I built two n8n pipelines that validate niche ideas against real data instead of vibes

The problem with "40 business ideas" listicles Every "micro-SaaS ideas" article is someone's shower thoughts. I wanted to know what the data actually says, so I built two automated research pipelines in n8n — one that checks demand (is anyone searching for this?) and one that checks traction (is anyone already building and paying for this?). Here's how they work and what they found. Pipeline 1: The Utility Opportunity Scanner (demand side) Most keyword research starts with a keyword. This pipeline starts with a shape: single-purpose web utilities — calculators, converters, generators, checklists, document tools. The boring tools everyone uses and nobody talks about. How it works: A seed grid of tool-shaped searches (age calculator, heic to jpg, packing list, qr code generator…) gets expanded via DataForSEO's keyword suggestions API — one task per seed — returning real phrases with monthly volume, 12-month trend, CPC, and keyword difficulty. A code node filters to tool-shaped queries (must match calculator|generator|converter|template|checklist|...) and excludes app-shaped terms (tracker, journal, reminder). It canonicalizes keywords so bmi calculator and bmi calc don't double-count. Scoring: 30% demand (log volume) · 25% rankability (inverse KD) · 20% build simplicity · 15% monetisation (CPC) · 10% cluster strength. The top 40 survivors get checked against live SERP top-10 results — who's actually ranking, and whether they're weak incumbents or untouchable ones — plus a Chrome Web Store check for extension-shaped competition. It pulls real Flippa comps filtered to ad-monetised content/tool sites — actual businesses that sold. Everything clusters by domain into "umbrella site" concepts, and OpenAI (strict JSON schema) names each site, picks the strongest tools, and writes per-tool briefs. Results land in Notion. Pipeline 2: The Proven Product Scanner (traction side) The demand side tells you if people are searching. This one tells you if people are building and buying. It scans: Product Hunt launches (official GraphQL API) Show HN posts (HN Algolia API) Chrome Web Store extensions iOS apps (iTunes Search API) AppSumo deals Reddit posts across consumer-problem subreddits Candidates get normalized, deduplicated by URL/title, matched against ~35 configured themes, and scored on popularity, category maturity, simplicity, price fit, and recency. Shortlisted app-like candidates get enriched with Play Store metadata, themes get Google Trends direction, and up to 8 real Flippa MRR/asking-price comps get injected into the OpenAI prompt so the model reasons against real acquisition prices, not hypotheticals. Accepted briefs (confidence ≥ 6) write to a second Notion database. A companion workflow polls that database every 5 minutes and generates structured PRDs on approval — so a validated idea can go from signal to spec without me touching it. What the data actually found Running the utility scanner against the "everyday tools" niche produced one umbrella-site concept with ~14M searches/month of combined demand across 40+ scored tools: Mortgage Calculator: 2.7M/mo BMI Calculator: 2.2M/mo Word Counter & Background Remover: 1.2M/mo each QR Code, GPA, Compound Interest: 823K/mo each The money signal is in the CPCs, not the volumes: JSON Formatter keywords run $15.05 CPC, lawn-care tools $15.37, Sales Tax $5.49. Those clusters are where display-ad revenue actually lives. And the build reality: most of these tools are trivial client-side builds — no backend, no API costs, deployable in days. The moat isn't any single tool (single tools die alone), it's the cluster — 40 cross-linked utilities under one domain feed each other in Google and eventually become a sellable asset. The Flippa comps prove there's a real exit market. The honest bit This is evidence, not a guarantee. Volumes are third-party estimates — sanity-check the big ones in Keyword Planner before building anything. Rankings and revenue depend on execution. The scanners tell you where to dig; they don't dig for you. What surprised me most: how many obviously good ideas failed the data (high volume, brutal SERP incumbents) and how many boring ones sailed through. The pipeline's most valuable output is arguably the watchlist of tools that failed — it kills months of wasted building. Stack n8n (self-hosted) — orchestration, all logic is Code nodes + HTTP requests Keyword everywhere API — keyword suggestions, volume, CPC, KD Serper — live SERP top-10 checks Apify — Chrome Web Store, Play Store, AppSumo, Reddit, Google Trends scrapers OpenAI — schema-validated briefs and site naming (gpt-4o-mini is plenty) Notion — Candidates, Briefs, Umbrella Sites, and PRD databases I turned the utility scanner's full output into a research report (the blueprint — ranked launch order, raw CSVs, evidence strings behind every score) and I now run the same pipeline on other people's niches. Links in the comments if you want to see the output format — but honestly the interesting part is the method, and everything above is enough to build your own version. Happy to answer questions about the pipeline design.