The 2 a.m. Stripe ping: anatomy of a review-mining micro-SaaS

A back-end dev in Lisbon builds a tool over one weekend: it watches Capterra reviews of an accounting app and flags every feature users keep begging for. Product managers pay for the digest. The story goes viral — but the mechanics underneath are learnable. Here is the actual pipeline, and the actual math.

The pipeline, live drag to orbit · tune the sliders

Reviews scanned per day 120 % that are feature requests 18% Paying subscribers 14 Price per month $29

$406 MRR

≈ $4,872 / year, before fees

Your classifier surfaces ≈ 22 feature requests/day — that's the product: a filtered, ranked digest nobody has time to compile by hand.

Why this works: you're selling attention, not code

Public review sites (Capterra, G2, app stores) are full of sentences like “great app, but I wish it had recurring invoices.” Each one is a validated, written-down feature request. A product team at the vendor — or a competitor — would pay to see them ranked by frequency. The tool is small: a scraper, a classifier, a weekly email. The value is the compression: 840 raw reviews a week in, one ranked list out.

The build is a weekend. The moat is the niche: “feature-request radar for accounting software” beats “AI review analyzer for everyone.”

The five-step playbook

Pick one niche app category

One vertical (accounting, dental practice software, gym CRMs). Buyers must recognize their exact world in your landing page.

Scrape ethically

Poll public review pages daily; respect robots.txt and rate limits, or use official/partner APIs where offered. Store review ID, date, stars, text.

Classify & cluster

An LLM call or a fine-tuned classifier tags each review: feature request / bug / pricing complaint / praise. Cluster similar asks (“recurring invoices” x47).

Ship a weekly digest

Email with top-10 requested features, trend arrows, and verbatim quotes. Charge $19–$49/mo. Stripe + a cron job is the whole stack.

Sell to both sides

The vendor's PM team wants a roadmap signal; competitors want an attack list. Two customer types, same pipeline.

Reality check

The viral version skips the boring parts: scrapers break when sites change markup, classification needs spot-checking (aim for >90% precision on “feature request” before charging), and the first 10 customers come from cold DMs to PMs, not from luck. But the shape is real: tiny surface area, recurring value, and revenue math you can model honestly — as the sliders above show, 14 subscribers at $29 is ~$406 MRR, and 50 at $49 is ~$2,450.