In most businesses client acquisition works like this: nobody looks for new customers while the pipeline is full, and when it empties everyone within reach gets messaged in a panic. The result is the same in both cases — an irregular flow and revenue you can't forecast. Yet done properly, finding clients becomes a system that runs quietly in the background. Here's the logic behind the one we built for our own sales operation.
The First Question Is: Who Is Not a Client?
The first step in client acquisition isn't finding — it's defining. A business whose target audience is "everyone" reaches no one, because the message ends up written for everyone and lands on no one.
Once the definition is sharp, the system shapes itself. If you offer a 24/7 voice AI assistant aimed at clinics, your target is a particular kind of clinic — not a hospital or a public institution. The same holds for real estate, education, manufacturing or local services: the sector changes, the logic doesn't. Writing down who is not a fit clears half your list before you start.
Which Channels Do We Use?
A system resting on a single channel is fragile; when that channel closes, the flow stops. In practice we run four streams together:
- Map and business listings. Almost every local service business appears here publicly: what they do, where they are, how established they look. For anything requiring geographic targeting, this is the most productive stream.
- Professional networks. The most direct route to a decision maker by job title. The work here is targeted, manual research rather than automated collection — both platform rules and hit rate demand it.
- Your own content. The most valuable stream is the one that comes to you. Someone arriving from a search result or an article is far warmer than someone you went looking for. This article is part of that stream.
- Referrals and partnerships. Recommendations from existing clients and from firms serving the same audience differently. Low in volume, highest in conversion.
They won't all run at the same speed, and they don't need to. What matters is that none of them is a matter of life and death on its own.
The Real Work: Filtering the List
A raw prospect list is close to useless. It contains records that don't fit your target at all, businesses that have closed, the same company written three different ways, and missing contact details. This is where the system earns its keep:
Candidate companies
↓
Deduplication → a company never enters the list twice
↓
Target filtering → records outside your audience never enter
↓
Detail completion → missing contact information is filled in
↓
Prioritisation → who to write to first is decidedThe most underrated step here is deduplication. Messaging the same company twice is worse than never messaging them: they may have missed the first one, but by the second they know who you are — and in the wrong way. So in this system a company enters the list only the first time it is seen.
Why Only a Few Per Day?
Our system runs once a day and deliberately adds a limited number of new companies. Speeding it up is technically easy and practically pointless.
Because the bottleneck isn't the number of prospects. The bottleneck is how many companies you can write to genuinely personally in a day. Having five thousand companies in a spreadsheet changes nothing; if you can write properly to the twenty at the top of the list, the system is working. That's exactly what prioritisation is for — deciding where your limited attention goes.
First Contact: Why Most Messages Get Ignored
Messages that never get answered share one trait: they could have been sent to anyone. "Hello, we offer digital transformation solutions" doesn't touch any business's actual agenda.
A message that gets answered carries a concrete detail about the recipient's business and offers one clear benefit. That's why the system prepares each message separately: the company name, its location and a few specifics about it are woven into the text. We go down to the small stuff — even the grammatical inflection of a place name is generated correctly for the language. It sounds trivial, but it isn't for the person reading: one wrong inflection announces in the first line that the message was mass-sent.
The second rule: lead with value instead of sales pressure. The job of a first message isn't to close a deal; it's to make the reader think "this person actually looked at my business".
The Brain of the System: AI Running the Flow
In most setups with "automation" in the name there is no artificial intelligence at all — only fixed rules and ready-made templates. This one is different: AI isn't a bolt-on at the edge, it sits at the centre and runs the process. Think of it as an agent — the data comes to it, it decides what to do, and it produces the result.
It evaluates the incoming data. Records arriving from the channels are raw and messy; the same company may appear written three different ways, some fields are empty, others are in the wrong place. It's the AI that converts them into one shape, merges the variants of the same company, and decides whether each record is usable at all.
It separates right from wrong. This is the real work. An institution outside the target audience, a business that has closed, a fake contact detail leaking from a website template — all of it is filtered out before it ever enters the list. That judgement can't be made with a fixed list of rules; it requires reading the record and assessing it.
It researches what's missing. When a record is missing contact details, it looks at the company's own public sources and fills the gap. When it can't find it, it doesn't invent one — it marks the field as missing. A fabricated contact detail costs far more than an empty field.
It sets the order and writes the message. It decides which company to write to first, then produces a message sequence specific to that company: an opening, a gentle follow-up and a close. The text is grounded in that company's own details, several subject-line alternatives are generated, the tone leads with value instead of sales pressure, and an opt-out line goes into every message. Producing generic templates is deliberately ruled out.
Two things are left to a human: the final read of the message and the decision to send. Everything else along the chain is run by the AI.
What Is and Isn't Automated
We keep this distinction sharp, because "we automated everything" is usually untrue:
- Automated: compiling candidate companies, removing duplicates, filtering out records outside the target, completing missing details, sorting the list and preparing message drafts.
- Left to a human: the final read of the message, the decision to send, the reply to an answer, and the decision to move a conversation forward. Those aren't automatable — and when they are automated, people notice.
Stating the Limits Up Front
Not saying what a system doesn't do makes it look stronger than it is. This pipeline doesn't know when a company was really founded or how many people it employs; it infers from publicly available information. Those inferences are good enough to sort a list, but they shouldn't be presented as hard data.
The Rules: Data Protection and Basic Manners
Cold outreach earned its bad name not from the channel but from its unregulated use. The boundaries we set:
- Only publicly available corporate contact details of businesses are used; personal addresses are not targeted.
- No automated collection that strains platform rules.
- Every message carries the sender's real identity and a clear way to opt out.
- Anyone who opts out is removed and never written to again. No exceptions.
Adapting It to Your Own Business
The skeleton here isn't sector-specific: define the target, search across several channels, remove duplicates and off-target records, complete the contact details, prioritise, and personalise the first contact. What changes is the weighting of channels and the filtering rules — the logic stays the same.
If finding new clients in your business still depends on someone having a free afternoon, we can work out together how to turn it into a system that runs steadily — see our AI sales automation page. For an example of a sector-specific build, take a look at AI automation for clinics.