Most "OSINT for business" content targets security teams. But the cheapest useful intelligence product I built ran on a different buyer: a reseller. Someone who needs to know, before their competitors do, that a product just became scarce in a region - because a supply channel started posting "back in stock" at 2 AM, or a price war just opened in a Telegram marketplace group.
The method transfers directly from analyst work, with three changes:
1. Track entities, not events. Analyst feeds alert on discrete events (leak, incident, announcement). Market-intelligence feeds track state: SKU present/absent, price point, currency, delivery zone. Every post is reduced to a tuple (entity, attribute, value, timestamp) and the chart is the product, not the post. A reseller pays for the line that moved, not for 4,000 messages.
2. Normalization is the whole game. "3500 грн", "3,500 UAH", "3.5k", "стоимость 3500" are one data point. The regex dictionary that collapses them is 80% of the build. It's also the moat: a buyer who tries to replicate the feed discovers they can parse posts, not markets.
3. Cadence follows the market, not the clock. Markets move in bursts after restocks, policy changes, and regional shocks. Poll slow (hourly) baseline; auto-accelerate (10-min) when a channel's post-rate triples. Volume spikes are the market's own signal that something is happening.
What a $5 product in this space can be: not a dashboard subscription (too heavy to run, too heavy to sell solo), but a weekly digest of state changes - the five SKUs whose line moved most, the two new sellers entering, one price-war forming. Twelve bullet points a week. The buyer's decision gets faster; your cost is zero after the crawler exists.
I built the collection layer (public t.me/s/ preview polling, no login, free GitHub Actions cron) as a reusable recipe: the cadence rules, entity-normalization dictionary, and digest template are the core of the Telegram & Web OSINT Bundle ($5). A free sample brief shows what one week of output looks like.
The lesson for engineers with data skills: your cheapest sellable product is usually one layer above the data - not the dataset, not the dashboard, the decision summary. Datasets need support. Dashboards need servers. A weekly PDF of twelve bullets needs none.
Runs free on GitHub Actions - no server, no paid APIs.
Top comments (0)