Emerging Product Demand Detector

Discover products that people are starting to want — before everyone else does.

Give it a product, niche, or keyword (e.g. portable ice maker, home gym, robot vacuum) and it discovers the products people are actually researching and buying, then ranks them by a transparent Opportunity Score built from real public demand signals.

How it works / responsible use: This Actor uses only public, no-login demand signals — YouTube search (content volume + view counts), and Google, YouTube & Amazon autocomplete (what people are actively searching). It does not log in, scrape private accounts, bypass CAPTCHAs/paywalls, or access anything gated. Every score is explainable.

---

Who it's for

Amazon affiliates, e-commerce sellers, product researchers, brands, agencies, and content creators who want an early-warning system for product demand.

The scores (all 0–100, transparent)

Score	What it measures
Demand	Total YouTube view volume + mentions + Amazon-search corroboration
Momentum	Peak recent view velocity + share of fresh content (acceleration)
Purchase Intent	Presence in Amazon search suggestions (buyers) + intent language ("best", "vs", "review")
Competition	Content saturation for the product + how crowded the niche is
Sentiment	Positive vs negative language in titles
Opportunity	Weighted blend favouring momentum, intent, demand, low competition, positive sentiment, multi-source

Opportunity weights (exposed in scoreComponents.weights): momentum 0.25 · purchase-intent 0.20 · demand 0.18 · low-competition 0.15 · sentiment 0.12 · source-diversity 0.10.

Every product ships a scoreComponents object showing the raw numbers behind each score — no black-box ranking.

Emerging detection

Each product is classified rising / stable / peaking / declining from within-run momentum, and — when you pass a previous run's dataset — from real growth across runs.

Input

Field	Type	Default	Description
keywords	array	–	Products/niches/keywords to research. One or many.
expansionDepth	integer	1	Autocomplete expansion: 0 = seed only, 1 = seed + a–z (recommended), 2 = deeper.
maxProducts	integer	40	Max ranked products per keyword.
affiliateMode	boolean	true	Flag affiliate-suitable products + generate content angles.
minOpportunityScore	integer	0	Drop products below this Opportunity Score.
previousDatasetId	string	–	Prior run's dataset ID → mention growth + acceleration across runs.
proxyConfiguration	object	Apify Proxy	Datacenter proxy is fine; no residential needed.

Example inputs

{ "keywords": ["portable ice maker"] }

{ "keywords": ["robot vacuum", "standing desk"], "maxProducts": 25, "affiliateMode": true }

{ "keywords": ["home gym"], "minOpportunityScore": 60, "previousDatasetId": "<PREVIOUS_RUN_DATASET_ID>" }

Output

1) Emerging products (default dataset, ranked by Opportunity Score)

{
  "productName": "Roborock",
  "brand": "Roborock",
  "category": "robot vacuum",
  "asin": null,
  "productUrl": "https://www.amazon.com/s?k=Roborock",
  "sourceUrls": ["https://www.youtube.com/watch?v=..."],
  "demandScore": 74, "momentumScore": 91, "purchaseIntentScore": 100,
  "competitionScore": 62, "sentimentScore": 100, "opportunityScore": 81,
  "trendDirection": "rising", "mentionGrowth": null, "mentionGrowthPercent": null,
  "sourceCount": 3, "platformsDetected": ["youtube","amazon","google"],
  "keywords": ["roborock qrevo", "roborock s8"],
  "commonUseCases": ["pet hair", "hardwood floors"],
  "pros": [], "cons": ["robot vacuum getting stuck"], "complaints": ["..."],
  "competingProducts": ["Ecovacs","Eufy","Deebot"],
  "scoreComponents": { "demand": {...}, "momentum": {...}, "weights": {...} },
  "affiliate": {
    "affiliateSuitable": true,
    "contentAngles": ["best Roborock","Roborock review","Roborock vs Ecovacs","best Roborock under $100","Roborock alternatives"],
    "amazonSearchUrl": "https://www.amazon.com/s?k=Roborock"
  },
  "observedAt": "2026-08-30T18:00:00.000Z"
}

2) Source evidence (named dataset source-evidence)

One row per supporting signal — which platform it came from and the URL/what it was:

{ "seed": "robot vacuum", "productName": "Roborock", "source": "youtube", "signalType": "video-mention", "url": "https://www.youtube.com/watch?v=...", "observedAt": "..." }

A run summary is written to the key-value store record OUTPUT.

Scheduling & historical acceleration

Schedule the Actor and pass the previous run's dataset ID as previousDatasetId. It then computes mentionGrowth / mentionGrowthPercent and upgrades trend detection to real cross-run acceleration — so you catch products that are suddenly taking off.

Scoring methodology (summary)

Signals are normalized within each keyword's candidate set (relative min-max for demand/momentum/competition; absolute lexicon scores for intent/sentiment; source count for diversity), then combined with the fixed Opportunity weights above. Full per-product math is in scoreComponents.

Limitations (honest)

• Product extraction is heuristic. Brand/product names are mined from YouTube titles and cross-checked against Amazon autocomplete. It catches the real brands in a niche well, but an occasional common word can slip through, and pure model-number products may be missed. It is not a trained NER model.
• ASIN is not resolved (asin: null) — Amazon product pages aren't scraped. productUrl is an Amazon search link, which is affiliate-ready.
• Sentiment is title-level (public metadata), so it is a light signal, not deep review analysis.
• Scores are relative demand intelligence, not official platform analytics.

Architecture (modular)

sources/ (suggest + youtube) · core/ (text extraction, sentiment, scoring, emerging, affiliate, state) · main.js (orchestration/output). New marketplaces or social platforms (Reddit, TikTok, etc.) drop in as additional sources/* modules and feed the same scoring engine — they need their own official API keys or residential proxies, which is why this v1 ships the reliably-public sources.