We tracked 1918 leads from Telegram chats across nine niches. Here is where they actually get worked

We run a service that reads open Telegram chats and pulls out messages that look like someone asking for a service. Over two months, from 8 September to 4 November 2026, nine paying accounts collected 1918 leads from 1360 chats. The niches range from equipment certification to private therapists.

We did not go into the database looking for a nice number. We wanted one thing: the gap between a lead the robot finds and a lead a human actually acts on. That gap turned out to be wider than we expected.

The numbers

Each account ran for a different stretch, so we also show leads per day. The last column is the share of leads the account owner marked as "taken into work", meaning they went and messaged the person.

What this table is really saying

Compare the two extremes. Certification produces 14.6 leads per day, five times more than cleaning. Yet 2% get worked versus 69%. In absolute terms that is 12 leads against 68. The niche that looks five times richer in a dashboard gave the human five times less actual work.

Fulfilment scored an honest zero. 123 leads in 19 days, not a single touch.

The conclusion we drew about our own product: lead volume says almost nothing about whether a channel is useful. It says how widely your topic gets discussed in chats.

Why certification sits at 2%

We read a sample by hand. In seller and manufacturer chats the words "certificate" and "declaration" come up constantly, but nine times out of ten it is not a request for a service. It is a conversation: which document does the marketplace require, who has dealt with this, how much did it cost. The robot sees a keyword plus a question mark and files a lead. The human opens it and finds a discussion that nobody wants a vendor to barge into.

Bot and API development behaves the same way. 170 leads, 2 touches. Developers are discussed in tech chats around the clock, and hired by referral in private messages.

Why cleaning sits at 69%

The mechanics are reversed. "Looking for a cleaner after renovation" or "can anyone recommend someone to wash windows" means exactly what it says. A person with money and a task this week. There are few such messages, 2.9 a day, but roughly every second or third deserves a reply.

Real estate at 36% works the same way. "Renting a one-bedroom, up to 40k" is a complete request with nothing to interpret.

What cleaning, real estate and therapy share: the buyer is a private person, the request is short, the budget is modest, and the decision happens within a day. What certification, fulfilment and development share: the buyer is a company, the request is long, the sales cycle runs for weeks, and the chats are full of professional chatter in which a real request is indistinguishable from a discussion.

What to do with this if you are picking a channel

Three things we would tell ourselves two months ago.

First. If your customer is a private person with a task this week, chats work for you. Cleaning, renovation, home services, rentals, independent professionals. You will not get many leads, but they will be real.

Second. If you sell to companies on a long cycle, chats will hand you a stream you will not act on. Not because the tool is bad, but because your buyer does not post "we need certification, who can do it" in an open chat. They ask someone they already know.

Third, and this one is about us. For two months we were proud of the account with 539 leads and treated it as our best. In revenue it produced one payment. The account with 99 leads produced more and the customer stayed. "Leads found" is a lying metric. The honest one is leads a human chose to work.

How we counted

The numbers come from the production database, not from a survey. We took accounts with at least one successful payment above 1000 roubles, meaning real subscriptions rather than 190-rouble trial runs. Leads come from the leads table, touches from the contacted status, which the account owner sets by hand. Each account has its own period from first to last lead, hence the per-day normalisation.

What we do not know and do not claim: how many of the worked leads turned into money. That happens in the customer's private messages and we do not look there. So the "taken into work" column is the most honest thing we can show.

One more caveat: the table shows the already filtered stream. Before filtering, 542,795 keyword matches produced 14,774 genuine requests, which is 97.3% noise. Our language-model filter cuts 95% of the junk and loses roughly 5% of the good leads.

The service all of this was measured on is called Leadolovets. Five-day trial, no card required.