction more accurate. A compound learning system at month 6 reflects more learned context than at month 1 because it has observed many decision-outcome pairs specific to your brand. In the agent commerce era, compound learning also applies to agent visibility optimization — the system learns which product data structures, descriptions, and attributes correlate with higher agent mention rates, and continuously refines recommendations. The contrast is with a static model, whose accuracy is fixed at training time and drifts as conditions change.

Why it matters

It is the argument for tolerating an early model that is worse than a static one, on the expectation that it passes it. That expectation is the thing to verify, not assume.

In practice

Score accuracy over time rather than at a point. A system described as compounding that shows no accuracy trend after six months is not compounding.

Read the full guide

Related terms