A B2B business where one salesperson was the ceiling on growth. We replaced manual prospecting with an engine that turns a city name into a worked pipeline — and knows which ground it has already covered.
Prospecting was done by hand. Someone opened a map, searched a category in whichever city came to mind, copied names and numbers into a spreadsheet, wrote an email, and — often — forgot the follow-up. The work was real but the coverage was accidental: the same dense districts got worked repeatedly while entire regions were never touched at all.
The deeper problem was arithmetic. Deals here are large enough to deserve a real conversation and thin enough on margin that volume matters, and every close sits on top of a great many qualified conversations before it. One person cannot hold that many in a month, so the funnel was never the bottleneck — attention was.
Every attempt to fix it with volume made it worse. Blasting a bought list burns the domain, burns the brand, and in some jurisdictions burns the account outright.
The goal was never "send more email". It was to make coverage systematic and contact deliberate, so the human is spent on the twenty conversations that are worth having.
We built an engine that treats a market as territory to be worked rather than a list to be blasted. You give it a city. It divides the surrounding geography into tiles, works each tile until that tile is exhausted, and records what it found so it never pays to search the same ground twice.
For each business it finds, it writes a message — not a mail-merge with a name slotted in, but a short note composed against that specific company, in the language of the country it operates in. It sends, watches for a reply, matches the reply back to its own thread, and then does the one thing that matters most: it stops and hands the conversation to a person.
Geography is stored as tiles, each with a state — untouched, in progress, exhausted. A tile that returns nothing three times decays out of rotation. The system knows the difference between "we have not looked there" and "we looked there and there is nothing", which is the distinction every manual process loses within a week.
Dense districts get subdivided further; empty ones do not. Search density follows the actual density of businesses rather than a fixed grid, so budget goes where the prospects are. The floor is set fine enough to catch a business park and coarse enough not to waste calls.
Every search costs money at the data source. Spend is tracked per run and per tile against a ceiling, with a hard cap on leads per batch. When the ceiling is hit the engine deactivates itself and reports — it does not keep going and drain the account. That single rule is the difference between an automation you can leave running and one you have to watch.
Country is detected from the lead’s own data, and the message is written in that language. Length is capped deliberately — a long cold email is a deleted cold email. Templates live outside the code and sync on a schedule, so the sales team can change the pitch without an engineer.
Replies are matched to their original thread by message ID and conversation ID, with an address fallback. A "Re:" sent three weeks later from a colleague’s address still lands against the right company instead of arriving as an orphan nobody recognises.
When a reply looks like interest, the engine stops and asks. An operator console — running in a chat app the team already has open — shows the thread and takes the decision. The system is autonomous right up to the point where being autonomous would be a liability.
These are properties of the running system rather than benchmark figures.
Coverage stopped being a function of who remembered which city. One command works a market, and the map of what has been covered is a table you can query rather than a feeling someone has.
The cost side became legible. When you know what a sourced lead costs and what a reply costs, you can argue about whether the channel is worth running — before that, everyone is guessing with confidence.
The engine was built as ten reusable modules rather than one monolith. That mattered more than expected: cloning the whole system for a business in a completely different industry, on a different mail provider, was a configuration exercise rather than a restart.
This shape fits any business where the buyers are findable but scattered, the deal is worth a real conversation, and the constraint is human attention rather than demand. Freight and logistics, industrial supply, trades and installers, wholesale distribution, B2B services with a defined service radius.
It fits badly where the total market is fifty companies and you already know all fifty. At that size, a system that finds prospects is solving a problem you do not have.
Tell us what's not working. The first conversation is free — whether it's a build, a long-term partnership, or just questions, you'll get an honest read and a clear next step.
Start the conversation →