For my last product I sent about 2,000 cold emails. Almost nobody replied. What bothered me more was what I found afterwards: the people who needed it had been asking for it in public the whole time. In forums, in review sites, in "what tool do you use for X" threads.
So I started building Crossroad (https://crossroadhq.com/?utm_source=devto). The idea is simple: find people and companies who are publicly asking for what you sell, show you exactly what they said, and help you write a reply you send yourself.
The idea is simple. Doing it honestly is not. Here is what I learned building the first version.
1. Searching is easy. Deciding is the product.
The pipeline has three steps:
- Read the website. What does it sell, to whom, and where are those buyers? The founder can correct anything.
- Search the last week. Communities, review sites, business news and AI answers, using questions a buyer would actually ask.
- Score every result, with a written reason.
Step 2 returns a lot. Step 3 is where almost everything gets thrown away.
In a real test week for a small CRM for independent insurance agencies, the tool read 48 results and kept 2. Both were business news: an agency opening a new office, and an agency acquiring another one. Those are moments when an agency suddenly needs better tooling. The other 46 were set aside, each with a reason.
2. Most of the open web is vendors talking
For broad products (a helpdesk, a developer tool, a generic AI assistant) the open web is mostly vendor content: comparison pages, "top 10 tools" lists, launch posts. They contain all the right keywords and zero buyers.
Keyword matching can't tell the difference. "Looking for a CRM for my agency" and "We built the best CRM for agencies" share almost every word. So every result gets a question: is this a person or company with the need, or someone selling to them? Vendor content gets dropped, however well it matches.
The honest consequence: the tool works best for products with a clear type of buyer. For very broad products I'd rather show nothing than show noise.
3. No source, no lead
Every match has to carry:
- the link to the original post or article
- the date it was captured
- the person's own words (a short quote)
- the reason it fits, and what we don't know yet
If any of that is missing, it isn't shown. A summary of "why they fit" without the original text is just a guess you can't check.
4. Finding isn't the bottleneck. Replying is.
This week another founder building something similar shared his numbers on Reddit: his users contacted about 2% of the people the tool found. Finding people wasn't the problem. "I don't know what to say to a stranger, so I'll do it tomorrow" was.
So the reply matters more than the list. A good reply helps with the actual question first, and only mentions the product if it's genuinely the answer. And the tool never posts for you. Communities are right to ban bots, and an automated reply is spam no matter how well it's targeted.
Where it is now
The site went live two days ago and early access is open. The first three matches for each product are free, checked by hand while I learn what makes people actually reply.
If you've ever found customers by answering questions in public, I'd love to hear what made you hit send.

