ecommerce-product-research git:20260921.31c406eB

Immutable. This exact content is served forever at /api/v1/blob/76017b6c9fadde9a.

---
name: ecommerce-product-research
description_zh: "为商家做跨平台选品与类目机会分析，比较需求、竞争、单件贡献利润、供应商和验证成本；用于选品、竞品比较、供应链评估，不负责采购下单或经营复盘。"
description_en: "Select ecommerce products and assess category opportunities across channels using demand, competition, contribution profit, sourcing and validation costs. Use for product selection, competitor comparison and supplier evaluation, not purchasing or operating-performance reporting."
category: ecommerce
---

# Ecommerce Product Research

Help a merchant decide what to test or source next. Start from their market, customer, budget, capabilities and decision horizon, rather than a platform's bestseller list.

1. Establish a bounded candidate set and the evidence available. Supplied exports and supplier quotations are usable without any external service. Read [evidence and channel context](references/data-provenance-and-confidence.md) and build its compact claim ledger from sources actually observed in this task. For open-ended early discovery without category-level evidence, prefer explicit hypotheses over broad market statistics. Research only a decision-changing external claim, and include its direct source in the answer; otherwise omit the claim, report name, link, ranking and current statistic.
2. Apply [selection and unit economics](references/product-selection-method.md) to the decision. Compare demand, competition, differentiation, contribution profit, sourcing, seasonality and return exposure. Treat an unrestricted dimension as permission to propose a bounded test scope: choose and label that scope as the researcher's proposal instead of returning the choice as a prerequisite. Choose one current validation stage—category comparison, supplier/sample qualification, inventory demand trial or replenishment—before specifying its plan. Keep the executable actions within that stage; describe a later inventory stage only by its missing inputs, advancement gate and calculation method until it is unlocked. Expand only a decision-changing evidence gap; freeze candidates before the final comparison.
3. For Amazon-specific signals read [Amazon research](references/amazon-market-research.md). For Chinese retail or sourcing evidence read [China research](references/china-platform-research.md). Other channels use the same method with their actual fields; neither reference grants data access.
4. Before delivery, reconcile external/current factual claims against the observed claim ledger. Build allocations and total-spend recommendations only from a supplied commitment ceiling. Without one, make the current stage executable through scope, sequence, observations and decision rules; a useful monetary scenario must be explicitly conditional on the merchant choosing that ceiling. Propose inventory quantity only when both a commitment ceiling and a compatible landed-cost basis are supplied. Remove or demote an unsupported fact to a hypothesis; do not treat a remembered report title as observed evidence. For a fixed total budget, prefer one executable base allocation with its explicit arithmetic total. Add ranges only when they help the decision, and then show both range totals within the ceiling. Then use [the decision output](references/product-research-output.md) at the requested depth. Return candidates and rejected alternatives, the conditions under which each works, and a reversible validation with bounded commitment that could change the decision. Do not call it cheapest or lowest-cost unless compared costs support that ranking.

Do not turn rankings, social interest, a supplier badge or a proprietary opportunity score into observed retail demand or guaranteed profit. Unknown costs remain unknown; conditional scenarios can still be useful. Detailed VOC, final copy and actual purchases belong to their corresponding capabilities.

No external dependency is required. Third-party seller tools are optional evidence sources; configured credentials or open-source client code do not imply free data or permission to incur charges. This Skill ships no marketplace scraper or live store connector.