The Opportunity
Our client is an early-stage B2B software company selling an operations layer to businesses in a single, fragmented vertical. Its buyers number in the thousands and each one is small, so the acquisition economics only work if the motion runs close to unattended. There was no sales team to hire against and no list to buy.
The addressable accounts were the harder half of that problem. They all ran on one industry booking platform, which meant they shared a signature, but no commercial database segments on it. A generic firmographic export would have returned the same companies every competitor was already writing to, most of which did not run the platform at all.
The second constraint was legal. Several of the company's European markets prohibit unsolicited commercial email outright, so a single-channel email motion was closed off before it began. Whatever we built had to route those markets elsewhere, and had to do it without depending on anyone to remember the rule.
The Solution
Assembling a list that did not exist
The booking platform publishes a directory of the businesses running on it. We captured it, stripped out the test, demo and template records that make up a surprising share of any such directory, and inferred country and business category from what remained. That left a little over a thousand live brands, each carrying the platform's own identifier, which became the deduplication key for everything downstream.
The directory holds no websites, so the domains were researched in parallel and written back against each company. Research of that kind returns a confidence rating, and we treated that rating as worthless on its own. It is a guess's opinion of itself. The check that counted came next.
Verifying, and failing closed
Each company's site was probed across its homepage and its booking paths for a genuine integration signal, read only from the markup that does the work and never from prose or comments that merely mention it. That alone was not enough. Because the domain had been researched from a brand name, a wrong site that happened to run the same platform would look identical to a right one. So a confirmation also required the site to name the company it was supposed to belong to.
Anything showing the signal without confirming identity was recorded as unconfirmed rather than waved through, and the exact page a confirmation was read from was stored alongside the verdict, then cleared if that verdict ever changed. Evidence never outlives the claim it supports. Only confirmed accounts reach outreach.
Two channels on one system of record
A CRM holds every company and person, and the two outreach channels run against it as separate funnels with separate stage columns. They are never merged. Both channels carry a stage called replied, for different populations, and one prospect can sit at a different point on each. A scheduled process claims a small batch of uncontacted people several times a morning, marks each as claimed before it acts, and routes by the contact details that exist: a professional network profile goes to the network channel, an address goes to email, and someone with both gets both.
Compliance as a gate, ahead of the send
Before any of that runs, a gate stamps each person as emailable or not. The markets where cold email is unlawful are held to the network channel, and the gate refuses to pass a record whose country it cannot establish, so an unknown fails closed. A company-level flag suppresses every person at that company across every channel, including people imported months later, which is how existing clients and partners stay off the list. Suppression is ordered by strength rather than by recency: an unsubscribe outranks everything and cannot be erased by a reply that arrives after it.
Drafting at volume, sending by hand
Outreach copy is drafted by model, one message per prospect, alongside a one-page summary built for that specific account and linked from the message. Neither goes out on its own. Drafts wait for a person to approve them, and a read-only board shows what is ready, what is blocked, and the delivery signals that would call for stopping. The scale comes from the research and the drafting. The decision to send stays with a human.
The Impact
A list worth writing to. More than a thousand accounts have been targeted, each one verified as running the platform the product attaches to rather than inferred from a firmographic filter. The qualification happened before the first message, not in the replies.
Reply rates between 12% and 15.5%. That is the range the outreach has held. Each message is drafted against what the verification work already established about that specific account, which is what the list was assembled to make possible.
A motion that runs without a sales team. Sourcing, verification, enrichment, routing and drafting all run on a schedule. A person approves the messages and takes the calls, and the same pipeline now carries a second programme on the same infrastructure.
The Engagement
We built the pipeline end to end: the directory capture and cleaning, the domain research, the verification layer and its evidence trail, the CRM schema, the two-channel routing, the compliance gate, the drafting system, and the read-only board the whole thing reports into.
Albert Abello and Joost Brok lead the engagement, covering the go-to-market design, the data and systems work, and the outreach programme itself.