Glossary

Cold-Start Problem

In one line: The cold-start problem is what happens when an AI system has no history to learn from — new products, new customers, new stores. Solvable, but worth understanding.

Where it shows up

You launch a new product — the recommendation engine doesn’t know who’ll like it. A new customer signs up — the personalization engine has nothing to personalize on. You open a new location — the forecasting model has no sales history. In all three cases, the AI is starting from zero data.

How good systems handle it

• Borrow from similar items. A new “navy medium” t-shirt? Predict based on existing similar SKUs.
• Use content features. For new products, the model uses category, brand, price, and description — not just purchase history.
• Bootstrap from a peer store. A new location forecasts based on the demographically closest existing store.
• Default to popular. Show new shoppers the bestsellers until you have enough behavior to personalize.

Why it matters in evaluations

When evaluating AI tools, ask explicitly: “how does this handle new products / new customers / new locations?” If the answer is “wait for data,” the tool will quietly fail you every time you launch something new. A serious system has a strategy.

Related terms

Recommendation Engine, Demand Forecasting, SKU-Level Forecasting