Earlier this year we wrote about why we stopped using Clay and built our own AI personalisation engine. That system solved one problem: turning real research into personalisation lines that actually sound like a human did their homework. But it left another problem untouched. It personalised every prospect on the list, whether or not that prospect deserved to be on the list in the first place. The Signal Engine is what it grew into: a single system that qualifies and personalises in one pass, so that every prospect we contact has been researched, verified, and greenlit before a single email goes out.
The Problem With Personalising Everyone
A beautifully personalised email to the wrong prospect is still a wasted send. Worse, it's a wasted send that looks like effort. Most outbound teams treat qualification and personalisation as separate stages, often handled by separate tools, and the result is a leaky funnel: the list gets built against loose criteria, then the personalisation layer dutifully writes a clever line for every row, including the companies that were never going to buy. The research needed to personalise a prospect well is the same research that tells you whether they're worth contacting at all. It made no sense to keep those jobs apart.
How the Signal Engine Works
The workflow is deliberately simple from the outside. We feed the engine a prospect list and train it on the client's ICP: who they sell to, what problem they solve, what a genuinely good-fit account looks like, and just as importantly, what a bad one looks like. From there, the engine takes over. It researches each prospect across multiple data sources, verifies they actually fit the ICP, looks for active buying signals, and produces research-based personalised messaging for every prospect that makes the cut.
Step 1: Verify the ICP Fit
Every list has noise in it. Companies that matched a filter but not the intent behind it. Job titles that sound right but sit in the wrong part of the org. The Signal Engine's first job is to catch these. Because it's trained on the specific ICP rather than a set of firmographic filters, it can reason about fit the way a good SDR would: not 'does this company have 50 to 200 employees' but 'is this actually the kind of business, at the kind of stage, with the kind of problem our client solves?' That reasoning catches disqualifiers a filter never would: a company that turns out to be a subsidiary of a larger group, a firm that sits just above a headcount threshold according to recently published sources, or a business that is itself the type of company the client sells to, a peer rather than a buyer.
Crucially, nothing gets silently dropped. Every prospect receives an explicit verdict, qualified or disqualified, alongside a written reason for the decision, with sources cited where the engine relied on them. Disqualified prospects stay in the list with that reasoning attached, so every judgement the engine makes can be reviewed, challenged, and used to sharpen the ICP for the next run.
Step 2: Look for Buying Signals
Fitting the ICP means a prospect could buy. Buying signals suggest they might buy now. The engine looks for the signals that matter for the specific offer, including:
- Hiring activity that indicates a gap the offer fills, like a first SDR hire or a surge of sales roles
- Recent funding, acquisitions, or leadership changes that typically precede new investment in growth
- Tech stack changes and tool adoption relevant to the client's product
- Public content: posts, interviews, or announcements that reveal a live priority or pain point
- Company milestones such as new markets, new products, or rapid headcount growth
Only prospects showing genuine fit and credible signals get greenlit. This is the opposite of spray-and-pray: the engine's job is as much about deciding who not to contact as it is about preparing the outreach for those who make it through.
Step 3: Produce Research-Based Messaging
For every greenlit prospect, the engine turns its research into personalised messaging that's specific to that person, that company, and the client's offer. To be precise about what that means: the engine doesn't send emails, and it doesn't write them end to end. It produces the research-backed personalised messaging for each prospect and adds it to the prospect list as a dedicated column, ready to be used in campaigns. That distinction matters. The messaging is grounded in what the research actually found, and a human retains full control over the final email that lands in someone's inbox.
Each line also arrives with its working shown. The engine records the specific signals behind the messaging, whether that's a client case study, a job posting, a leadership appointment, or a founder's recent LinkedIn activity, and grades each personalisation by tier, so the strongest research-backed lines are distinguished from the merely solid ones. When we wrote about moving off Clay, the sharpest criticism was that you can't audit the reasoning behind a bad line because there isn't any. The Signal Engine is built as the opposite: every qualification verdict and every personalisation can be traced back to the evidence that produced it.
Why This Beats the Two-Tool Approach
When qualification and personalisation share the same research, three things happen. First, the messaging gets sharper, because it can reference the same signal that justified the outreach in the first place. Second, list quality compounds, because every campaign starts from a set of prospects that have each been individually verified rather than bulk-filtered. Third, reply rates reflect both improvements at once: the right people, contacted for a reason they can recognise, with a message that demonstrates someone actually looked.
The Signal Engine is the natural conclusion of the argument we made when we moved off Clay: enrichment gives you data, but outbound that converts requires judgement. Now that judgement is applied twice, once to decide whether a prospect should be contacted at all, and once to decide what would actually make them reply.